two-layer.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "weekly-tablet",
   "metadata": {},
   "source": [
    "# Two-Layer Networks on CIFAR10, Cyclical Learning Rate Schedule, and Bayesian Hyperparameter Search"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "antique-attention",
   "metadata": {},
   "source": [
    "## Contents\n",
    "\n",
    "* [Introduction](#introduction)\n",
    "* [Imports](#imports)\n",
    "* [Importing Dataset and Data Pre-Processing](#importing-dataset-and-data-pre-processing)\n",
    "* [Numerical Gradient Check](#numerical-gradient-check)\n",
    "* [Training and Results](#training-and-results)\n",
    "* [Hyperparameter Search](#hyperparameter-search)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "assisted-employee",
   "metadata": {},
   "source": [
    "## Introduction <a class=\"anchor\" id=\"introduction\"></a>\n",
    "\n",
    "In this notebook, a two-layer network with multiple outputs is trained and tested to classify images from the CIFAR-10 dataset. The network is trained using mini-batch gradient descent applied to a cost function that computes the cross-entropy loss of the classifier applied to the labelled training data and an L2 regularization term on the weight matrix. The learning rate uses a cyclical learning rate schedule. Furthemore, a Bayesian hyperparameter search is conducted with the use of external libraries.\n",
    "\n",
    "In my implementation, I built on top of the code presented in the ```one-layer.ipynb``` notebook. At the core of my implementation is the ```Dense``` layer (fully-connected layer that can have activations ```SoftmaxActivation```, ```LinearActivation```, and ```ReLUActivation```. The learnable (or trainable) parameters of the ```Dense``` layer can be initialized with the ```XavierInitializer```, and can be regularized with the ```L2Regularizer```. A ```Model``` comprises any number of layers, its loss function is ```CategoricalCrossEntropyLoss```. The trainable parameters are optimized with the mini-batch gradient descent algorithm via the ```SGDOptimizer``` and the learning rate schedule of the optimizer is [```LRCyclingSchedule```](https://arxiv.org/abs/1506.01186) . The ```Model``` is first compiled with the loss function, some metrics such as the ```AccuracyMetrics```, and the optimizer, and then it is fit to the data with the ```Model.fit``` method.\n",
    "\n",
    "In this notebook, I implemented a two-layer neural network for multi-class classification such that both of the ```Dense``` layers' trainable parameters are initialized with the ```XavierInitializer``` and regularized with the ```L2Regularizer```. The input dimension and the output dimension of the first ```Dense``` layer are 3072 (=32x32x3) and 50, respectively. The input and output dimensions of the second ```Dense``` layer are 50 and 10 (=the number of class labels), respectively. The first ```Dense``` layer has ```ReLUActivation```, and the second ```Dense``` layer has ```SoftmaxActivation```. The ```Model``` is then compiled with the ```CategoricalCrossEntropyLoss``` loss function, the ```AccuracyMetrics``` performance metric, and the ```SGDOptimizer``` optimizer that uses the ```LRCyclingSchedule``` learning rate schedule."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "flush-saskatchewan",
   "metadata": {},
   "source": [
    "## Imports <a class=\"anchor\" id=\"imports\"></a>\n",
    "\n",
    "Import the library parts and the required packages."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "compatible-retail",
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from tqdm.notebook import tqdm\n",
    "from sklearn.model_selection import train_test_split\n",
    "import pickle\n",
    "from copy import deepcopy\n",
    "from math import sqrt, ceil\n",
    "import datetime\n",
    "import time\n",
    "import sys\n",
    "from itertools import product\n",
    "import pandas as pd\n",
    "from hyperopt import hp, fmin, tpe, space_eval, STATUS_OK, Trials\n",
    "\n",
    "\n",
    "from data_utils import load_cfar10_batch, load_label_names\n",
    "from losses import CategoricalCrossEntropyLoss, LossSmootherConstant\n",
    "from activations import LinearActivation, ReLUActivation, SoftmaxActivation\n",
    "from initializers import NormalInitializer, XavierInitializer\n",
    "from layers import Dense\n",
    "from regularizers import L2Regularizer\n",
    "from models import Model\n",
    "from metrics import AccuracyMetrics\n",
    "from optimizers import SGDOptimizer\n",
    "from opt_utils import GradClipperByNothing\n",
    "from lr_schedules import LRConstantSchedule, LRExponentialDecaySchedule, LRCyclingSchedule\n",
    "from viz_utils import plot_losses, plot_costs, plot_accuracies, plot_lrs\n",
    "from grad_check import numerical_gradient_check_model"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ancient-photograph",
   "metadata": {},
   "source": [
    "## Importing Dataset and Data Pre-Processing <a class=\"anchor\" id=\"importing-dataset-and-data-pre-processing\"></a>\n",
    "\n",
    "In this notebook, for the sake of speed, the training set is ```data_batch_1```, the validation set is ```data_batch_2```, and the test set is ```test_batch```.\n",
    "\n",
    "Pre-process datasets by standardization of features (subtract mean of each feature and divide by its standard deviation). Use the training mean and standard deviation on the validation and the test datasets! After this, the mean of each feature in the subsets should be around 0 and the standard deviation should be around 1.\n",
    "\n",
    "For a finer breakdown of these steps, see ```one-layer.ipynb```."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "detected-trace",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "There are 10 labels which are: \n",
      "[0 1 2 3 4 5 6 7 8 9]\n",
      "The images are of shape: (32, 32, 3)\n",
      "The set sizes are train: 10000, validation: 10000, and test: 10000\n",
      "The class ratios in the training set are: [0.1005 0.0974 0.1032 0.1016 0.0999 0.0937 0.103  0.1001 0.1025 0.0981]\n",
      "The class ratios in the validation set are: [0.0984 0.1007 0.101  0.0995 0.101  0.0988 0.1008 0.1026 0.0987 0.0985]\n",
      "The class ratios in the test set are: [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1]\n",
      "The means of the training set: [ 1.31672451e-16  5.32907052e-18  3.59934305e-17 ... -1.10966791e-16\n",
      " -1.01096909e-16  2.62401212e-17] (#features = 3072)\n",
      "The stds of the training set: [1. 1. 1. ... 1. 1. 1.] (#features = 3072)\n",
      "The means of the validation set: [-0.02151247 -0.01731957 -0.01718778 ... -0.00928345 -0.00494739\n",
      " -0.00583879]\n",
      "The stds of the validation set: [0.9972366  0.99489016 0.99599598 ... 0.99822343 0.98979069 0.99003225]\n",
      "The means of the test set: [-0.01997713 -0.01194233 -0.00644388 ... -0.01584402 -0.00696293\n",
      "  0.00331293]\n",
      "The stds of the test set: [0.99457784 0.98900477 0.98955298 ... 0.98649573 0.97809437 0.98249776]\n"
     ]
    }
   ],
   "source": [
    "# train set is batch 1, val set is batch 2, test set is test\n",
    "path = os.path.join(\"data\", \"data_batch_1\")\n",
    "x_train_img, y_train = load_cfar10_batch(path)\n",
    "\n",
    "path = os.path.join(\"data\", \"data_batch_2\")\n",
    "x_val_img, y_val = load_cfar10_batch(path)\n",
    "\n",
    "path = os.path.join(\"data\", \"test_batch\")\n",
    "x_test_img, y_test = load_cfar10_batch(path)\n",
    "\n",
    "print(f\"There are {np.unique(y_train).size} labels which are: \\n{np.unique(y_train)}\")\n",
    "\n",
    "# check counts in datasets\n",
    "print(f\"The images are of shape: {x_train_img.shape[1:]}\")\n",
    "print(f\"The set sizes are train: {x_train_img.shape[0]}, \"\n",
    "      f\"validation: {x_val_img.shape[0]}, and test: {x_test_img.shape[0]}\")\n",
    "\n",
    "# assert balanced dataset\n",
    "train_counts = np.unique(y_train, return_counts=True)[1]\n",
    "train_ratios = train_counts / train_counts.sum()\n",
    "\n",
    "val_counts = np.unique(y_val, return_counts=True)[1]\n",
    "val_ratios = val_counts / val_counts.sum()\n",
    "\n",
    "test_counts = np.unique(y_test, return_counts=True)[1]\n",
    "test_ratios = test_counts / test_counts.sum()\n",
    "\n",
    "# This is approximating the ratios, good enough\n",
    "np.testing.assert_allclose(train_ratios, val_ratios, rtol=1e-1, atol=0)\n",
    "np.testing.assert_allclose(val_ratios, test_ratios, rtol=1e-1, atol=0)\n",
    "\n",
    "print(f\"The class ratios in the training set are: {train_ratios}\")\n",
    "print(f\"The class ratios in the validation set are: {val_ratios}\")\n",
    "print(f\"The class ratios in the test set are: {test_ratios}\")\n",
    "\n",
    "x_train_un = x_train_img.reshape(x_train_img.shape[0], -1)\n",
    "x_val_un = x_val_img.reshape(x_val_img.shape[0], -1)\n",
    "x_test_un = x_test_img.reshape(x_test_img.shape[0], -1)\n",
    "\n",
    "#x_train = x_train_un / 255.\n",
    "#x_val = x_val_un / 255.\n",
    "#x_test = x_test_un / 255.\n",
    "\n",
    "mean = np.mean(x_train_un, axis=0).reshape(1, x_train_un.shape[1])\n",
    "std = np.std(x_train_un, axis=0).reshape(1, x_train_un.shape[1])\n",
    "\n",
    "x_train = (x_train_un - mean) / std\n",
    "x_val = (x_val_un - mean) / std\n",
    "x_test = (x_test_un - mean) / std\n",
    "\n",
    "print(f\"The means of the training set: {np.mean(x_train, axis=0)} \"\n",
    "      f\"(#features = {np.mean(x_train, axis=0).size})\")\n",
    "print(f\"The stds of the training set: {np.std(x_train, axis=0)} \"\n",
    "      f\"(#features = {np.mean(x_train, axis=0).size})\")\n",
    "\n",
    "print(f\"The means of the validation set: {np.mean(x_val, axis=0)}\")\n",
    "print(f\"The stds of the validation set: {np.std(x_val, axis=0)}\")\n",
    "\n",
    "print(f\"The means of the test set: {np.mean(x_test, axis=0)}\")\n",
    "print(f\"The stds of the test set: {np.std(x_test, axis=0)}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "natural-investing",
   "metadata": {},
   "source": [
    "## Numerical Gradient Check <a class=\"anchor\" id=\"numerical-gradient-check\"></a>\n",
    "\n",
    "The correctness of the gradient computation and backpropagation was assured via numerical gradient check. I implemented the double centered gradient check with the ```numerical_gradient_check_model``` that computes the analytical and numerical gradients, and then compares them in each layer in the model. As outlined in [CS231n: Convolutional Neural Networks for Visual Recognition](https://cs231n.github.io/neural-networks-3/#gradcheck), it is satisfactory to assert that the analytical gradients of the model are correct based on the values shwon below . To avoid kinks in the objective function, only a few data points were used in the gradient check. The step size for computing the numerical gradient was set to 1e-06. In computing the relative error, the formula provided in [CS231n: Convolutional Neural Networks for Visual Recognition](https://cs231n.github.io/neural-networks-3/#gradcheck) was used. Furthermore, the ```numpy.testing.assert\\_array\\_almost\\_equal``` function was also used to make the comparison."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "difficult-thunder",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "network with reg. rate = 0.6000\n",
      "layer=0, param_name=w\n",
      "max rel error=8.604779596268731e-07\n",
      "layer=0, param_name=b\n",
      "max rel error=6.4432917294973935e-09\n",
      "layer=1, param_name=w\n",
      "max rel error=8.890298402076154e-08\n",
      "layer=1, param_name=b\n",
      "max rel error=3.840511462274298e-09\n",
      "test_grad_check passed\n",
      "\n",
      "network with reg. rate = 0.7000\n",
      "layer=0, param_name=w\n",
      "max rel error=1.9550366957949666e-07\n",
      "layer=0, param_name=b\n",
      "max rel error=6.4432917294973935e-09\n",
      "layer=1, param_name=w\n",
      "max rel error=1.6942324342327626e-07\n",
      "layer=1, param_name=b\n",
      "max rel error=3.840511462274298e-09\n",
      "test_grad_check passed\n",
      "\n",
      "network with reg. rate = 0.9000\n",
      "layer=0, param_name=w\n",
      "max rel error=3.700825705542563e-07\n",
      "layer=0, param_name=b\n",
      "max rel error=6.4432917294973935e-09\n",
      "layer=1, param_name=w\n",
      "max rel error=1.47348804853943e-07\n",
      "layer=1, param_name=b\n",
      "max rel error=3.840511462274298e-09\n",
      "test_grad_check passed\n",
      "\n",
      "network with reg. rate = 0.7000\n",
      "layer=0, param_name=w\n",
      "max rel error=1.9550366957949666e-07\n",
      "layer=0, param_name=b\n",
      "max rel error=6.4432917294973935e-09\n",
      "layer=1, param_name=w\n",
      "max rel error=1.6942324342327626e-07\n",
      "layer=1, param_name=b\n",
      "max rel error=3.840511462274298e-09\n",
      "test_grad_check passed\n",
      "\n",
      "network with reg. rate = 0.2000\n",
      "layer=0, param_name=w\n",
      "max rel error=9.811004013439056e-07\n",
      "layer=0, param_name=b\n",
      "max rel error=6.4432917294973935e-09\n",
      "layer=1, param_name=w\n",
      "max rel error=2.471622680768831e-07\n",
      "layer=1, param_name=b\n",
      "max rel error=3.840511462274298e-09\n",
      "test_grad_check passed\n",
      "\n"
     ]
    }
   ],
   "source": [
    "def build_model(reg_rate_l2, in_dim, mid_dim, out_dim, seed):\n",
    "    \n",
    "    # kernel and bias initializer params, normal distribution with Xavier os std is None (adaptively computed)\n",
    "    init_params = {\"coeff\": 1.0, \"mean\": 0.0, \"std\": None}\n",
    "\n",
    "    # define the dense layer\n",
    "    dense_1 = \\\n",
    "        Dense(in_dim=in_dim, out_dim=mid_dim, \n",
    "              kernel_initializer=XavierInitializer(seed=seed, **init_params), \n",
    "              bias_initializer=XavierInitializer(seed=seed+1, **init_params), \n",
    "              kernel_regularizer=L2Regularizer(reg_rate=reg_rate_l2), \n",
    "              activation=ReLUActivation()\n",
    "             )\n",
    "\n",
    "    dense_2 = \\\n",
    "        Dense(in_dim=mid_dim, out_dim=out_dim,\n",
    "              kernel_initializer=XavierInitializer(seed=seed+2, **init_params), \n",
    "              bias_initializer=XavierInitializer(seed=seed+3, **init_params), \n",
    "              kernel_regularizer=L2Regularizer(reg_rate=reg_rate_l2), \n",
    "              activation=SoftmaxActivation()\n",
    "             )\n",
    "\n",
    "    layers = [\n",
    "        dense_1,\n",
    "        dense_2\n",
    "    ]\n",
    "\n",
    "    model = Model(layers)\n",
    "\n",
    "    # define the loss\n",
    "    loss = CategoricalCrossEntropyLoss(loss_smoother=LossSmootherConstant())\n",
    "    \n",
    "    return model, loss\n",
    "\n",
    "\n",
    "def test_model(x, y, seed=6):\n",
    "    in_dim = x.shape[1]\n",
    "    mid_dim = 50\n",
    "    out_dim = np.unique(y_train).size\n",
    "\n",
    "    build_model_loss_func_dict = {\n",
    "        \"model\": build_model\n",
    "    }\n",
    "\n",
    "    np.random.seed(seed + 3)\n",
    "    reg_rates = 10e-2 * np.random.randint(low=1, high=10, size=5)\n",
    "\n",
    "    for reg_rate in reg_rates:\n",
    "        for key, build_model_func in build_model_loss_func_dict.items():\n",
    "            print(f\"network with reg. rate = {reg_rate:.4f}\")\n",
    "            model, loss = build_model_func(reg_rate, in_dim, mid_dim, out_dim, seed)\n",
    "            numerical_gradient_check_model(x, y, model, loss)\n",
    "            print()\n",
    "            \n",
    "test_model(x_train[:2, :10], y_train[:2])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "proper-frequency",
   "metadata": {},
   "source": [
    "## Training and Results <a class=\"anchor\" id=\"training-and-results\"></a>\n",
    "\n",
    "A two-layer network is trained with ReLu activations, Xavier initilaization, L2 regularization, and a cyclical learning rate schedule.\n",
    "\n",
    "The training data is ```data_batch_1```, the validation data is ```data_batch_2```, and the testing data is ```test_batch```. All of the training, validation, and test set included 10000 images. The data was pre-processed as described earlier. For the cyclical learning rate schedule, we need the number of data points in the training set.  \n",
    "\n",
    "The ```LRCyclingSchedule``` cyclical learning rate schedule's base learning rate was 1e-5 and the maximum learning rate was 1e-1. The step size (half-cycle) of the ```LRCyclingSchedule``` was n\\_s = 800 (i.e.: one fully cycle is 1600 steps). The training on the training set of 10000 images went on for 48 epochs with a mini-batch size of 100 that is equivalent to 48 000 / 100 = 4800 update steps, or equivalently for 3 full-cycles."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "alpine-reset",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "step size of cyc. lr: 800 update steps\n",
      "full cycle of cyc.lr : 1600 update steps\n",
      "16.0 epochs = 1 full cycle = 1600 update steps\n",
      "3 cycle = 48 epochs = 4800 update steps\n"
     ]
    }
   ],
   "source": [
    "batch_size = 100\n",
    "\n",
    "step_size = int(8*np.floor(x_train.shape[0] / batch_size))\n",
    "print(f\"step size of cyc. lr: {step_size} update steps\")\n",
    "\n",
    "cycle_steps = 2*step_size\n",
    "print(f\"full cycle of cyc.lr : {cycle_steps} update steps\")\n",
    "\n",
    "epochs_one_full_cycle = (cycle_steps * batch_size) / x_train.shape[0]\n",
    "print(f\"{epochs_one_full_cycle} epochs = 1 full cycle = {cycle_steps} update steps\")\n",
    "\n",
    "n_cycle = 3\n",
    "n_epochs = int(n_cycle*epochs_one_full_cycle)\n",
    "print(f\"{n_cycle} cycle = {n_epochs} epochs = {n_cycle*cycle_steps} update steps\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "played-expense",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "model summary: \n",
      "layer 0: dense: \n",
      "\t shape -- in: 3072, out: 50\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.018042^2), reg: l2 with 2.5000e-02\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: relu\n",
      "\n",
      "layer 1: dense: \n",
      "\t shape -- in: 50, out: 10\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.141421^2), reg: l2 with 2.5000e-02\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: softmax\n",
      "\n",
      "categorical cross-entropy loss with loss smoother constant\n",
      "sgd with cycling lr schedule and clipper who does nothing\n",
      "\n",
      "starting epoch: 1 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 72.12it/s]\n",
      "epoch 1/48 \n",
      " \t -- train loss = 1.877934537240929 / train cost = 2.6066589455396585\n",
      "\t -- {\"accuracy_train\": 0.3417} \n",
      "\t -- val loss = 1.93937876576476 / val cost = 2.6681031740634893\n",
      "\t -- {\"accuracy_val\": 0.3172} \n",
      "\n",
      "\n",
      "starting epoch: 2 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 72.09it/s]\n",
      "epoch 2/48 \n",
      " \t -- train loss = 1.686525858267782 / train cost = 2.359775230345653\n",
      "\t -- {\"accuracy_train\": 0.411} \n",
      "\t -- val loss = 1.8015764640320848 / val cost = 2.474825836109956\n",
      "\t -- {\"accuracy_val\": 0.365} \n",
      "\n",
      "\n",
      "starting epoch: 3 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 69.66it/s]\n",
      "epoch 3/48 \n",
      " \t -- train loss = 1.5509888255004947 / train cost = 2.143277777061135\n",
      "\t -- {\"accuracy_train\": 0.4602} \n",
      "\t -- val loss = 1.7129603939095654 / val cost = 2.3052493454702057\n",
      "\t -- {\"accuracy_val\": 0.3991} \n",
      "\n",
      "\n",
      "starting epoch: 4 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 72.32it/s]\n",
      "epoch 4/48 \n",
      " \t -- train loss = 1.5166295344861647 / train cost = 2.0185511405918577\n",
      "\t -- {\"accuracy_train\": 0.4714} \n",
      "\t -- val loss = 1.7113717596753035 / val cost = 2.213293365780997\n",
      "\t -- {\"accuracy_val\": 0.4009} \n",
      "\n",
      "\n",
      "starting epoch: 5 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.38it/s]\n",
      "epoch 5/48 \n",
      " \t -- train loss = 1.4808396892312703 / train cost = 1.8980300519254047\n",
      "\t -- {\"accuracy_train\": 0.4829} \n",
      "\t -- val loss = 1.695937854143655 / val cost = 2.113128216837789\n",
      "\t -- {\"accuracy_val\": 0.4037} \n",
      "\n",
      "\n",
      "starting epoch: 6 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 71.67it/s]\n",
      "epoch 6/48 \n",
      " \t -- train loss = 1.5263257669794068 / train cost = 1.8718079410232646\n",
      "\t -- {\"accuracy_train\": 0.462} \n",
      "\t -- val loss = 1.7469594072999046 / val cost = 2.0924415813437625\n",
      "\t -- {\"accuracy_val\": 0.3833} \n",
      "\n",
      "\n",
      "starting epoch: 7 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 69.64it/s]\n",
      "epoch 7/48 \n",
      " \t -- train loss = 1.4768988394779616 / train cost = 1.7686306660925848\n",
      "\t -- {\"accuracy_train\": 0.4818} \n",
      "\t -- val loss = 1.712422444516535 / val cost = 2.004154271131158\n",
      "\t -- {\"accuracy_val\": 0.3952} \n",
      "\n",
      "\n",
      "starting epoch: 8 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 80.40it/s]\n",
      "epoch 8/48 \n",
      " \t -- train loss = 1.4363425842700834 / train cost = 1.6897014698308053\n",
      "\t -- {\"accuracy_train\": 0.4955} \n",
      "\t -- val loss = 1.6709842568352629 / val cost = 1.9243431423959847\n",
      "\t -- {\"accuracy_val\": 0.4042} \n",
      "\n",
      "\n",
      "starting epoch: 9 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 90.15it/s]\n",
      "epoch 9/48 \n",
      " \t -- train loss = 1.4086599108477633 / train cost = 1.6421543854815093\n",
      "\t -- {\"accuracy_train\": 0.509} \n",
      "\t -- val loss = 1.661781078960803 / val cost = 1.895275553594549\n",
      "\t -- {\"accuracy_val\": 0.4154} \n",
      "\n",
      "\n",
      "starting epoch: 10 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 86.89it/s]\n",
      "epoch 10/48 \n",
      " \t -- train loss = 1.3591101100596692 / train cost = 1.5862853410437\n",
      "\t -- {\"accuracy_train\": 0.528} \n",
      "\t -- val loss = 1.6407296094234618 / val cost = 1.8679048404074925\n",
      "\t -- {\"accuracy_val\": 0.4239} \n",
      "\n",
      "\n",
      "starting epoch: 11 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 93.18it/s]\n",
      "epoch 11/48 \n",
      " \t -- train loss = 1.3051873463634884 / train cost = 1.5326399750858788\n",
      "\t -- {\"accuracy_train\": 0.556} \n",
      "\t -- val loss = 1.6001581995387144 / val cost = 1.8276108282611048\n",
      "\t -- {\"accuracy_val\": 0.4386} \n",
      "\n",
      "\n",
      "starting epoch: 12 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 64.10it/s]\n",
      "epoch 12/48 \n",
      " \t -- train loss = 1.276836029216563 / train cost = 1.5080649245243487\n",
      "\t -- {\"accuracy_train\": 0.5579} \n",
      "\t -- val loss = 1.6035193342249936 / val cost = 1.8347482295327793\n",
      "\t -- {\"accuracy_val\": 0.4372} \n",
      "\n",
      "\n",
      "starting epoch: 13 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 76.54it/s]\n",
      "epoch 13/48 \n",
      " \t -- train loss = 1.2324111003084817 / train cost = 1.4681841596199967\n",
      "\t -- {\"accuracy_train\": 0.5831} \n",
      "\t -- val loss = 1.578465460007519 / val cost = 1.814238519319034\n",
      "\t -- {\"accuracy_val\": 0.4488} \n",
      "\n",
      "\n",
      "starting epoch: 14 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.72it/s]\n",
      "epoch 14/48 \n",
      " \t -- train loss = 1.1941477505736648 / train cost = 1.4343588504049758\n",
      "\t -- {\"accuracy_train\": 0.604} \n",
      "\t -- val loss = 1.5528410852056205 / val cost = 1.7930521850369314\n",
      "\t -- {\"accuracy_val\": 0.4577} \n",
      "\n",
      "\n",
      "starting epoch: 15 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:02<00:00, 40.30it/s]\n",
      "epoch 15/48 \n",
      " \t -- train loss = 1.1659873222009676 / train cost = 1.409764636221284\n",
      "\t -- {\"accuracy_train\": 0.6166} \n",
      "\t -- val loss = 1.5399103789476853 / val cost = 1.7836876929680017\n",
      "\t -- {\"accuracy_val\": 0.4608} \n",
      "\n",
      "\n",
      "starting epoch: 16 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 54.76it/s]\n",
      "epoch 16/48 \n",
      " \t -- train loss = 1.154524582471445 / train cost = 1.3999868411697656\n",
      "\t -- {\"accuracy_train\": 0.6204} \n",
      "\t -- val loss = 1.537323596973523 / val cost = 1.7827858556718437\n",
      "\t -- {\"accuracy_val\": 0.4642} \n",
      "\n",
      "\n",
      "starting epoch: 17 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 61.29it/s]\n",
      "epoch 17/48 \n",
      " \t -- train loss = 1.1566572423787294 / train cost = 1.4036349622299789\n",
      "\t -- {\"accuracy_train\": 0.619} \n",
      "\t -- val loss = 1.538284186405923 / val cost = 1.7852619062571724\n",
      "\t -- {\"accuracy_val\": 0.4622} \n",
      "\n",
      "\n",
      "starting epoch: 18 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 79.82it/s]\n",
      "epoch 18/48 \n",
      " \t -- train loss = 1.171908006195833 / train cost = 1.422263044695139\n",
      "\t -- {\"accuracy_train\": 0.6014} \n",
      "\t -- val loss = 1.564924392278839 / val cost = 1.815279430778145\n",
      "\t -- {\"accuracy_val\": 0.4544} \n",
      "\n",
      "\n",
      "starting epoch: 19 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 85.08it/s]\n",
      "epoch 19/48 \n",
      " \t -- train loss = 1.193374569634954 / train cost = 1.4482120148651285\n",
      "\t -- {\"accuracy_train\": 0.598} \n",
      "\t -- val loss = 1.5944077890193005 / val cost = 1.8492452342494747\n",
      "\t -- {\"accuracy_val\": 0.4422} \n",
      "\n",
      "\n",
      "starting epoch: 20 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 91.05it/s]\n",
      "epoch 20/48 \n",
      " \t -- train loss = 1.2011315382467929 / train cost = 1.4596673435909668\n",
      "\t -- {\"accuracy_train\": 0.5922} \n",
      "\t -- val loss = 1.598968127508331 / val cost = 1.857503932852505\n",
      "\t -- {\"accuracy_val\": 0.4474} \n",
      "\n",
      "\n",
      "starting epoch: 21 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 83.18it/s]\n",
      "epoch 21/48 \n",
      " \t -- train loss = 1.3231564103141282 / train cost = 1.5829221933707176\n",
      "\t -- {\"accuracy_train\": 0.5407} \n",
      "\t -- val loss = 1.7078421816336538 / val cost = 1.9676079646902433\n",
      "\t -- {\"accuracy_val\": 0.4057} \n",
      "\n",
      "\n",
      "starting epoch: 22 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 80.56it/s]\n",
      "epoch 22/48 \n",
      " \t -- train loss = 1.3148610866805261 / train cost = 1.5743227510141462\n",
      "\t -- {\"accuracy_train\": 0.5435} \n",
      "\t -- val loss = 1.711329251022461 / val cost = 1.970790915356081\n",
      "\t -- {\"accuracy_val\": 0.4102} \n",
      "\n",
      "\n",
      "starting epoch: 23 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 66.33it/s]\n",
      "epoch 23/48 \n",
      " \t -- train loss = 1.3409448283071546 / train cost = 1.5967094592592412\n",
      "\t -- {\"accuracy_train\": 0.5247} \n",
      "\t -- val loss = 1.6999881773575312 / val cost = 1.9557528083096178\n",
      "\t -- {\"accuracy_val\": 0.4069} \n",
      "\n",
      "\n",
      "starting epoch: 24 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.38it/s]\n",
      "epoch 24/48 \n",
      " \t -- train loss = 1.3304064146366525 / train cost = 1.580623496353543\n",
      "\t -- {\"accuracy_train\": 0.5419} \n",
      "\t -- val loss = 1.6511340832776393 / val cost = 1.9013511649945298\n",
      "\t -- {\"accuracy_val\": 0.4268} \n",
      "\n",
      "\n",
      "starting epoch: 25 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 65.94it/s]\n",
      "epoch 25/48 \n",
      " \t -- train loss = 1.2709943081763355 / train cost = 1.5220875310214286\n",
      "\t -- {\"accuracy_train\": 0.5668} \n",
      "\t -- val loss = 1.6155913187832598 / val cost = 1.8666845416283528\n",
      "\t -- {\"accuracy_val\": 0.434} \n",
      "\n",
      "\n",
      "starting epoch: 26 ...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 63.50it/s]\n",
      "epoch 26/48 \n",
      " \t -- train loss = 1.2589578465182971 / train cost = 1.5127400187862357\n",
      "\t -- {\"accuracy_train\": 0.5639} \n",
      "\t -- val loss = 1.6234433640083938 / val cost = 1.8772255362763324\n",
      "\t -- {\"accuracy_val\": 0.432} \n",
      "\n",
      "\n",
      "starting epoch: 27 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 63.68it/s]\n",
      "epoch 27/48 \n",
      " \t -- train loss = 1.2305698445653181 / train cost = 1.4912248658394298\n",
      "\t -- {\"accuracy_train\": 0.5772} \n",
      "\t -- val loss = 1.5987669614216513 / val cost = 1.859421982695763\n",
      "\t -- {\"accuracy_val\": 0.4442} \n",
      "\n",
      "\n",
      "starting epoch: 28 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 72.59it/s]\n",
      "epoch 28/48 \n",
      " \t -- train loss = 1.1762081238092186 / train cost = 1.4410945977345335\n",
      "\t -- {\"accuracy_train\": 0.5994} \n",
      "\t -- val loss = 1.5676381237532242 / val cost = 1.8325245976785391\n",
      "\t -- {\"accuracy_val\": 0.4568} \n",
      "\n",
      "\n",
      "starting epoch: 29 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 70.14it/s]\n",
      "epoch 29/48 \n",
      " \t -- train loss = 1.1507813552843493 / train cost = 1.4208250400977285\n",
      "\t -- {\"accuracy_train\": 0.6156} \n",
      "\t -- val loss = 1.5626182796493366 / val cost = 1.8326619644627158\n",
      "\t -- {\"accuracy_val\": 0.4542} \n",
      "\n",
      "\n",
      "starting epoch: 30 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.10it/s]\n",
      "epoch 30/48 \n",
      " \t -- train loss = 1.1090126817198072 / train cost = 1.3839547828986618\n",
      "\t -- {\"accuracy_train\": 0.6396} \n",
      "\t -- val loss = 1.5429413234709448 / val cost = 1.8178834246497995\n",
      "\t -- {\"accuracy_val\": 0.4636} \n",
      "\n",
      "\n",
      "starting epoch: 31 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 66.72it/s]\n",
      "epoch 31/48 \n",
      " \t -- train loss = 1.0877790690659297 / train cost = 1.3665182562733111\n",
      "\t -- {\"accuracy_train\": 0.6437} \n",
      "\t -- val loss = 1.5327670363052486 / val cost = 1.81150622351263\n",
      "\t -- {\"accuracy_val\": 0.4616} \n",
      "\n",
      "\n",
      "starting epoch: 32 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.93it/s]\n",
      "epoch 32/48 \n",
      " \t -- train loss = 1.0730691215411419 / train cost = 1.3529396364237671\n",
      "\t -- {\"accuracy_train\": 0.6537} \n",
      "\t -- val loss = 1.523658842441614 / val cost = 1.8035293573242392\n",
      "\t -- {\"accuracy_val\": 0.4672} \n",
      "\n",
      "\n",
      "starting epoch: 33 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 66.07it/s]\n",
      "epoch 33/48 \n",
      " \t -- train loss = 1.07949044239803 / train cost = 1.360684085952728\n",
      "\t -- {\"accuracy_train\": 0.6535} \n",
      "\t -- val loss = 1.5313677358271105 / val cost = 1.8125613793818085\n",
      "\t -- {\"accuracy_val\": 0.4663} \n",
      "\n",
      "\n",
      "starting epoch: 34 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 68.26it/s]\n",
      "epoch 34/48 \n",
      " \t -- train loss = 1.1016764402796484 / train cost = 1.3864397321788828\n",
      "\t -- {\"accuracy_train\": 0.6435} \n",
      "\t -- val loss = 1.556724221229677 / val cost = 1.8414875131289115\n",
      "\t -- {\"accuracy_val\": 0.4613} \n",
      "\n",
      "\n",
      "starting epoch: 35 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.61it/s]\n",
      "epoch 35/48 \n",
      " \t -- train loss = 1.1180075236037257 / train cost = 1.4066934197922643\n",
      "\t -- {\"accuracy_train\": 0.6319} \n",
      "\t -- val loss = 1.5764901977616466 / val cost = 1.865176093950185\n",
      "\t -- {\"accuracy_val\": 0.4514} \n",
      "\n",
      "\n",
      "starting epoch: 36 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 63.65it/s]\n",
      "epoch 36/48 \n",
      " \t -- train loss = 1.1214221167313732 / train cost = 1.4126178322882974\n",
      "\t -- {\"accuracy_train\": 0.6228} \n",
      "\t -- val loss = 1.5919889712478468 / val cost = 1.883184686804771\n",
      "\t -- {\"accuracy_val\": 0.444} \n",
      "\n",
      "\n",
      "starting epoch: 37 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 70.04it/s]\n",
      "epoch 37/48 \n",
      " \t -- train loss = 1.1635853064159047 / train cost = 1.4542017473012678\n",
      "\t -- {\"accuracy_train\": 0.6016} \n",
      "\t -- val loss = 1.6196746175911931 / val cost = 1.9102910584765564\n",
      "\t -- {\"accuracy_val\": 0.4358} \n",
      "\n",
      "\n",
      "starting epoch: 38 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.08it/s]\n",
      "epoch 38/48 \n",
      " \t -- train loss = 1.2386385914492235 / train cost = 1.5278395374799136\n",
      "\t -- {\"accuracy_train\": 0.5725} \n",
      "\t -- val loss = 1.6685422103315446 / val cost = 1.9577431563622347\n",
      "\t -- {\"accuracy_val\": 0.4271} \n",
      "\n",
      "\n",
      "starting epoch: 39 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.03it/s]\n",
      "epoch 39/48 \n",
      " \t -- train loss = 1.255208905671776 / train cost = 1.5360878884466151\n",
      "\t -- {\"accuracy_train\": 0.5638} \n",
      "\t -- val loss = 1.6582119231523957 / val cost = 1.939090905927235\n",
      "\t -- {\"accuracy_val\": 0.4189} \n",
      "\n",
      "\n",
      "starting epoch: 40 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 72.33it/s]\n",
      "epoch 40/48 \n",
      " \t -- train loss = 1.317606736077483 / train cost = 1.5912511154756641\n",
      "\t -- {\"accuracy_train\": 0.5353} \n",
      "\t -- val loss = 1.6941600107521824 / val cost = 1.9678043901503635\n",
      "\t -- {\"accuracy_val\": 0.4078} \n",
      "\n",
      "\n",
      "starting epoch: 41 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 69.24it/s]\n",
      "epoch 41/48 \n",
      " \t -- train loss = 1.226462451369888 / train cost = 1.495659190944959\n",
      "\t -- {\"accuracy_train\": 0.5831} \n",
      "\t -- val loss = 1.6073880201945823 / val cost = 1.8765847597696532\n",
      "\t -- {\"accuracy_val\": 0.4384} \n",
      "\n",
      "\n",
      "starting epoch: 42 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 66.12it/s]\n",
      "epoch 42/48 \n",
      " \t -- train loss = 1.2811114612735541 / train cost = 1.5532485242373153\n",
      "\t -- {\"accuracy_train\": 0.5479} \n",
      "\t -- val loss = 1.663424764178018 / val cost = 1.935561827141779\n",
      "\t -- {\"accuracy_val\": 0.4165} \n",
      "\n",
      "\n",
      "starting epoch: 43 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.62it/s]\n",
      "epoch 43/48 \n",
      " \t -- train loss = 1.2276012807639782 / train cost = 1.5045690386142279\n",
      "\t -- {\"accuracy_train\": 0.574} \n",
      "\t -- val loss = 1.6242599223196195 / val cost = 1.901227680169869\n",
      "\t -- {\"accuracy_val\": 0.4371} \n",
      "\n",
      "\n",
      "starting epoch: 44 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.22it/s]\n",
      "epoch 44/48 \n",
      " \t -- train loss = 1.1378535836628119 / train cost = 1.4197870457936097\n",
      "\t -- {\"accuracy_train\": 0.6169} \n",
      "\t -- val loss = 1.5538253206137878 / val cost = 1.8357587827445858\n",
      "\t -- {\"accuracy_val\": 0.4541} \n",
      "\n",
      "\n",
      "starting epoch: 45 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 66.08it/s]\n",
      "epoch 45/48 \n",
      " \t -- train loss = 1.1024585109839162 / train cost = 1.3893077860019072\n",
      "\t -- {\"accuracy_train\": 0.6371} \n",
      "\t -- val loss = 1.5478556173404323 / val cost = 1.8347048923584233\n",
      "\t -- {\"accuracy_val\": 0.4585} \n",
      "\n",
      "\n",
      "starting epoch: 46 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.57it/s]\n",
      "epoch 46/48 \n",
      " \t -- train loss = 1.0734025074284819 / train cost = 1.3646117638141726\n",
      "\t -- {\"accuracy_train\": 0.6518} \n",
      "\t -- val loss = 1.5330962142493558 / val cost = 1.8243054706350466\n",
      "\t -- {\"accuracy_val\": 0.4643} \n",
      "\n",
      "\n",
      "starting epoch: 47 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 67.56it/s]\n",
      "epoch 47/48 \n",
      " \t -- train loss = 1.052573735500182 / train cost = 1.347257871394629\n",
      "\t -- {\"accuracy_train\": 0.6598} \n",
      "\t -- val loss = 1.5253137809192068 / val cost = 1.8199979168136537\n",
      "\t -- {\"accuracy_val\": 0.468} \n",
      "\n",
      "\n",
      "starting epoch: 48 ...\n",
      "batch 100/100: 100%|██████████| 100/100 [00:01<00:00, 68.58it/s]\n",
      "epoch 48/48 \n",
      " \t -- train loss = 1.040124208315318 / train cost = 1.3363510990069802\n",
      "\t -- {\"accuracy_train\": 0.6702} \n",
      "\t -- val loss = 1.5181348630003424 / val cost = 1.8143617536920047\n",
      "\t -- {\"accuracy_val\": 0.4713} \n",
      "\n",
      "\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEWCAYAAAB8LwAVAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAAsTAAALEwEAmpwYAABQXklEQVR4nO3dd3hUZfbA8e9JI4WSECSUAAm99yqgwQoKomtDUewsdl11LetPXdd1dV1X11UXUbGiiIiK0kUiIC2A9JbQklBTSEhCet7fH3ciQ5gkk5DJZDLn8zzzZObOLefeJHPmvlWMMSillFJl+bg7AKWUUnWTJgillFIOaYJQSinlkCYIpZRSDmmCUEop5ZAmCKWUUg5pglBKVUpEbheRle6OQ9UuTRDKrUQkVkROiEgDd8eilDqTJgjlNiISBYwEDHBVLR/brzaPp5Qn0gSh3GkSsAb4GLjN/g0RaSMic0QkRUTSRORtu/fuEZGdIpIlIjtEpL9tuRGRjnbrfSwiL9mex4hIsog8KSJHgY9EJExEfrQd44TteaTd9k1F5CMROWx7/zvb8m0iMs5uPX8RSRWRvmVP0BbnWLvXfrZ1+4tIoIh8bju/DBGJE5EIZy6ciAwVkVW27TaLSIzde7Ei8g8RWScimSLyvYg0tXv/KhHZbts2VkS6OXPdbe//y3Yt9ovIGLvlt4vIPtvvZL+ITHTmPFTdpglCudMkYIbtcXnph6OI+AI/AgeBKKA1MNP23vXAC7ZtG2PdeaQ5ebwWQFOgHTAZ6+//I9vrtkAuYP+B+BkQDPQAmgNv2JZ/Ctxit94VwBFjzCYHx/wSuMnu9eVAqjFmI1ZSbAK0AcKBKbYYKiQirYF5wEu283kc+EZEzrNbbRJwJ9AKKALesm3b2RbTI8B5wHzgBxEJqOi62wwBdgPNgH8CH4olxLb/McaYRsD5gKNroTyNMUYf+qj1BzACKASa2V7vAh61PR8GpAB+DrZbBDxczj4N0NHu9cfAS7bnMUABEFhBTH2BE7bnLYESIMzBeq2ALKCx7fVs4M/l7LOjbd1g2+sZwHO253cCq4DeVbx2TwKfObgut9mexwKv2L3X3XbuvsD/AbPs3vMBDtmuT0XX/XYgwe51sO16twBCgAzgWiDI3X9b+qi5h95BKHe5DVhsjEm1vf6C08VMbYCDxpgiB9u1AfZW85gpxpi80hciEiwi74nIQRE5CSwHQm3fpNsA6caYE2V3Yow5DPwKXCsiocAYrA/+sxhjEoCdwDgRCca64/nC9vZnWB/sM23FWP8UEX8nzqMdcL2tiChDRDKwEm5Lu3WS7J4fBPyxvvm3sr0uja/Etm5rKr7uAEfttjtle9rQGJMD3Ih1B3REROaJSFcnzkPVcVpRp2qdiAQBNwC+tvoAgAZYH859sD6w2oqIn4MPqySgQzm7PoX1zbZUCyDZ7nXZoYsfA7oAQ4wxR211CL8BYjtOUxEJNcZkODjWJ8DdWP9Dq40xh8o7X04XM/kAO2xJA2NMIfBX4K+2Cvv5WEU4H1awL2yxfWaMuaeCddrYPW+LdbeWChwGepW+ISJiW/cQkE/5171CxphFwCLb7/Yl4H2sBgjKg+kdhHKHq4FirKKPvrZHN2AFVtn5OuAI8IqIhNgqc4fbtv0AeFxEBtjKvzuKSDvbe5uAm0XEV0RGAxdWEkcjrDL/DFsl7vOlbxhjjgALgHdtldn+InKB3bbfAf2Bh7HqJCoyE7gMuJfTdw+IyCgR6WW7YzmJ9SFeXMm+AD7HuiO53HaugWJVwkfarXOLiHS33bW8CMw2xhQDs4ArReRi293KY1iJYRUVX/dyiUiEreI7xLavbCfPQ9VxmiCUO9wGfGSMSTTGHC19YFUQT8T6Bj8Oq/w+Eesu4EYAY8zXwN+xPmizsD6oS1voPGzbLsO2n+8qieNNIAjrm/UaYGGZ92/F+tDeBRzHqtjFFkcu8A0QDcyp6CC2ZLMaq/L2K7u3WmDVX5zEKob6BevDHxGZKiJTy9lfEjAeeAarziAJeIIz/58/w6qDOQoEAg/Ztt2NVcH+X9t5jwPGGWMKbAnE4XWvhA9WojkMpGMl5vuc2E7VcWKMThikVHWIyHNAZ2PMLZWuXItEJBb43BjzgbtjUZ5N6yCUqgZbkdRdWHcZStVLWsSkVBWJyD1YxToLjDHL3R2PUq6iRUxKKaUc0jsIpZRSDtWrOohmzZqZqKioam2bk5NDSEhIzQbkYbz9Gnj7+YNeA/C+a7Bhw4ZUY8x5jt6rVwkiKiqK9evXV2vb2NhYYmJiajYgD+Pt18Dbzx/0GoD3XQMROVjee1rEpJRSyiFNEEoppRzSBKGUUsqhelUH4UhhYSHJycnk5eVVuF6TJk3YuXNnLUVV8wIDA4mMjMTf35nBQJVSqnL1PkEkJyfTqFEjoqKisAaudCwrK4tGjRrVYmQ1xxhDWloaycnJREdHuzscpVQ9Ue+LmPLy8ggPD68wOXg6ESE8PLzSuySllKqKep8ggHqdHEp5wzkqpWqXVyQIpZSqrzYmnmDa8upOslgxTRAulpGRwbvvvlvl7a644goyMjJqPiClVL2xdOcxbn5/DV+sTSQnv0qTADpFE4SLlZcgiosrnnBr/vz5hIaGuigqpZSnmxWXxOTPNtCpeSNm33s+IQ1qvs1RvW/F5G5PPfUUe/fupW/fvvj7+9OwYUNatmzJpk2b2LFjB1dffTVJSUnk5eXx8MMPM3nyZOD0sCHZ2dmMGTOGESNGsGrVKlq3bs33339PUFCQm89MKeUOxhjeWZbAvxbvYWSnZky9ZYBLkgN4WYL46w/b2XH4pMP3iouL8fX1rfI+u7dqzPPjepT7/iuvvMK2bdvYtGkTsbGxXHnllWzbtu335qjTp0+nadOm5ObmMmjQIK699lrCw8PP2Ed8fDxffvkl77//PjfccAPffPMNt9xSpyYxU0rVguISwwtzt/PZmoNc0681r17bmwA/1xUEeVWCqAsGDx58Rl+Ft956i2+//RaApKQk4uPjz0oQ0dHR9O3bF4ABAwZw4MCB2gpXKVVH5BUW88jMTSzcfpQ/XtCeJ0d3xcfHta0XvSpBVPRNv7Y6ytkPIxwbG8tPP/3E6tWrCQ4OJiYmxmFfhgYNGvz+3NfXl9zcXJfHqZSqO45m5vHQl7+x7kA6z17ZjbtHtq+V43pVgnCHRo0akZWV5fC9zMxMwsLCCA4OZteuXaxZs6aWo1NK1XU/bD7Ms99tI7+omLdu6sdVfVrV2rE1QbhYeHg4w4cPp2fPngQFBREREfH7e6NHj2bq1Kn07t2bLl26MHToUDdGqpSqSzJPFfJ/329j7ubD9GkTyhs39KH9eQ1rNQaXJggRGQ38B/AFPjDGvOJgnRjgTcAfSDXGXGhbfgDIAoqBImPMQFfG6kpffPGFw+UNGjRgwYIFDt8rrWdo1qwZ27Zt+335448/XuPxKVURYwybkzM5VVCEv68Pvj6Cn4/g5+ODn6/QJiyYoICqN/BQ5VsRn8ITX28hNTufP13amftiOuDnW/u9ElyWIETEF3gHuBRIBuJEZK4xZofdOqHAu8BoY0yiiDQvs5tRxphUV8WolKpYYXEJz367ja/WJ5W7TqsmgUy/YxBdWzSuxcjqp9yCYl5ZsJNPVh+kw3khTJt0Pr0jQ90WjyvvIAYDCcaYfQAiMhMYD+ywW+dmYI4xJhHAGHPchfEopargZF4h98/YyIr4VO6N6cAFnc6juMRQVFJCUbGhqMSQk1/EPxft4rr/rebdif25oLPDqY2VE0pKDPfN2MCy3Sncfn4UT43pSqC/e+/MXJkgWgP2XzuSgSFl1ukM+ItILNAI+I8x5lPbewZYLCIGeM8YM82FsSql7BzKyOXOj+LYm5LNP6/rzQ0D25S77vkdw7njozju+DiOl67uyU2D29ZipPXHf5bGs2x3Ci+O78GkYVHuDgcAMca4Zsci1wOXG2Putr2+FRhsjHnQbp23gYHAxUAQsBq40hizR0RaGWMO24qdlgAPGmOWOzjOZGAyQERExICZM2ee8X6TJk3o2LFjpfFWt6NcXZKQkEBmZma1t8/OzqZhw9qtBKtLvP38wboGqcVBvLExn4Jiw4P9AukeXvn/RW6R4Z1N+WxLLebKaH+u7eyPj4eOMOyOv4NNx4t4c2M+w1v5cXevgFodnXnUqFEbyqvjdeUdRDJg/7UjEjjsYJ1UY0wOkCMiy4E+wB5jzGGwip1E5FusIquzEoTtzmIawMCBA01MTMwZ7+/cudOp/g2ePGFQqcDAQPr161ft7WNjYyl7/byJt58/wBuzfmLa1kKahgTy0R2D6Bzh/P/EJaNKeO777Xy5LhFp1Ix/Xd/H7UUk1VHbfwcHUnN48O2V9GjVmA/vPb9OXTNXVovHAZ1EJFpEAoAJwNwy63wPjBQRPxEJxiqC2ikiISLSCEBEQoDLgG0opVwi/lgWL8/fyVsb8+nYvCHf3n9+lZIDgL+vDy9f05OnxnTlxy1HuOWDtWSeKnRRxPXDqYIipny+AV8fYeotA+pUcgAX3kEYY4pE5AFgEVYz1+nGmO0iMsX2/lRjzE4RWQhsAUqwmsJuE5H2wLe22yw/4AtjzEJXxVqXNGzYkOzsbHeHobzA0cw85m4+xHe/HWbHkZP4CAxp6cv0Pw4lOKB6Hw0iwpQLOxAZFsSfvtrMLR+u5fO7htAkWOdKL8sYw1PfbGX3sSw+uWMwbZoGuzuks7i0H4QxZj4wv8yyqWVevwa8VmbZPqyiJqVUDTLG8P2mw8yMS2Tt/nSMgT6RTXhubHfG9mnJjg1rqp0c7I3t3Yogf1/u/Xwjt05fy2d3DaFJkCYJex/9eoC5mw/zxOVd6mzrL+1J7WJPPvkk7dq147777gPghRdeQERYvnw5J06coLCwkJdeeonx48e7OVLlDX7edZxHvtpEVHgwD13UifF9W53RO3dHBdtW1cXdIvjfLf2Z8vkGbv1Qk4S9tfvSeHn+Ti7rHsG9F3Zwdzjl8q4EseApOLrV4VtBxUXgW43L0aIXjDmrg/jvJkyYwCOPPPJ7gpg1axYLFy7k0UcfpXHjxqSmpjJ06FCuuuoqnVdauZQxhv/+nECbpkEs+dOF+NdCz9yLu0Uw9ZYBTPl8A5M+XMunmiTIzC3kgS9/o23TYF6/oY/LR2Q9FzqjnIv169eP48ePc/jwYTZv3kxYWBgtW7bkmWeeoXfv3lxyySUcOnSIY8eOuTtUVc/9mpDGpqQMplzYoVaSQ6mLu0Xwv4kD2HHkJJOmryMz17srrv+7NJ7U7HzeuqkfjQLrdrL0rjuICr7p57qwmet1113H7NmzOXr0KBMmTGDGjBmkpKSwYcMG/P39iYqKcjjMt1I16e1l8UQ0bsB1AyJr/diXdLeSxL0zNjBp+jo+u2swjev4h6Mr7EvJ5uNVB7hxYBt6tm7i7nAqpXcQtWDChAnMnDmT2bNnc91115GZmUnz5s3x9/dn2bJlHDx40N0hqnpu/YF01uxLZ/IFHWjg556mlJd0j+DdiQPYcTiTG99bQ2LaKbfE4U4vz99JoL8vj13Wxd2hOEUTRC3o0aMHWVlZtG7dmpYtWzJx4kTWr1/PwIEDmTFjBl27dnV3iKqee3tZAk1DArhpcPlDZtSGS7tH8MFtgzh04hTj3l7Jst3eM/zaivgUftp5nAcu6sh5jRpUvkEd4F1FTG60devpyvFmzZqxevVqh+tpHwhV07YdyiR2dwpPXN6lRpqwnqsLO5/HDw+OYMrnG7nz4zgevaQzD4zqWKcra89VUXEJf/txB22bBnPH8Ch3h+M0vYNQqp57++cEGgX6ceuwdu4O5XftwkOYc+/5XN23Nf9esofJn62v15XXX65LZM+xbJ65opvbiviqQxOEUvVY/LEsFm4/yu3nR9W5SuGgAF/+fUMf/npVD2J3pzD+7ZXsPup4el5PlnmqkH8v2cPQ9k25vEdE5RvUIV6RIFw1Ym1d4g3nqKru3di9BPn7csfwaHeH4pCIcNv5UcycPJScgmKufudXEo7XryTxn6XxZOQW8tzYHh7X16neJ4jAwEDS0tLq9QeoMYa0tDQCAwPdHUqVGWPYdiiT1xfv5sUfdvDhyv0s3HaUbYcyOZFTUK9/b652MC2H7zcd4pahbWkaEuDucCo0MKopcx8Yjp+P8MqC3e4Op8YkHM/m09UHmDCoDd1bed6Me+6vsXKxyMhIkpOTSUlJqXC9vLw8j/yALRUYGEhkZO23b68OYwxbkjOZv+0IC7YeJTH9FL4+QoCvD7mFxWesGxzgS/eWjXl3Yn+aN/bc3487TP1lL36+Ptwzsr27Q3FKyyZBTInpwGuLdrNufzqDo5u6O6Rz5mnNWsuq9wnC39+f6OjKb69jY2PPaS4FVbmTeYW883MCP245wqGMXPx8hOEdm3H/qA5c2r0FYcH+ZJwq5FBGLskncm0/T/HF2kT+8t02pt06wONu0d3lcEYuszckc+OgNh6VWO8cHs2nqw/w8vydfHvf+R79+14Rn8LPu47z9JiuNGvoGc1ay6r3CULVDXmFxdz98Xo2JJ7gws7n8cglnbise4uzhoEOCwkgLCTgjF6mLZsE8vL8Xfyw5QhX9WlV26F7pP/+HI8x8McL6u5AcI4EBfjy2KVd+PM3W1iw7ShX9Grp7pCqbc7GQzQNCeB2D2rWWla9r4NQ7ldUXMIDX/xG3MF03ryxL9NvH8T1A9s4PUfAXSPa06dNKC/M3U5adr6Lo/V8P+04xpfrkrhjeFSdnGOgMtcOiKRzREP+uXAXhcUl7g6nWowxrExIZXjHZh7VrLUsTRDKpYwxPDVnKz/tPMaLV/VgXDXuAHx9hNeu601WXiHPz93ugijrj+NZefz5my10b9mYxy/3zHJvXx/hqTFdOZB2ii/XJbo7nGrZfSyLlKx8RnZs5u5QzokmCOVS/1iwi9kbknnkkk7cOiyq2vvpHNGIhy7qxI9bjrBw29GaC7AeKSkxPP71FnLyi3jrpr4e/c11VJfmDG3flP/8FE92fpG7w6mylfGpAIzopAlCKYem/rKXacv3MWlYOx6+uNM5729KTAe6t2zMs99tI+NUQQ1EWL98tOoAy/ek8OzY7nRs7pqRiWuLiPD0mG6k5RQw7Ze97g6nylbEp9LhvBBahQa5O5RzoglCucSsuCReWbCLsb1b8sK4mukg5O/rw2vX9ybjVAEv/liTc595vh2HT/Lqgl1c0i2CW4a0dXc4NaJPm1DG9m7J+yv2c/yk5wyHn19UzNr9aYzsVDenEa0KlyYIERktIrtFJEFEnipnnRgR2SQi20Xkl6psq+qmJTuO8dScLYzs1Ix/39C3Rgdh69GqCffGdGDOxkMs2+U9I4FWJK+wmIdn/kaTYH9evbaXRzcNLeuJy7tQVFLCGz/FuzsUp204eIK8whJGeHj9A7gwQYiIL/AOMAboDtwkIt3LrBMKvAtcZYzpAVzv7LaqbsrMLeSJ2Zvp2boJU28ZQIBfzf+JPXBRRzo1b8jTc7ZyMq/+DvDmrJfn7yT+eDavX9+HcA9tb1+eduEhTBzSjq/iEj1mCI6V8an4+QhDO4S7O5Rz5so7iMFAgjFmnzGmAJgJjC+zzs3AHGNMIoAx5ngVtlV10LuxCWTmFvKPP/QipIFrutk08PPltev7cDwrj6v+u5J/zN/Jmn1pHtsk8lws3XmMT1cf5K4R0VzQ2fOLNBx56OJOhAT4ecxdxMqEVPq1DaWhi/7+a5MrE0RrIMnudbJtmb3OQJiIxIrIBhGZVIVtVR2TlH6Kj349wB/6RdKjlWunU+zbJpS3b+5PZFgw03/dz4Rpa+j/tyXc/8VG5mxMJj2n/ldiZ54q5M+zt9C1RSP+PNozm7Q6o2lIADcOasOibUdJyarb/WBO5BSw9VAmIzrWj2TtyhTnqCC07MhrfsAA4GIgCFgtImuc3NY6iMhkYDJAREQEsbGx1Qo2Ozu72tvWF+d6DaZuzsOUlDC8cVqtXMtg4O6OMDEqiO2pxWxOKWbFrqPM23IEX4GRkX6Ma+9PeJBz34M87W9g9p4C0nMKeaiPD6tXrqiRfdbVa9CeEopKDK99/QtXtHftwIPncg3WHS3CGAjJSiQ29lDNBuYGrkwQyYD9/IaRwGEH66QaY3KAHBFZDvRxclsAjDHTgGkAAwcONDExMdUKNjY2lupuW1+cyzXYkpzBmoW/cv+oDvzh8tqfQnWM7WdJiWHb4UxmrU/iq7gkfj1czA0D23DfqI60rqTJoSf9DRzPymPp0ljG9WnFbVfV3BhidfkafJe8mnVpebxy+4UunX3uXK7BojlbaBR4hNuvGoWfr+c3EnXlGcQBnUQkWkQCgAnA3DLrfA+MFBE/EQkGhgA7ndxW1RHGGP4+byfhIQFMudC9Y//4+Ai9I0N56epexD4xihsHtWHW+iRiXlvGX77dyqGMXLfGV1PeXbaXguISHr20s7tDqTU3DWnDgbRTrNmX5u5QHDLGsCI+lWHtw+tFcgAXJghjTBHwALAI60N/ljFmu4hMEZEptnV2AguBLcA64ANjzLbytnVVrOrc/LTzOGv3p/PIpZ1pVIdmLWsdGvR7orhh4OlE8cNmhzejHuNQRi5frE3k+gGRRDcLcXc4tWZMz5Y0CfLny7ikyld2g4Npp0g+kctID+89bc+l1ezGmPnA/DLLppZ5/RrwmjPbqrqnsLiEfyzYSfvzQpgwqE3lG7hB69Ag/n5NL+6N6cAtH6zlq7ikao0JVVe8ZWvN81AN9E73JIH+vlzTrzVfrE0kLTu/zjXpXZFQOrxG/aigBu1Jrc7RzLgk9qXk8PSYbvjX8dvqyLBgRnY6j98ST1Bc4pkz1e1LyWb2xmQmDm3r8cM4VMdNg9tSUFzCnI11rwJ4ZXwKrUODiAr3vBF0y1O3/6NVnZaVV8ibS/YwJLopl3Rr7u5wnNK/XSg5BcXsOeYZna7KeuOneAJ8fbgvpqO7Q3GLLi0aMaBdGF/GJdap6WiLiktYtTeNkZ2a1aue7JogVLVN/WUvaTkF/OXKbh7zTzGgrTWN5YaDJ9wcSdXtOHySHzYf5s4RUZzXqG4Vr9Smmwa3ZV9KDuv2p7s7lN9tOZRJVl6Rx4/eWpYmCFUtRzPz+GDFfsb3bUXvyFB3h+O0Nk2DaNYwgI2Jnpcg/r1kN40C/Zg80rNmiatpV/ZqSaNAP5fMFfHvJXvYmlL14cVXxqciAsM7aIJQig9X7qOoxPC4h03GLiL0bxvGRg+7g9iYeIKfdh5nyoUdnJ6Jr74KCrAqq+dvO8qJGuwxn3GqgLeWxvP93qqP77UyPpWerZoQFuLaTny1TROEqrLM3EK+WJvI2N4tPXJKy/7twjiQdsqjpi99ffFumjUM4Pbzo9wdSp0wYVBbCopKmPNbzVVWxx2wvjQkZJRwJNP5/jLZ+UVsTDxR74qXQBOEqoYv1iaSU1DM5AvauzuUaunfNgyAjYkZ7g3ESasSUvk1IY17Yzq6bABET9O9VWP6tgll5rqaq6xetz8NX1sP7arMWrh2XxpFJcbjpxd1RBOEqpL8omKm/7qfkZ2auXxAPlfpHdkEPx/xiHqIo5l5PPb1ZlqHBjGxnkwEVFNuHtyW+OPZNdbgYN3+dAa0CyOyobBgq/MJYkV8KoH+PgyICquROOoSTRCqSr7/7TApWfkee/cAVoerHq0a1/mWTNn5Rdz5cRwncwuZNmkAgf6eO8e0K4zt05KGDfz4ogYqq7Pzi9h2+CRDopsyqIUfcQfTnZ7FbmVCKoOjwz16DvDyaIJQTispMUxbsY/uLRt7/GxZ/duFsSU5o87OIVFUXML9Mzay+1gW70zs77F3a64UHODH+L6tmLflCNsOZZ7TvjYetDpPDo5uysAWfhgDi7ZXfhfxW+IJEo5ne0w/oKrSBKGc9vOu4yQcz+aPF7b3mH4P5enfNoy8whJ2Hjnp7lDOYozh/77fzi97Unjp6p7EdKmfHz41YfIF7QkLDuAP/1vF52sOVrs+Yt3+dHx9rBZurRv60LF5Q+Y7Ucz04cr9NAr049r+kdU6bl2nCUI5bdryfbQODeKKXi3dHco5G9DOVlFdB4uZpv6yjy/XJXJfTAduGqz1DhVpFx7CvIdGMLR9OM9+t42HZm4iO7/q/RjW7U+nZ+smvzcCuKJnC9buTyO1gpZuhzJyWbDtKDcNbltvGw9oglBO2Zh4gnUH0rlrRHSdH3PJGa1Cg2jROJANtdySaduhTF5ZsIvF24+Seers9vZzNx/m1YW7uKpPK4/rY+Iu4Q0b8PHtg3ji8i7M23KYcf9dyY7Dzt8Z5hUWsykpgyHRTX9fdkXvlpQYWLz9WLnbfbLqAAC31eOmx/Uz7akaN+2XfTQJ8ufGOjpia3UMaFe7HeaMMfzlu21sTsoAQAS6t2zM0PbhDG0fjp+v8PiszQyOaspr1/d26aQ49Y2Pj3D/qI4MbBfGg1/+xjXv/soLV/VgwqA2lRaHbk7KoKC4hMFRpxNEl4hGtG8WwvytR7jZQeux7PwivlybyJieLSqdiMqTef5XQeVy+1KyWbTjKLcObVevbqX7tQ3lUEYux5xsrXKu1uxLZ3NSBv83tjuz/jiMRy/pTONAfz5bc5B7Pl3PHR/FEdk0iGmTBtTLFjG1YUj7cOY/PJLB0U15es5W3v45odJtSsd0GmjXTFVEGNOrBav3pTmc3/zr9Ulk5Rdx90jPbc3njPrz365c5oOV+/H39al3t9L29RBjaqFe5b3lewkPCWDikLYE+vsyOLopD13cibzCYjYnZfBbUgbj+rQiNLh+DddQ25o1bMAndwzmnk/X8+Gv+7nngvYVNhFedyCdri0anXXdx/RsyTvL9rJkx1FuHHT6LqK4xDD91/0MaBdG3zahrjqNOkHvIFSFUrLymb0hmWv7R9a7EUR7tGpCgJ9PrfSH2HnkJLG7U7j9/KizPqwC/X0Z0j6cKRd2qNfFFbXJx0e4a0Q0GacKmb/1SLnrFRaXsOHgCQbb1T+U6tGqMW2bBp/VmmnJjmMkpedy94joGo+7rtEEoc5yMq+QjYknmBWXxJPfbKGwuIR7Rta/f4YAPx96t25SKz2q3/tlL8EBvtw6rJ3Lj6UswzqE075ZCDPWlt+Rbvvhk5wqKHaYIEqLmX5NSD2jQcGHK/cRGRbEZT1auCTuukSLmBQFRSW8vSyBpZtyeWrVUo7alckH+Plwz8j2tD+voRsjdJ3+7cL4+NcD5BcVu+wYSemn+GHLEW4/P0qLj2qRiHDzkLa8NG8nu46epGuLxmets25/GsAZFdT2rujZkvd+2ceSnce4bkAkW5IziDtwgv8b2/33cZvqM5feQYjIaBHZLSIJIvKUg/djRCRTRDbZHs/ZvXdARLbalq93ZZze7tvfknlraTxZBXB+h3D+PLoL708aSOzjMex8cTTPXNHN3SG6TP+2YRQUl7DtkOs6zH24cj8C3OUFRRJ1zbX9Iwnw8+GLcu4i1u1PJ7pZCM0bBzp8v3dkE1qHBrHAVkz14cr9NGzgxw0D62fHuLJcdgchIr7AO8ClQDIQJyJzjTE7yqy6whgztpzdjDLGpLoqRmU1vfx09UG6RDTiqb7FjBrV190h1ar+7UIBq6K6kwv2n55TwMy4RMb3be2Vc0i7W1hIAGN7tWTOxkM8ObrrGa3wSkoM6/anM6Zn+Q0URIQxPVvw6eqDxB/LYp7tTrBRoHfMyeHKO4jBQIIxZp8xpgCYCYx34fFUNWxKymD74ZPcMqydxw+fUR3NGwXSpmmQy+ohPll1gLzCEqZcWL+bQ9ZlE4e2JTu/iB82Hz5j+e5jWZzMK3JY/2BvTK+WFBSXMOXzDZQYU+9a81XElQmiNZBk9zrZtqysYSKyWUQWiEgPu+UGWCwiG0Rksgvj9GqfrTlIiG2GLm/Vv20YGw6eqLF5BUqdKiji09UHuKRbczpFNKrRfSvn9W8bRpeIRmdVVpf2f6gsQfRrE0qLxoHsTclhdM8WHjlJVnW5spLa0dfRsv+BG4F2xphsEbkC+A5+v9Mfbow5LCLNgSUisssYs/ysg1jJYzJAREQEsbGx1Qo2Ozu72tt6quwCw9xNp7igtR/rV6/0ymsA0Ci/kONZBSSlmRo9/yUHCzlxqpAhjU96zHWtr38Dg5oW8vnOAj76finRTaxmxvM25dE0UEjYvJa9dnfPjq5Br7Aijp6E/sEZ9fL6lMsY45IHMAxYZPf6aeDpSrY5ADRzsPwF4PHKjjlgwABTXcuWLav2tp7qvV8STLsnfzQ7j2QaY7zzGhhjzNbkDNPuyR/NP2YsqbF9FhQVm/P/sdRc++6vNbbP2lBf/wYycwtM12cXmCdnbzbGGFNSUmIGvrTEPPzlxrPWdXQNUrPyzA+bD7k6TLcA1ptyPlNdWcQUB3QSkWgRCQAmAHPtVxCRFmIr+BaRwVhFXmkiEiIijWzLQ4DLgG0ujNXrlJQYPl+TyOCopg6b/3mTri0aEeTvS0LGmU1dc/KLWBGfwltL49l9NKtK+5y35QiHMnL544UdajJUVU2NA/0Z37cV3286zMm8Qg6knSIlK5/B0eFObR/esAFje7dycZR1j8uKmIwxRSLyALAI8AWmG2O2i8gU2/tTgeuAe0WkCMgFJhhjjIhEAN/acocf8IUxZqGrYvVGy+NTSEw/xeOX64ihfr4+9GnThN0pGfy04xjrDqSzdn862w5lUlxilYrO3XyY+Q+NJMCv8u9UxSWGqb/spVPzhlzcVedyqCsmDmnHzLgkvvvtEA1sv8fK6h+8nUs7yhlj5gPzyyybavf8beBtB9vtA/q4MjZv9/maRJo1DGC0F/QGdcaAdmGs2ZfO3Z+uJ8DXh75tQrn3wg4Mjm7KybxCHvjiNz5YuY/7YjpWuq+Pft3PrqNZvHVTPx2RtQ7pFdmE3pFNmLEmkR6tGhMeEkCH80LcHVadpj2pvVDyiVP8vOsY98Z0cOobsTe47fwo0o8kcfUF/enTJvSs8ZJ+2HyYt5bGM653qwpbsSQcz+a1Rbu5pFsE43p7/sRK9c3EIW158putHEjL4aKuzb2yaXdV6KeDF/rSNsm7zlZ2WvNGgVwe5c+Q9uEOR/58flwPfER4fu72cpvDFpcYHv96M0EBvrz8h5764VMHjevTikYN/MgvKtHiJSdogvAyBUUlfBWXxEVdI4gM85723OeqVWgQj17SmZ93HWdRObOMvb9iH5uSMvjrVT1o3sjx0A3KvYID/PhDf6vPz6Byxl9Sp2mC8DILtx8lNbuAW4bq3UNV3T48iq4tGvHXH7aTU2be4/hjWfx78R7G9GzBVX28r7WLJ3no4k68dHVPerTy7tZ7ztAE4WU+X32Qtk2DuaDTee4OxeP4+/rw92t6ciQzj/8sjf99eVFxCY99vZmGgX787WotWqrrwhs24Jah3jm0TFVpgvAiu46eZN2BdG4Z2lZb11TTgHZNmTCoDR+u3M/OI9YIsFN/2cuW5Ez+Nr4nzRrWr0mVlHerNEGIyFgR0URSD3z86wEC/Hy4fkAbd4fi0Z4c3ZUmQf48+902th/O5D9L4xnbuyVXaqslVc8488E/AYgXkX+KSP2dGKCe25qcyVfrk7h5cFvCQnTSmnMRFhLA02O6suHgCW5+fy1Ngvx5cXxPd4elVI2rNEEYY24B+gF7gY9EZLWITC4dCkPVfcUlhme/20p4SAP+dFlnd4dTL1w3IJLB0U3JzC3kpat70VSTrqqHnCo6MsacBL7BmtOhJXANsFFEHnRhbKqGzIxLZHNyJv83thuNvWSiE1cTEd65uT8fTBrI6J7aG13VT87UQYwTkW+BnwF/YLAxZgzWUBiPuzg+dY5Ss/P558LdDGsfrs0va9h5jRpwSfcId4ehlMs4M9TG9cAbpsxcDMaYUyJyp2vCUjXllQW7OFVQxN+u7qHN+pRSVeJMEdPzwLrSFyISJCJRAMaYpS6KS9WAuAPpzN6QzN0j29OxuVYZKaWqxpkE8TVQYve62LZM1WGFxSU8++02WocG8eBFlY9AqpRSZTmTIPyMMQWlL2zPtclGHffJqgPsPpbF8+O6Exygg/YqparOmQSRIiJXlb4QkfFAqutCUufqSGYubyzZw0Vdm3OpVqIqparJma+WU4AZIvI2IEASMMmlUalz8tKPOykqMbwwTiumlVLVV2mCMMbsBYaKSENAjDFVm5xX1apZ65OYt/UIf7q0M23DdThvpVT1OVU4LSJXAj2AwNJvpMaYF10Yl6qGVQmpPDNnKyM7NePemA7uDkcp5eGc6Sg3FbgReBCriOl6oJ0zOxeR0SKyW0QSROQpB+/HiEimiGyyPZ5zdlt1poTjWfzx8w20Py+Edyb2x99Xx1dUSp0bZz5FzjfGTAJOGGP+CgwDKh0OVER8gXeAMUB34CYR6e5g1RXGmL62x4tV3FZh9Za+4+M4Gvj5Mv32QTqchlKqRjiTIPJsP0+JSCugEIh2YrvBQIIxZp+taexMYLyTcZ3Ltl4lr7CYez5dT0pWPh/cNlCnEVVK1RhnEsQPIhIKvAZsBA4AXzqxXWusFk+lkm3LyhomIptFZIGI9Kjitl6tpMTw2Neb2ZSUwRs39KVvm1B3h6SUqkcqrKS2TRS01BiTAXwjIj8CgcaYTCf27ah9pSnzeiPQzhiTLSJXAN8BnZzctjTGycBkgIiICGJjY50I7WzZ2dnV3tZdZu8pYN6+Qm7o4k9Q2m5iY3ef0/488RrUJG8/f9BrAHoN7FWYIIwxJSLyOla9A8aYfCDfyX0nc2ZdRSRwuMz+T9o9ny8i74pIM2e2tdtuGjANYODAgSYmJsbJ8M4UGxtLdbd1h1nrk/hx3xZuGtyGl6/pVSP9HTztGtQ0bz9/0GsAeg3sOVPEtFhErpWqfwLFAZ1EJFpEArBmpptrv4KItCjdr4gMtsWT5sy23mz13jSembOVER2b8eL4ntoZTinlEs70g/gTEAIUiUgeVvGPMcY0rmgjY0yRiDwALAJ8genGmO0iMsX2/lTgOuBeESkCcoEJxhhjO9ZZ21bvFOuX/ak53DtjA+3Cg7U5q1LKpZzpSV3tcaKNMfOB+WWWTbV7/jbwtrPbervMU4Xc9XEcAky/fRBNgrQ5q1LKdSpNECJygaPlZScQUq5VWFzCvTM2kHTiFDPuHkq78BB3h6SUquecKWJ6wu55IFYfhQ3ARS6JSJ3FGMNz329n1d40/nV9HwZHN3V3SEopL+BMEdM4+9ci0gb4p8siUmf5cOV+vlyXyH0xHbhuQKS7w1FKeYnq1HAmAz1rOpD6Lq+wmP8ujedAak6VtvtpxzH+Pn8nY3q24PHLurgoOqWUOpszdRD/5XQnNR+gL7DZhTHVO3mFxfzxsw38sieFFfGpfPXHoU41TU0+cYqHZ/5Gz1ZN+PcNffHx0easSqna40wdxHq750XAl8aYX10UT72TV1jM5M82sCI+hUu6RfDTzmMs2XGMy3q0qHTbf8zfRbEx/O+W/gQF+NZCtEopdZozCWI2kGeMKQZrpFURCTbGnHJtaJ6vdCC9lQmpvPqH3vyhf2suf3M5ryzYxaiuzSvsw7Bqb+rvE//oAHxKKXdwpg5iKRBk9zoI+Mk14dQf9snhn9f25oZBbfDz9eGZK7qxLzWHL9cllrttUXEJL/6wg8iwICZf0L4Wo1ZKqdOcSRCBxpjs0he25/qVtgK5BcXc/YmVHF67rg/XDzw9rNRFXZsztH1T3vwpnpN5hQ63/2JdIruOZvHsld0I9NeiJaWUeziTIHJEpH/pCxEZgDUshnIgt6CYuz6J49e9qfzruj5nNUsVEf5yRXfScwqYGrv3rO1P5BTw+uI9nN8hnMudqKdQSnm5ogJI2eOSXTtTB/EI8LWIlI6m2hJrClJVRk5+EXd9Esfa/em8fn0f/tDfcZ+FXpFNuKZfaz5cuZ+JQ9vROvR0Cd7rS3aTnV/E8+N66CB8SinHMpIgYQnE/wT7f4GAhvDYLqjhzwxnOsrFiUhXoAvWQH27jDGOy0a82Mm8Qu74KI5NSRm8eWNfxveteH6jxy7rzLytR3h90W7+fWNfAHYcPskXaxOZNCyKLi2qPQSWUqq+KS6Eg6sgfjEk/AQpu6zlTdpC7xug46VgSkBqtkjamX4Q9wMzjDHbbK/DROQmY8y7NRqJB8s4VcCk6evYcfgkb9/UjzG9Wla6TWRYMHcOj2bqL3u5c0Q0PVo15oUfttMkyJ9HL+lcC1Er5YTCPNj0OWz6AnwDIDgcgpvaftoeLXpDC+07W+MK82DfMtj5A+yeD7knrN9Bu/Oh363Q6VJo1rnG7xrsOVPEdI8x5p3SF8aYEyJyD6AJAkjLzufWD9eRcDybqbcM4JLuEU5ve9+oDnwVl8jf5+3k5iFtWbc/nb9f05MmwV44SmtuBmAgKMzdkSiAghxY/xGs+i9kH7WSgH8wpO+D5PVwKg1K7AoS2p4PQ6dAlyvB15mPFeVQwSnYs8BKCvFLoCAbAptA5zHQbSy0HwUNGtZaOM78Jn1ERGzzNCAivkCAa8PyDMez8pj4/loS00/x/m0DubDzeVXavnGgPw9f3IkXftjB1kOZdG/ZmAmD2roo2jrGGDi2HeIXwZ7FkLzOukVu1BKad4fm3ayfEd2hWRcI0IZzNeLIZljyHDRofPr6Nu8OYdHg64dvUQ4s/xeseddKAtEXwLXvQ9TIM7+pGgP5WZCTArsXwLr3YNYkaNIGBt0N/SdZdxrKOQWnYP10+PVN65qGNIde10O3cda193PPR64zCWIRMEtEpmINuTEFWODSqDzAkcxcJr6/liOZeXx0xyDO79CsWvu5eUg7Pll9kP2pObxwVQ986/NwGvnZcGDl6aRwMtla3rIPjHzc+mZ0fCcc3wFxH0BRnvW+XxBc9BcYeh/4aLPfatv2DXx3v3WdGzS2vqWWjqLj2wDO68ywlH1QnAOdLrN+J22HON6XCAQ2th7nPwBD77USxdqp8NPzEPsK9L0JLn5O7worUnAKNnwEK9+EnOPQPgZG/AmiRtSJv3VnEsSTwGTgXqxK6t+wWjJ5reISw8QP1nI8K5/P7hrMwKjqf1MK8PPhf7f0Z8fhk/VvGO/iIjiyCfYus8pSk9ZZxRL+IdBhFFz4Z+uDqLGDP6eSYkjfbyWLzV/C4mdh549w9bsQ3qHWT8WjlRTDzy/Byn9DmyFww2fQKML6cErdfTopH9tBWkkoEVf/DVr1q9oxfHytIpBuY+HoVlj7Hmz8zPpCcPMsaBrtmnPzVIW5VhHer29C9jGIvhBiPrHqF+oQZ1oxlYjIGqA9VvPWpsA3rg6sLtt9NIt9KTn889re55QcSnVt0ZiuLSqcwdWz7JoPm2bA/hWQn2kta9Ebht1nfUNqNxz8GlS8Dx9faNbRenQbB1tmwYIn4H/D4dK/wqB7wEenW61UXiZ8c49119Z/Elzxr9PXPiDYSgR2yWBnbCwRVU0OZbXoBePfhj4TYOZE+OBimPBl+Xcj3uT4Tvjtc9jylVWUFH0BXPcRRA13d2QOlZsgRKQzMAG4CUgDvgIwxoyqndDqrvUH0wEY1iHczZHUMUUFsPgvsG4aNG4N3a+y7hSiL4SQ6hXBAVZxRp8bIXokzH0IFvzZKh4Z/zaERdVY+B4neYNV7h8QAm0G2x5DILyTlTxTE+DLCXBiv5UYBt3t0hYvZ4kaAXcvhS+uh0/GWXd/va6rvePXFbkZVvHephlwaAP4+EHn0VaxXNQId0dXoYruIHYBK4BxxpgEABF5tCo7F5HRwH8AX+ADY8wr5aw3CFgD3GiMmW1bdgDIAoqBImPMwKoc25XiDpygReNAIsOCKl/ZW2Qkwde3w6H1MOwBuOQF8K3h1liNW8HEr+G3z2DhM9bdxFVvQc9ra/Y4niBlN8y4zuogFRYFu360rgtYrV4iB0FSnNWi6NbvrOTqDs06Wkli5kT45i4rWY18vHYTlbucOAg//836MlOUB817wOUvQ68boGHVGrS4S0UJ4lqsO4hlIrIQmIlVB+EUW2und4BLsSYZihORucaYHQ7WexWrMrysUcaYVGePWRuMMcTtT2dgVJj2dC6VsBS+udvqzHP9J9DjatcdS8QqKmk/CubcYx3Xxw+6j3fdMeuazGT47BrrvCd9Z9XJGANpCZC01qrrSVpn9U24+n8Q1s698QY3teKc+6BVF5K+H8a+6baWObWiqAC+usVqFtzvFug70SrK87DPjHIThDHmW+BbEQkBrgYeBSJE5H/At8aYxZXsezCQYIzZByAiM4HxwI4y6z2IVacxqFpnUMsOZeRy9GRe/atQro6SElj+GsT+w2qWesNn1jfG2hDaBibOhs//ALPvgptCoNMltXNsdzqVbiWH/Cy4fd7pCnsRaNbJevS7xb0xOuLXAK55D5q2t/5eUnZbRYTNu7k7MtdY8S84ugVunGFV3HuoSmv5jDE5xpgZxpixQCSwCXjKiX23BpLsXifblv1ORFoD1wBTHR0aWCwiG0RkshPHqxXrD5wAYGA7L08QeZlW2XLsy9D7Rrj7p9pLDqUaNLRayDTvCl9NhAP1fB6r/GyrWOnEQbjpS2jZ290RVY0IxDwF139sfbOeOhKWvQxF+e6OrGYd2mj1Jek9waOTA4DY+r/V/I5FrgcuN8bcbXt9KzDYGPOg3TpfA68bY9aIyMfAj3Z1EK2MMYdFpDmwBHjQGLPcwXEmYzXDJSIiYsDMmTOrFW92djYNG1beQ/Hj7fmsPVLEOxcH4+Nht4uVcfYa+Bbl0HvLX2mUlUB8p8kcaXm5W2+d/Qsy6bvpGRrkp7G5z9/IatzJ4XqNM3fT8sgiUs4bTnr4gLPed/b83UFKCum19SXCTmxhW8+nSGvmmhZBtXUN/Asy6bB3Oi2OxZITHMmezveTGdrd5cd1xrlcA5/iAgZseBS/olziBr1FkX/d/HuyN2rUqA3l1vEaY1zyAIYBi+xePw08XWad/cAB2yMbOA5c7WBfLwCPV3bMAQMGmOpatmyZU+td+u9YM+nDtdU+Tl3m1DXIzTTm/YuN+WtTY3b84PKYnJZ5yJg3ehnzSjtjjm4/vby4yJjt3xvzwaXGPN/Yenx0pcNdOPs3UOuKi4yZdbsV+8bPXHqoWr8G8UuM+XdP69x+eNSY3IzaPb4D53QNFv3FOpf4JTUWj6sB6005n6mubEgeB3QSkWgRCcCq8J5bJjlFG2OijDFRWFOb3meM+U5EQkSkEYCtDuQyYJsLY3VKxqkC9hzLZlCUl/YMzTtplfkf/s2qjK5Lt8+NW8Ftc8EvED4dD0e2wNpp8N8BMOtWyDoKo1+1+k8krrGKazyBMTD/Cdg+By75a92sXzgXHS+B+1bD0PutHsXvDIW9P7s7quo5uBpWvQ0D77TOqx5wWYIwxhQBD2C1TtoJzDLGbBeRKSIypZLNI4CVIrIZWAfMM8YsdFWsztpw0Kp/GFQDneM8Tt5J+PxaW3L4uG4lh1JhUTDpezDF8N5Iq2NdSDMrmT30mzWYXLdxVm/uAyvcHW3ljLGGrVj/IQx/GEY84u6IXKNBQxj9slWP1aCRVQm/8BlrNFNPkZ8N302B0LZw6d/cHU2Ncemwi8aY+cD8MsscVUhjjLnd7vk+oI8rY6uOuAMn8PcV+rQJdXcotev35LDR6vXZbZy7IyrfeV2sJLF+ulVJWLb3btuh1qikCUuhyxj3xOisFf+CX/8DA++y7h7qu9YD4I+/WIMJrnnHmgjn2g88o6XTT89bjQdun1ero626mo7LWwXrD6TTq3WT+jdPdPp+mPcYvU5kQPZcaBJpfRNq0sbq0PPtFKsH6PUfWb2j67oWvWDsG47f82tgjY65d2ntxlRVa/5n9RnoPcHqBV3PGkSUyz8IrnjNKqL5/n5470K47G8weHLtXIPYVwlLDwBinN9m78/W4JLDHqizQ2ZUlyYIJ+UVFrMlOZM7hke5O5SaVXAKvroVMg4S4N/MKuvOPXHmOuJrSw71pDNax4utsYnS91nt8uuajZ/Bwqeg61gY/453jjnV+XK4d5WVJBb82ZobYfw71iCDrpIaD7Ev06VBOBTeC/6BlW9TkAPfP2gNSX/Rs66LzU00QThpS3ImBcUlNTI4X51hDPz4CBzbBhNns+GQHzExMVYnrMxka/iMzESI6FW/BlrrcLH1M2EpDK7lBFGQY92xNW3veI6Lbd9YPY47XAzXTffuyXcaNrf6ucR9YI3mO3WEVeTU/kLXHG/TDAAC89Osep9h91e+zaq3rWHr71ho3f3UM17811c1cQesAfoGtqtHLZjWvW+NKjnqWasX8qFYa3mDRrYJezyg7Lc6wjtYRWh7f4bB99TecXfNs1oknTwEiDUEtv3kSEV5VnJoOwxu/LzyEW+9gYj1O4oaAbNus1qoxTwFFzxRs/MllBTD5pnQ6XLSU4/RdMXr1pAuDSqYG/7kEWu47u7jod2wmoulDtEE4aT1B9Lp1LwhYSH1ZPyYg6th0dPWVIYjH3N3NLVLxPqGvvVra8wcV48JlJkM8/8Mu+dZA7aN+ou17PgO67F7gdXyCqzxem7+SmfQK6t5N5i8DOY9Zg3VcXCVdTfRsHnN7H/vz5B1BMa8yv6E4zTd+DisfsdKRuVZ9hKUFFkDU9ZTmiCcUFJiWH/wBGN7t3J3KDUj6yh8fRuEtoM/vOedZdwdL7ba3Sevc92Qy8VF1lScP//dmk710hetWfHKjnJbmAepeyAj0ZofILAezQ1SkwJCrMEH2w2H+Y+fLnKKvuDc971pBgQ1hc5jyDq+CrpdZRUfDboHQhwM639kC/w2w5pNry7WY9UQL/xkqLo9x7PIyiuqHx3kigqsW/X8LKsYI7CJuyNyj+gLrMr3BBe1Zjq0Ed6PgUXPWC1b7l9r9WVwNAS6f6A1rlK3sZocKiMC/W+Fe362/nY/HQ+//NMqIqquU+lW8V/vG07fTV70LBTmWLPwlWWMNe9JUJg1dHk9pgnCCXH7rfqHetFBbvGzkLTGGkkzom6MfeMWgU2sCXZc0dw1fR9MHw3ZKVYnvZtnuX/I7fomogfcswx6XQ/L/m5NSJSRWL19bfsGigug782nl53XBfrcbNXTZSafuf6ehbB/OcQ8DUGh1T4FT6AJwgn1ZoKg9R9ZRR7DHvDOSXbK6nAxHNlsfZDXpEV/seZqmLzMmhvDW/ow1LYGDa0hxK95zyry+d8I2Dq76vv57XOrpV7LMn1zY54CDMTazXNWXAiL/8+atW/gHecUvifQBOGE9Qc8fIKgghz4/gGrSWv7Ud7RK9cZHS+yfu5bVnP7TFgKu+fDBY9b40Mp1xKx5r6essL61v/NXTBnsjUcvTOObYcjm6DfxLPfC21jTdO6aYbVRwKsL1lp8VbnvZqeMbEO0gRRiUMZuRzOzPPc4qVj22HaKOtb0sjHrUl2vLltvb2Wfa2KyZqqhyguhIVPQ1i0c23oVc1pGg13LLCKfbbOtu4mDq6ufLtNX4CPvzUNqCMjH7OGZvn5JasDaew/rPqrzqNrNv46ShNEJdaX9n/wtApqYyDuQ3j/IsjLsKZ8vPj/NDnY8/GFDqOsJo4lJee+v7gPIHW3Ne+w9mGofb5+VrHQnQutlnkfX2GNZVWe4kKrH1CX0Y5bKoE12OOw+2HHd9adSe4JuOzvXlNsqAmiEnEH0mnYwI+uLTyodUluBsyaBPP+ZDUJnLIS2se4O6q6qcPFkHPc6k1+LnJSYdk/oMNFdX8QwPquzWDrb777eGvgv6V/s74wlRW/GHJSrPmiKzLsAetOM36xta6nzeR3DvTrZCXi9p+gf7swfH085BtDarw1Z8PJw1a7+2EPemc/B2d1sNVD7F0K9Kv+fn5+CQqy4fJ/eM23yzqtQSO49kPr54p/QWEuXF7mm/+mLyCkOXS8tOJ9BTaGS56HX16rl+MtVUQTRAUyTxWy+1gW4/q0dHcozjm2w2oXjrHGhmkzyN0R1X2NW1q9mxOWQlQ1E8SRLbDhYxgyxZofW9UNPr4w9j/gF2QNH16UC1e8bn1hyk6xmqsOvde5YtcBt0O/SV73ZUsTRAU2JJbWP3hABfWRzfDp1eAbALf9AOd1dndEnqPjRbBmKr6RuVXf1hhY8CQEN614WAblHj4+MOZVa+iSlW9YdxJXvQ1bZ1nDZFRWvFR2X15GE0QFNiVm4Osj9IkMdXcoFTu0wZqFK6CRNe1meAd3R+RZOlwMq/5LaMY2wK7+IOuoVQyx7Rur12ynS63iiObdThdVbJ8Diatg7Jv1vtOUxxKBi58H/xBr/KTCXGtok1b96++AlDVEE0QFdh/Lol14MEEBdXiCoMQ18Pl1ViuMSXO1x251tB0GfkGEnfjNGj8pYQls/BT2LLIG0Wsz1BqOYclz1qNxpDWWU8eLYfFz1gRF/Se5+yxURUTgwiesIbkX/8VaduXr7o3JA2iCqED88Ww6N69guF93278CvrjRKkefNBeatHZ3RJ7JPxCiRhBxYDm80QOyj1qVl+c/CP1uhWYdrfUyD0HCT1YC2TYHNn5iLb/2/Zodelq5zvkPWIP+bZoBPa9zdzR1nksThIiMBv4D+AIfGGNeKWe9QcAa4EZjzOyqbOsq+UXFHEw7xdheNVRBXZhrVSK36FkzbeT3LIZZt0JYlJUcXDnTljfocTV+CUuh1XDrbqDTZWf3lG3SGgbcZj2KCiBprdVyqd357olZVc/AO7ximIya4LIEISK+wDvApUAyECcic40xOxys9yqwqKrbutK+lByKSwwdI6p5B5F3EpLWwcFfrbHrD2+0BgQbMsWqNKuukhJY/prVo7NFL7j1W6szjzo3fSeyIr05F1x8mXPr+wVA9EjXxqSUm7nyDmIwkGCM2QcgIjOB8UDZD/kHgW+AQdXY1mX2HMsCoHNEw4pXLCmGjINW/4PUPdbjyBY4usWaA8DHz5oEZsgUa5TPde/DwLuq18roVDp8+0erw07vCTD2DZ1YpqaIUOJbTyaDUqqGuDJBtAaS7F4nA2dMbCwirYFrgIs4M0FUuq2rJRzPxtdHiG4WcvabxlgTqR9YCWkJ1p1BqeBmVsuIC56wih4iB1llnmD1tt2/3Bpye+KsqgV0ZDN8dYs1zeGVr1tJRjtkKaVcyJUJwtGnV9n+7m8CTxpjisuMlOrMttaKIpOByQARERHExsZWOVCA7OzsM7ZdtT2P84Jg9coVZ63b6GQ8AzZOI6NJd062upJTwZGcCm7NqeDWFPnbDcmRCCTGnbFtm9bX0CH+EzbPeZMTTfs6FVuLIz/Rec9UCgIas73v38nK6Qi//FKNs6xY2Wvgbbz9/EGvAeg1sOfKBJEMtLF7HQkcLrPOQGCmLTk0A64QkSIntwXAGDMNmAYwcOBAExMTU61gY2Njsd/2xfWx9I1qREzMgLNXXrwUfPwI/eM8QoOr2ImuaBi88wt9jn4F4x+ouBdncRHMfwx2fwzRFxJ43XQGuLC+oew18Dbefv6g1wD0GthzZdfAOKCTiESLSAAwAZhrv4IxJtoYE2WMiQJmA/cZY75zZltXyi8q5kBajuP6B2Ng+3fWGD5VTQ5gtWC69EVrsvrSZpKOlBRb9Q0bPoYRj2pltFKq1rksQRhjioAHsFon7QRmGWO2i8gUEZlSnW1dFWtZ+1JyKDHQyVELpuT1kJkIPf5Q/QN0u8oaZXXZy44nNikpgbkPwrbZcMkL1kPb2SulaplL+0EYY+YD88ssm1rOurdXtm1tKW3B1MnRHcT2OdZ4R12vqP4BRKw5A6bFwPJ/WbNTlTIG5j1qdeSJeca6e1BKKTfwvtGnnBB/rJwWTCUlVvFSx0utSe/PRau+1iTpa6dC+n5rWWnrqA0fw4g/wYV/PrdjKKXUOdAE4UD88SyiwoNp4FemWCdpLWQdhh7X1MyBLvo/a7rDJc9ZyWHxs7BumjVBycXPaTNWpZRbaYJwIP5YNp0cjcG0fQ74BVpTFNaExi1hxCOwcy58fRusfhsG3QOXvaTJQSnldpogysgrLKcFU0kx7PjeGqOnQQ0O4DfsAWjc2tp3/0kw5p+aHJRSdYKO5lpGuS2YDv4K2ceg5zm0XnIkIBium26N1zT8Ea+clEQpVTdpgigj/njpGExlEsS2OdaEI50ur/mDth1qPZRSqg7Rr6tllLZgimpmNwhecZFVT9BltA6Op5TyGpogythzzEELpgPL4VRazbVeUkopD6AJooz449mOi5cCGln9H5RSyktogrCTV1jMwbQcOjW3a8FUVAA7f7B6TvsHui84pZSqZZog7DhswbQvFvIyzm3sJaWU8kCaIOw4bMG0fY41rEaHi9wUlVJKuYcmCDt7jmWdOQZTUT7smgddx1lzECullBfRBGEn/lg2UeHBBPjZLkvCUsg/qa2XlFJeSROEnbNaMCUssVovtb/QfUEppZSbaIKwKSg2Vgsm+wRxcJXVw9nX332BKaWUm2iCsDmaU2K1YCpt4pqTCim7oN357g1MKaXcRBOEzaFsA9i1YEpcbf1sN9xNESmllHtpgrA5lF2Cn30LpoOrrLkfWvVzb2BKKeUmLk0QIjJaRHaLSIKIPOXg/fEiskVENonIehEZYffeARHZWvqeK+MEK0FENQs53YLp4K8QOUibtyqlvJbLEoSI+ALvAGOA7sBNItK9zGpLgT7GmL7AncAHZd4fZYzpa4wZ6Ko4Sx3OLjld/5CXCUe3avGSUsqrufIOYjCQYIzZZ4wpAGYC4+1XMMZkG2OM7WUIYHCDvMJijp8yp1swJa0DU6IV1Eopr+bKBNEaSLJ7nWxbdgYRuUZEdgHzsO4iShlgsYhsEJHJLoyTvSnZGDg9zejBX8HHzypiUkopL+XKGeUcTax81h2CMeZb4FsRuQD4G3CJ7a3hxpjDItIcWCIiu4wxy886iJU8JgNEREQQGxtb5UBXHS4CIOPgLmLT99BvywJo2IHfVq2r8r48WXZ2drWuX33h7ecPeg1Ar4E9VyaIZKCN3etI4HB5KxtjlotIBxFpZoxJNcYcti0/LiLfYhVZnZUgjDHTgGkAAwcONDExMVUONG7RLnxlLzeMiSHA5MPyvTDsPqqzL08WGxvrdedsz9vPH/QagF4De64sYooDOolItIgEABOAufYriEhHERHb8/5AAJAmIiEi0si2PAS4DNjmqkD3HMsmIlisFkzJ66GkUCuolVJez2V3EMaYIhF5AFgE+ALTjTHbRWSK7f2pwLXAJBEpBHKBG40xRkQisIqdSmP8whiz0FWxxh/LolXD0uatqwCBNkNcdTillPIIrixiwhgzH5hfZtlUu+evAq862G4f0MeVsZUqLC4ht7CY1qF2/R9a9ISg0No4vFJK1Vle35Pa39eHtc9cwviO/tb0oknrtHhJKaXQBPE7HxE4shmKcrX/g1JKoQniTAd/tX621QShlFKaIOwdXAXNOkPD89wdiVJKuZ0miFKmGBLXQNth7o5EKaXqBE0QNg2zD0J+plZQK6WUjSYImyaZ260nWkGtlFKAJojfhWZshyZtIbRN5SsrpZQX0AQBYAxNMnfo3YNSStnRBAGQGk9AYaYmCKWUsqMJAk73f9AKaqWU+p0mCICDqyjwD4XwDu6ORCml6gxNEAAHV5ER2gPE0RxHSinlnVw6mqtHKMyDDjGk5jWnubtjUUqpOkTvIPwDYfw7HI+40N2RKKVUnaIJQimllEOaIJRSSjmkCUIppZRDmiCUUko55NIEISKjRWS3iCSIyFMO3h8vIltEZJOIrBeREc5uq5RSyrVcliBExBd4BxgDdAduEpHuZVZbCvQxxvQF7gQ+qMK2SimlXMiVdxCDgQRjzD5jTAEwExhvv4IxJtsYY2wvQwDj7LZKKaVcy5UJojWQZPc62bbsDCJyjYjsAuZh3UU4va1SSinXcWVPakfjVpizFhjzLfCtiFwA/A24xNltAURkMjDZ9jJbRHZXL1yaAanV3La+8PZr4O3nD3oNwPuuQbvy3nBlgkgG7GffiQQOl7eyMWa5iHQQkWZV2dYYMw2Ydq7Bish6Y8zAc92PJ/P2a+Dt5w96DUCvgT1XFjHFAZ1EJFpEAoAJwFz7FUSko4g1Qp6I9AcCgDRntlVKKeVaLruDMMYUicgDwCLAF5hujNkuIlNs708FrgUmiUghkAvcaKu0dritq2JVSil1NjndiMi7ichkW3GV1/L2a+Dt5w96DUCvgT1NEEoppRzSoTaUUko5pAlCKaWUQ16fILxxzCcRmS4ix0Vkm92ypiKyRETibT/D3Bmjq4lIGxFZJiI7RWS7iDxsW+4V10FEAkVknYhstp3/X23LveL87YmIr4j8JiI/2l573TUoj1cnCC8e8+ljYHSZZU8BS40xnbDGyKrvybIIeMwY0w0YCtxv+917y3XIBy4yxvQB+gKjRWQo3nP+9h4Gdtq99sZr4JBXJwi8dMwnY8xyIL3M4vHAJ7bnnwBX12ZMtc0Yc8QYs9H2PAvrA6I1XnIdjCXb9tLf9jB4yfmXEpFI4EpsA4XaeNU1qIi3Jwgd8+m0CGPMEbA+PIHmbo6n1ohIFNAPWIsXXQdb0com4DiwxBjjVedv8ybwZ6DEbpm3XYNyeXuCcHrMJ1U/iUhD4BvgEWPMSXfHU5uMMcW2ofYjgcEi0tPNIdUqERkLHDfGbHB3LHWVtyeIKo0XVc8dE5GWALafx90cj8uJiD9WcphhjJljW+x118EYkwHEYtVLedP5DweuEpEDWMXLF4nI53jXNaiQtycIHfPptLnAbbbntwHfuzEWl7ONAfYhsNMY82+7t7ziOojIeSISansehDWK8i685PwBjDFPG2MijTFRWP/7PxtjbsGLrkFlvL4ntYhcgVUOWTrm09/dG5HriciXQAzWsMbHgOeB74BZQFsgEbjeGFO2IrvesE1vuwLYyuny52ew6iHq/XUQkd5YFbC+WF8UZxljXhSRcLzg/MsSkRjgcWPMWG+9Bo54fYJQSinlmLcXMSmllCqHJgillFIOaYJQSinlkCYIpZRSDmmCUEop5ZAmCKUqISLFIrLJ7lFjg7eJSJT9qLpK1SUum5NaqXok1zYkhVJeRe8glKomETkgIq/a5lVYJyIdbcvbichSEdli+9nWtjxCRL61zcGwWUTOt+3KV0Tet83LsNjWsxkReUhEdtj2M9NNp6m8mCYIpSoXVKaI6Ua7904aYwYDb2P1yMf2/FNjTG9gBvCWbflbwC+2ORj6A9ttyzsB7xhjegAZwLW25U8B/Wz7meKaU1OqfNqTWqlKiEi2Maahg+UHsCbd2Wcb+O+oMSZcRFKBlsaYQtvyI8aYZiKSAkQaY/Lt9hGFNdR2J9vrJwF/Y8xLIrIQyMYaBuU7u/kblKoVegeh1Lkx5Twvbx1H8u2eF3O6bvBKrBkPBwAbRETrDFWt0gSh1Lm50e7natvzVVijgwJMBFbani8F7oXfJ+tpXN5ORcQHaGOMWYY1oU0ocNZdjFKupN9IlKpckG3mtVILjTGlTV0biMharC9bN9mWPQRMF5EngBTgDtvyh4FpInIX1p3CvcCRco7pC3wuIk2wJrZ6wzZvg1K1RusglKomWx3EQGNMqrtjUcoVtIhJKaWUQ3oHoZRSyiG9g1BKKeWQJgillFIOaYJQSinlkCYIpZRSDmmCUEop5dD/A6+Y41ePXOURAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The test metrics are: {'accuracy': 0.4832}\n"
     ]
    }
   ],
   "source": [
    "coeff = 1.0\n",
    "mean = 0.0\n",
    "std = 0.01\n",
    "params = {\"coeff\":coeff, \"mean\": mean, \"std\":None}\n",
    "\n",
    "reg_rate_l2 = 0.025\n",
    "\n",
    "in_dim = x_train.shape[1]\n",
    "out_dim = np.unique(y_train).size\n",
    "mid_dim = 50\n",
    "\n",
    "seed = 200\n",
    "\n",
    "dense_1 = \\\n",
    "    Dense(in_dim=in_dim, out_dim=mid_dim, \n",
    "          kernel_initializer=XavierInitializer(seed=seed, **params), \n",
    "          bias_initializer=XavierInitializer(seed=seed+1, **params), \n",
    "          kernel_regularizer=L2Regularizer(reg_rate=reg_rate_l2), \n",
    "          activation=ReLUActivation()\n",
    "         )\n",
    "\n",
    "dense_2 = \\\n",
    "    Dense(in_dim=mid_dim, out_dim=out_dim,\n",
    "          kernel_initializer=XavierInitializer(seed=seed+2, **params), \n",
    "          bias_initializer=XavierInitializer(seed=seed+3, **params), \n",
    "          kernel_regularizer=L2Regularizer(reg_rate=reg_rate_l2), \n",
    "          activation=SoftmaxActivation()\n",
    "         )\n",
    "\n",
    "layers = [\n",
    "    dense_1,\n",
    "    dense_2\n",
    "]\n",
    "\n",
    "model = Model(layers)\n",
    "\n",
    "loss = CategoricalCrossEntropyLoss(loss_smoother=LossSmootherConstant())\n",
    "\n",
    "# defined before\n",
    "# n_epochs = 48\n",
    "# batch_size = 100\n",
    "\n",
    "lr_initial = 1e-5\n",
    "lr_max = 1e-1\n",
    "# defined before\n",
    "# step_size = 800\n",
    "lr_schedule = LRCyclingSchedule(lr_initial, lr_max, step_size)\n",
    "optimizer = SGDOptimizer(lr_schedule=lr_schedule, grad_clipper=GradClipperByNothing())\n",
    "\n",
    "metrics = [AccuracyMetrics()]\n",
    "\n",
    "model.compile_model(optimizer, loss, metrics)\n",
    "print(model)\n",
    "\n",
    "# verbosity level of fit\n",
    "verbose = 2\n",
    "history = model.fit(x_train, y_train, x_val, y_val, n_epochs, batch_size, verbose, None)\n",
    "\n",
    "path_save_losses = \"assets/two_layer/losses.png\"\n",
    "path_save_costs = \"assets/two_layer/costs.png\"\n",
    "path_save_accuracies = \"assets/two_layer/accuracies.png\"\n",
    "path_save_lrs = \"assets/two_layer/lrs.png\"\n",
    "plot_losses(history, path_save_losses)\n",
    "plot_costs(history, path_save_costs)\n",
    "plot_accuracies(history, path_save_accuracies)\n",
    "plot_lrs(history, path_save_lrs)\n",
    "\n",
    "params_test = {\"mode\": \"test\"}\n",
    "scores_test = model.forward(x_test, **params_test)\n",
    "y_hat_test = np.argmax(scores_test, axis=1)\n",
    "metrics_test = model.compute_metrics(y_test, scores_test)\n",
    "\n",
    "print(f\"The test metrics are: {metrics_test}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "aquatic-charleston",
   "metadata": {},
   "source": [
    "Note that the reason for the learning rate curve to be shifted is that the learning rate is always recorded at the end of an epoch.\n",
    "\n",
    "As shown in the results, the cylcial learning rate schedule increases and decrease the learning rate in cycles. This helps escape local minimia during optimization and achieve better model performance."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "potential-omaha",
   "metadata": {},
   "source": [
    "Let's see some of the correct and incorrect predicitions of the trained model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "innovative-necklace",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAEICAYAAACZA4KlAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAAsTAAALEwEAmpwYAAAiEUlEQVR4nO2de3Bkd3XnP6e71Wq9H6MZjTRPezw2sQ02MNikbIgB4zhUZW3+IBUnFcwuG5MKFFALu3FIipgUu3nsQkLVbrE7xBATHN4QCCGJjZfgOGWMx854bDN+zsPz0EiaGWn0nJa6++wffWcjD7/zk0ZqdQvu+VSp1P07/bv31797T9++v2+fc0RVcRznZ59MowfgOE59cGd3nJTgzu44KcGd3XFSgju746QEd3bHSQnu7HVERH5dRO5bQf93ichDtRzTBez7JhH5m0bse7mISL+I7BeR5kaPZS3gzl5HVPVeVb2p0eNYJv8N+ONGDyKGiBwSkRvPPVfVYeD7wB2NG9XawZ19jSAiuUaPwUJEXgd0qeoPa7zdn3jPqzAP9wLvqfE2fypxZ68xInKniLwoIpMi8mMRefsC28u+houIish7ReR54PkFbe8XkQMiclJE/ruIBI+TiHxKRI6IyISIPCYib1hgu0tEviIin0/G8rSI7FpgHxSRr4vIqIgcFJH3R97WLwE/OG/fV4jI/SJyWkSGReQjSXuziPy5iBxP/v783NdoEblBRI6KyO+IyAngc8k4vyYiXxCRCeBdItIlIneLyJCIHBORj4tIdsG+fzP5en5ujl8jIn8FbAX+VkSmROS/JC9/BLhYRLZFD1wKcGevPS8CbwC6gI8BXxCRgcjrbwWuBS5f0PZ2YBfwGuAW4D8YfR8FrgZ6gb8GvioihQX2fwd8CegGvg38T4Dkw+NvgSeATcBbgA+KyC8a+3kl8Oy5JyLSAXwP+AdgELgEeCAx/x7w+mRcVwHXAL+/YFsbk/Fu49++Xt8CfC0Z573APUAp2e6rgZuA/5js+x3AXcA7gc7kPZ5S1d8AXgJ+WVXbVfVPAVS1BLyQjCXdqKr/reIfsBe4JXn8LuChBTYF3nze6xW4ecHz3wYeCPUP7GsMuCp5fBfwvQW2y4HZ5PG1wEvn9f1d4HPGdu8HfmvB89uAfzVe+yLwtgXPfxE4lDy+AZgDCgvsdwEPLnjeDxSBlvP29/3k8T8CHzD2fQi4MdD+L8A7G30uNPpvzd4n/rQiIu8E/hOwPWlqB/oiXY4s0naY6tUztK8PUb3iDVL9kOg8b18nFjyeAQrJPfE2YFBExhfYs8A/G2McAzoWPN9C1alDDCZjtsY/qqpnz+uz8P1uA5qAIRE515ZZ8JrYvi06gPEL7PMzhzt7DUnuCz9D9Wvxw6paFpG9gES6hcIOtwBPJ4+3AscD+3oD8DvJvp5W1YqIjC2yr3McAQ6q6s4lvBZgH3Dpef1vM157nKrDWuMPvd+FbUeoXtn7tPoV/HyOADuMff/EtpMPt0uo3rKkGr9nry1tVE+4UQAR+ffAlcvYzn8WkR4R2QJ8APhy4DUdVO9rR4GciHyU6pV9KfwImEgWylpEJCsiVyar7iG+C/zCguffATaKyAeTBbkOEbk2sX0R+H0RWS8ifcBHgS8scVyo6hBwH/AJEekUkYyI7BCRc/v/C+DDIvJaqXLJgsW3YeDi8zZ5DdXbiMOkHHf2GqKqPwY+ATxM9cR7JdX7xQvlW8BjVO/3/w64O/CafwT+HniO6lfls4RvCULjLAO/THUR7SBwkqoTdRmvfxw4c86hVXUSeGuyjRNUlYQ3JS//OLCH6reBJ4HHk7YL4Z1AHvgx1VuIrwEDyb6/CvxXqguSk8DfUF3wA/gjqh804yLy4aTt14H/fYH7/5lEkgUMZ40gIgrsVNUXGj2WhYjITcBvq+qtjR7LUhGRDVQlw1cH1glShzv7GmOtOrvz049/jXeclOBXdsdJCX5ld5yUUFedvb29Q3vWrb/gfqZwLLakHBWbl6JE14jIEFey1dXY6DJY+98Kl/XFNdrHNi6zm2lczvZOnz7J9NRk8ARZkbOLyM3Ap6j++uovVDUaAtmzbj0f+t0/vOD9ZDLZYHs2a38xkYiXZTJ2v0wmMsVy4WdObF+xMcZYbr9aE7sFrPXt4XK3Z/XTSqRPZXnvq1KxN7ocW6yP5eyf/JOPmV2W/TU+iUL6X1Qjoi4HbhORy+O9HMdpFCu5Z78GeEFVD6jqHNXoqltqMyzHcWrNSpx9Ey//xdbRpO1liMgdIrJHRPZMT02sYHeO46yElTh76MbxJ+4kVHW3qu5S1V1t7Uv96bbjOLVmJc5+lGp01jk2E4jOchxnbbCS1fhHgZ0ichFwDPhV4NfiXZRyJRS1CHLBUaBAOFsTAJmITSNLsbE13+UsgsdWb2Or6vF9rY3V+Bi1Xo2PrkwvZ3vlyIr7Mlfja71SH+tj78oew7KdXVVLIvI+qtFXWeCzqvr0It0cx2kQK9LZVfW7VGOdHcdZ4/jPZR0nJbizO05KcGd3nJTgzu44KaG+2WVVqaY/CxDRmtSKCopJHREJIhORriqVWABN2BaT0JYfwGHb1kgcTF2DXWoeCBPZ3HIltFqPcTlyY2wEfmV3nJTgzu44KcGd3XFSgju746QEd3bHSQl1XY1X4kEoZr9lrFYaJc2rxFJFRVbjqyEAoX3ZPWK2aNBNxLZWWNbqc2w+lrkKHsNOSxULhLnw7UF8jHNzc6bNVnOWERwWwa/sjpMS3NkdJyW4sztOSnBnd5yU4M7uOCnBnd1xUkJ9pTetMDdfDNsinzuWMpGL6Fq5XJNpq0hYQoNoWjuami48qCKWWy9jSHkAlejHcCSvnWXT2KGOyVqRwI9l5MmLKq+ReaxEpDKJVOqx5LB4RRgjWAvAqE4EMDw6bdqefmKfaXvtrp8LthcK9jmsaowjcjL6ld1xUoI7u+OkBHd2x0kJ7uyOkxLc2R0nJbizO05KqKv0NjdX5PChA0FbOfK5kzEktkKTLYO0tLSZtqZcwbTlcraelM+H9zc3P2/2yUbeV1vOHuN8NiI1RY6aVMLRVaK2jJPL29uDcLkugOJcLF9feJBG9S8gHuM1b0i2ANmcPcczM2E5LBPRDfN5e0IK3RtM26N7D5q22dO2LJeVsNQ3Nzdr9pkvhs/FStmWDVfk7CJyCJgEykBJVXetZHuO46wetbiyv0lVT9ZgO47jrCJ+z+44KWGlzq7AfSLymIjcEXqBiNwhIntEZM/ZWfsexHGc1WWlX+OvU9XjIrIBuF9EnlHVBxe+QFV3A7sB1vf317aqgOM4S2ZFV3ZVPZ78HwG+CVxTi0E5jlN7ln1lF5E2IKOqk8njm4A/jPVpKTRz5WXbgraRU2Nmv8lpQ7ZQW/LqaG82bX196+1+EcmuozVsGxkdNftMjI2btlNHjpi2GbE1qqbIe0ONqMKI1JTJ2nJNJhIG2NFqz+PU2dPB9mLxrNlHIxF2zQV7HMWiLcudtWxle1/tHX2mrWPjTtMWk2C7e2xXm5w8EWwvq52kslIOS2/lij2GlXyN7we+mWTGzAF/rar/sILtOY6ziizb2VX1AHBVDcfiOM4q4tKb46QEd3bHSQnu7I6TEtzZHScl1DXqrdCc59KLtwdt2UjyyPbWsNSUb7ajk1ojEtrmwX7TdtGWi0xboSm8v5G+XrPPE4/vNW1TWfs9t3Z1mrZSRCqbK4all/l5+/dM8/MxiceOlrt4wJbeDo6H5aTjQy+afbp77XncvGW7aXvqqUOmrWS875ikuH7zZabt0JFTpq1cnjFt3X32sR47PRTeXiRE0JJSy2VbevMru+OkBHd2x0kJ7uyOkxLc2R0nJbizO05KqOtq/OzZIs88G87TNT1rrwjnjFVwI3UXAM1Ndp652Sk7rv7ZZ/ebNisef/asvb1Mk70Ke9nVl5u2tt5207Y+svqvRiBEsWSv7JYr9iry6LD93qZH7QCUrtbwKn5/f6vZZ9vFW0xbIXI8C00tpi1XCNvae2xFhubNpuml546att52+9q5Y9ugaSuXwvM4HwmsKRu55rJZ26X9yu44KcGd3XFSgju746QEd3bHSQnu7I6TEtzZHScl1FV6UxXmS+Fd5prsvGpni2FpaD6Se+zMuJ3fbccOO4+YZCIJcI1qU6VI7rRyZHujM2fscXTaASjFoi2HCeF8ffM6Zfbp7LBlvukJu8TWM4fsoJYrrgrLVzta7fxuFbEDm0aGJk3bxvUDpk0IS2/zTfY47vu/e03b1OkJ03ZJv30+TozZORYhfKxVY8mYDUnXPhX9yu44acGd3XFSgju746QEd3bHSQnu7I6TEtzZHScl1Fd6q1QoFsPS0GzRjsrq6gznYxvstyWXyUlbqulot/PTtXV2mLazc2FpZWDQjtaamrIlr5iy0tNjj6M0b5TDAjLGEc2ILW02yUbTtu/hh01bGTsq68x0eP5Lk7ESSXb02sBGO0ptYJ197mSy4Tx5Dz1hl+wa3NBl2prX26GWV+7YatpmZ8dN28REeK5i50c2G5ZEJRMp82VvLuks8lkRGRGRpxa09YrI/SLyfPK/Z7HtOI7TWJbyNf4vgZvPa7sTeEBVdwIPJM8dx1nDLOrsSb3180ty3gLckzy+B7i1tsNyHKfWLHeBrl9VhwCS/xusF4rIHSKyR0T2TFullx3HWXVWfTVeVXer6i5V3dXWZi+MOY6zuizX2YdFZAAg+T9SuyE5jrMaLFd6+zZwO/DHyf9vLaVTNit0dIQloN4+e0H/Fa8IJ2YsZG05KVJNilIk+WJzi73NHqNck0R21tNpyzhGBR8gXpIJtUObVMLy1dyMHZH1yN59pu3Hz+8xba+62o4eLGs4gq2zw06WmcvaSSUzGTv6rtBpR+0dHgof63yLfVx2bLfPgf5euyxXt60ccnLUlhynZsLybGnePk+jJ7jBUqS3LwIPA5eJyFEReTdVJ3+riDwPvDV57jjOGmbRK7uq3maY3lLjsTiOs4r4z2UdJyW4sztOSnBnd5yU4M7uOCmhrlFvmUyG1tawrHHgcLgGHMD4WDgxY2nWDgvqW28nFOztteWfjQPhKCmA1rawNFQp21LYXERCy+RsOcmqb1fdn33Y8s3hWmrl3Fmzz8ZBO+rtt97/btPW27POtDXlw3OVydjXl5ialGuyE3COjtuRaJ/7+peD7S+9ZEe9bd5oz/21t73ZtI2fOmzaYr8ezciFX3Pn5sLnVSUSKudXdsdJCe7sjpMS3NkdJyW4sztOSnBnd5yU4M7uOCmhrtLb/Pw8w8PhaNiRYVsKme8Ot99wvS2DXHbZZaats8NO5igROaxsRptFkvxFEgBWKrZkNx+JeGppscOrJBuWoUTs7YleYdq0Ys9H5K0xXwpLQ7H6ZbG5ymZs6e0Hj+41bU/vPxRsv/QiOznka6/abtpymZgkattOnx43bcViWBaNzZUpvZVtGdKv7I6TEtzZHScluLM7TkpwZ3eclODO7jgpoa6r8eWKMjkxG7QVi/bK9PXXvzHYft3115p98pFAkvmSnQ/sxMkJ0zY2GR67RnLCxcYxO2EHR8zM2OPYedlFpq25Ofz5rdir8eWyHawTWdylHDFWKoYqEAmE0Yo9xokz4bkH+Oa3/t60jR4/Emzftt5e6T548IRp27fvpGkrztjBRhpRXmIqhIWl5JR8Nd5xHHd2x0kJ7uyOkxLc2R0nJbizO05KcGd3nJRQV+kNVUrlsLxyw/XXm90u2xkuMzQxOWn2mYnk/Jo7a0tNpybsbVrCUDZrB4vMRWxd7XZATl9Pv2krluxSTlZ+unwhVirLlrwyYktGM8axBJidDpc0mp21j8vU1Ixp+8ZXv23a/uk+W3prN3IeRobBxLh9XCoV+9zJN9nHGrVdTSthGTAmo5m1w2xFcUnlnz4rIiMi8tSCtrtE5JiI7E3+3rbYdhzHaSxL+Rr/l8DNgfY/U9Wrk7/v1nZYjuPUmkWdXVUfBE7XYSyO46wiK1mge5+I7Eu+5pv1lkXkDhHZIyJ7ZmbsezLHcVaX5Tr7p4EdwNXAEPAJ64WqultVd6nqrtbWcAEDx3FWn2U5u6oOq2pZqxEgnwGuqe2wHMepNcuS3kRkQFWHkqdvB56Kvf4chZYCl1/xiqDtikvC8hpAaS4sNY3OhuUdgLHRU6ZtZtLWXUpqSys5Q1rp6uoy+2TElmPGxsZN2+yY/d6KeTsfW2dLWDaardhyXXHWlhuzakfttXSad2/0doQlL52zb+VODB0zbadODpu2n3/dpaattT08/liutpMj4XJjAOVIqa9cJH9ha4s9j01GP4nlNjRKRkVLaNmmc53li8ANQJ+IHAX+ALhBRK6mquodAt6z2HYcx2ksizq7qt4WaL57FcbiOM4q4j+XdZyU4M7uOCnBnd1xUoI7u+OkhLpGvWlFmTfK1tz/wPfMfpu2hhMsTs7YEslsJCJuNhIRl83aU2IlS2xrazP7jJ0ZN20v7N9v2q7f9fOmbdebbjBt02fC7/s73/yS2WdsxE6weOONdoxTj1HiCWDkYDgx4+HDB8w+kZyjXHrJZtNWaLbLYeVy4eM5Pjlm9jlyzJYAx8ds6XDiTCSib9I+V9vbCsH2WLLSfHPYFqkY5Vd2x0kL7uyOkxLc2R0nJbizO05KcGd3nJTgzu44KaGu0psI5Axp68yEXdvs5JNPBtuHR46afQqR5H8ZtSOeJPL5p0aSv6IhJwJMRCSXIy++aNreep2dgHPb1gHTduJAuI7diWeeMftkI+O/9c1vsvv19Jq2PT98KNj+Lz+wJdbjp+yESJORZJTt7Z2mrX9DOHFnwZC7APr715m2zZu2mbYnn7Cl1OlIpGWhEI5itOq5AahayT5t7c2v7I6TEtzZHScluLM7TkpwZ3eclODO7jgpoa6r8U35PJs2bbrgfg/98LFg+8SZWbPP2Zy9krlti11aqZCPlEkypssKtgCYL4dXxwEu3jpo2g4dtlfqd9/9f0zb8OHjwfbjJ+xgl83d3abtkR89Ytqy6zaYtkMHDgXbNw7ax3/dpq32viK5/LJZO/FaPm9dz+xV67JRjgkgl7GDU6679srINu1SWbl8+PyJpJMjY1j3PPZEpI/jOKnAnd1xUoI7u+OkBHd2x0kJ7uyOkxLc2R0nJSylIswW4PPARqAC7FbVT4lIL/BlYDvVqjC/oqp2Yi9gdmaWJ58IB7UUi2fNfnmjPE5ba7jUEUBXhx3okG9uN21ilNUB6DRyzW0a3Gj2KWNLgOVyJIfbUVsqe+7F50xbxZCNrn3rL5h92iOf+ff/6GHTdmrCPmbjw2EJsLnFLl3V1bvetK3vsktN9fXZgTCdXeFiou2RIqP5nC2/VtQ+nqWSLbMW7dgryuXwMZOMLb6JcZybcvb8LuXKXgI+pKo/B7weeK+IXA7cCTygqjuBB5LnjuOsURZ1dlUdUtXHk8eTwH5gE3ALcE/ysnuAW1dpjI7j1IALumcXke3Aq4FHgP5zlVyT//bPqRzHaThLdnYRaQe+DnxQVe1MEz/Z7w4R2SMie6Yj+dodx1ldluTsItJE1dHvVdVvJM3DIjKQ2AeAkVBfVd2tqrtUdVesmILjOKvLos4uIkK1RPN+Vf3kAtO3gduTx7cD36r98BzHqRWisXoxgIhcD/wz8CT8fx3pI1Tv278CbAVeAt6hqnYSMaC3p1ffcsONQVtrzv7cyTWHpbfpUiSXXEQiGRmz70J6evpMW29LeIzre+1vLL39doRdU96WBzNqq6IRZYhsPvy+NWNvb27GlgCHhsNlnABmI3nhMhKWjZpaI+85a58D2XIsb2Ak75qxzXyTLVE1N9sT3N5my7aFgv3erNJhABXDB8uR91wph4/zx//okxw6fCQ4+Yvq7Kr6EHa03VsW6+84ztrAf0HnOCnBnd1xUoI7u+OkBHd2x0kJ7uyOkxLqmnASyZDJtwRNUzNTZrfZseDvdZgt233K87acNFu2P+OOHD1i2gZ6uoPtQy22VNO/yZanSkY5KYDOTjvKa/MW+5fJExNh9bMrEjUWCfRj04CdFDMmX2UNASebtRNHxiS0cqRkV6lkJ3O0bCdP2irx+Lj9S08RO+FkqRSJUsvY723eOFdj0luxGI6wK0XkaL+yO05KcGd3nJTgzu44KcGd3XFSgju746QEd3bHSQl1lt4AI7rt1JQtozUbtbDy2JJXNhIaVp61EwMObuy2t2m0j5y2x55pGjdtGpGuMk1hiRLguRdsefDkSFimnJkOJ/oEyDfbn/l9fd2mrbury7S1NIXnX+ZtaciSoADy7fZ8xOI2540kkFmxT/2xSFTk4YPHTJsYkX4A5UrRtDXlw+fBXNHuo4ZsOztr1z/0K7vjpAR3dsdJCe7sjpMS3NkdJyW4sztOSqjranxpfp7R0aGgbcOAHdzR2x0u81SaigQsRAInClP2Kmeu2Z6S4ly43FHfBrv80+zEuGmLrWZ3d9n5zI4dGzZtnZ3hbUqkDFUma69nnz5tV/QaOTFq2i7etj3YPnw8fPwBRg0lAWB6zg4o2rDBPnesfHKq9nXu7Kx9fszP20rO2Pi4adu40S5t1dERzmF46tQps491nCuVyHE2LY7j/Ezhzu44KcGd3XFSgju746QEd3bHSQnu7I6TEhaV3kRkC/B5YCPV8k+7VfVTInIX8JvAOf3lI6r63di28s1NXLQ1XA4p32yXUFrXE5besl12PrAzY7Zk1LXelsrGJ2w5b2YuPF2Dm+ztladsCW1mxpZ4tm3uNW0njh42bVNnwkE5V77qMrPP6Kidj23ouC3zzUzZASNzRu63ucjl5ZJXXmGP48hLpi1WMDSTCQeMzMzYASPrN6wzbdPTdtBT/0a7dFgsQMXKGyeR5ID5pvC5GAvGWYrOXgI+pKqPi0gH8JiI3J/Y/kxV/8cStuE4ToNZSq23IWAoeTwpIvuBTas9MMdxassF3bOLyHbg1VQruAK8T0T2ichnRcTOVew4TsNZsrOLSDvwdeCDqjoBfBrYAVxN9cr/CaPfHSKyR0T2nJ0N/9zUcZzVZ0nOLiJNVB39XlX9BoCqDqtqWVUrwGeAa0J9VXW3qu5S1V2FFnuxynGc1WVRZ5fq8t7dwH5V/eSC9oEFL3s78FTth+c4Tq1Yymr8dcBvAE+KyN6k7SPAbSJyNdUUYIeA9yy2oUK+iZ0XbwnaXnrpqNlvbjYclbVl0Ja8urttOWZq2r6dmJ21pZW+vnDkUv8Ge7livsPOM3fsqJ3PzFBWAHjFpTtMWy4bliNV7fxuE6fsaLPXXWXLYWMn7ai3M9NhCbO1vd3sM3XWHmNLq51TsFSyJczLrwiPf37O3lcsq90LL54xbVdc+QrTNjU1adrKhvQ2e9Y+Ty3pLW/ks4OlrcY/BMHCXVFN3XGctYX/gs5xUoI7u+OkBHd2x0kJ7uyOkxLc2R0nJdS3/BNAJSwzDEYS8jXnDTnJzq1HLmtLNa0FOzKotyscYQfQ3tEabG+JlHFqzdvRa+3tnaatULC3ObjZ7jc3G5aUThw7ZPbZPBCORATYvnXQtK1fZ4/j0JGwrNg6Zx+048O2BNjTEyk1FfmxVreRrHTakAYBVG3pbevWzabt5El7/JFgNAYHB4LtFcNXAObnw1GF2ZxVpMyv7I6TGtzZHScluLM7TkpwZ3eclODO7jgpwZ3dcVJC3aU3KyGeVZMLwlE4ACUjqSFANmtLELF9bd4cjsoDmDLkmoMHD5p9tm7datqI1F+bnLSjpHLZFtNWLoe32b3OlgBbCvb2jh4/btqGTtpJPVtawzLlQE+4HWDroJ2wMdtkJxcdGjph2s6cCUepxc6BAwcOmLZcznaZ2DHr6ek2bda5Gqsrp6aWZ2t8fmV3nJTgzu44KcGd3XFSgju746QEd3bHSQnu7I6TEuoqvYkIOTMqJyIZWDJDxY5OilGp2JJXsWgn+ZuYCNc2O3nylNmnPZJgMZu13/MLLzxv2np6Npi2jPH53T9g1y+bKdp1yEZOn4z0s6XPpuawVCbYkVyU7e3NRY517JhZCRhjkW19fbYEePq0XRevXLbfW3e3HbVn9Yudpzb2+/Iru+OkBHd2x0kJ7uyOkxLc2R0nJbizO05KWHQ1XkQKwINAc/L6r6nqH4hIL/BlYDvV8k+/oqp2ZMS/bdHajz3IXHhFtRIJhIkFEcRWOcuRBX4rb9nGjXYOt0zG/jwtlewxxnKkHT6817RtWB8eS3uXnaetZJQfApiPrDCv67Vz0FnBHaXIivt8JOdaRKyht9cO8rFWumNBVB0ddh7CWCBMW5tdcqxQsOd/zihFFVvdNyckcv4u5cpeBN6sqldRLc98s4i8HrgTeEBVdwIPJM8dx1mjLOrsWuVctcOm5E+BW4B7kvZ7gFtXY4CO49SGpdZnzyYVXEeA+1X1EaBfVYcAkv/2Lz0cx2k4S3J2VS2r6tXAZuAaEblyqTsQkTtEZI+I7JmenlnmMB3HWSkXtBqvquPAPwE3A8MiMgCQ/A9myFfV3aq6S1V3tbXZWUocx1ldFnV2EVkvIt3J4xbgRuAZ4NvA7cnLbge+tUpjdBynBiwlEGYAuEdEslQ/HL6iqt8RkYeBr4jIu4GXgHcstqFKpcLZs+GghZj0Vs6FZZJc1h5+NiKRlItF02bJIABq1JuKlfaJaSGx97xz507TtmnQHn+hJZxPrhIJ/IjJWp0ddiBPoXDheQM1Nh9Zu+RVxcitB3F505r/mDRbjsiDzUaAD8TnIyb3Wudc7Pwwg8MiLOrsqroPeHWg/RTwlgveo+M4DcF/Qec4KcGd3XFSgju746QEd3bHSQnu7I6TEiSWi6vmOxMZBQ4nT/sAO8FZ/fBxvBwfx8v5aRvHNlVdHzLU1dlftmORPaq6qyE793H4OFI4Dv8a7zgpwZ3dcVJCI519dwP3vRAfx8vxcbycn5lxNOye3XGc+uJf4x0nJbizO05KaIizi8jNIvKsiLwgIg1LVCkih0TkSRHZKyJ76rjfz4rIiIg8taCtV0TuF5Hnk/89DRrHXSJyLJmTvSLytjqMY4uIfF9E9ovI0yLygaS9rnMSGUdd50RECiLyIxF5IhnHx5L2lc2Hqtb1D8gCLwIXA3ngCeDyeo8jGcshoK8B+30j8BrgqQVtfwrcmTy+E/iTBo3jLuDDdZ6PAeA1yeMO4Dng8nrPSWQcdZ0TqukA2pPHTcAjwOtXOh+NuLJfA7ygqgdUdQ74EtVMtalBVR8Ezi8HWvdsvcY46o6qDqnq48njSWA/sIk6z0lkHHVFq9Q8o3MjnH0TcGTB86M0YEITFLhPRB4TkTsaNIZzrKVsve8TkX3J1/xVv51YiIhsp5ospaEZjM8bB9R5TlYjo3MjnD2UT6dR+t91qvoa4JeA94rIGxs0jrXEp4EdVAuCDAGfqNeORaQd+DrwQVWdqNd+lzCOus+JriCjs0UjnP0osGXB883A8QaMA1U9nvwfAb5J9RajUSwpW+9qo6rDyYlWAT5DneZERJqoOti9qvqNpLnucxIaR6PmJNn3OBeY0dmiEc7+KLBTRC4SkTzwq1Qz1dYVEWkTkY5zj4GbgKfivVaVNZGt99zJlPB26jAnUs2seDewX1U/ucBU1zmxxlHvOVm1jM71WmE8b7XxbVRXOl8Efq9BY7iYqhLwBPB0PccBfJHq18F5qt903g2so1oz7/nkf2+DxvFXwJPAvuTkGqjDOK6neiu3D9ib/L2t3nMSGUdd5wR4FfCvyf6eAj6atK9oPvznso6TEvwXdI6TEtzZHScluLM7TkpwZ3eclODO7jgpwZ3dcVKCO7vjpIT/B+l5l3xpcJVBAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "label_names = load_label_names()\n",
    "correct_preds_indices = np.argwhere(y_hat_test == y_test).flatten()\n",
    "incorrect_preds_indices = np.argwhere(y_hat_test != y_test).flatten()\n",
    "np.random.seed(123)\n",
    "np.random.shuffle(correct_preds_indices)\n",
    "np.random.seed(123)\n",
    "np.random.shuffle(incorrect_preds_indices)\n",
    "\n",
    "for preds_idx in np.hstack([correct_preds_indices[:5], incorrect_preds_indices[:5]]):\n",
    "    plt.imshow(x_test_img[preds_idx])\n",
    "    \n",
    "    pred_label = label_names[y_hat_test[preds_idx]]\n",
    "    true_label = label_names[y_test[preds_idx]]\n",
    "    title = \\\n",
    "        f\"{pred_label} (correct)\" \\\n",
    "        if pred_label == true_label \\\n",
    "        else f\"{pred_label} (incorrect, true: {true_label})\"\n",
    "    plt.title(title) \n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "pediatric-civilization",
   "metadata": {},
   "source": [
    "## Hyperparameter Search <a class=\"anchor\" id=\"hyperparameter-search\"></a>\n",
    "\n",
    "The hyperparameter search is conducted for the combination of two hyperpaparameters - the learning rate and the L2 regularization strength. To do that, [hyperopt](#https://github.com/hyperopt/hyperopt), a Bayesian, distributed hyperparameter optimization library is used."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "radio-evening",
   "metadata": {},
   "outputs": [],
   "source": [
    "def objective(x_train, y_train, x_val, y_val, **kwargs):\n",
    "    \n",
    "    reg_rate_l2 = kwargs[\"reg_rate_l2\"]\n",
    "    lr_initial = kwargs[\"lr_initial\"]\n",
    "\n",
    "    params = {\"coeff\": 1.0, \"mean\": 0.0, \"std\":None}\n",
    "\n",
    "    in_dim = x_train.shape[1]\n",
    "    out_dim = np.unique(y_train).size\n",
    "    mid_dim = 50\n",
    "\n",
    "    seed = 200\n",
    "\n",
    "    dense_1 = \\\n",
    "        Dense(in_dim=in_dim, out_dim=mid_dim, \n",
    "              kernel_initializer=XavierInitializer(seed=seed, **params), \n",
    "              bias_initializer=XavierInitializer(seed=seed+1, **params), \n",
    "              kernel_regularizer=L2Regularizer(reg_rate=reg_rate_l2), \n",
    "              activation=ReLUActivation()\n",
    "             )\n",
    "\n",
    "    dense_2 = \\\n",
    "        Dense(in_dim=mid_dim, out_dim=out_dim,\n",
    "              kernel_initializer=XavierInitializer(seed=seed+2, **params), \n",
    "              bias_initializer=XavierInitializer(seed=seed+3, **params), \n",
    "              kernel_regularizer=L2Regularizer(reg_rate=reg_rate_l2), \n",
    "              activation=SoftmaxActivation()\n",
    "             )\n",
    "\n",
    "    layers = [\n",
    "        dense_1,\n",
    "        dense_2\n",
    "    ]\n",
    "\n",
    "    model = Model(layers)\n",
    "\n",
    "    loss = CategoricalCrossEntropyLoss(loss_smoother=LossSmootherConstant())\n",
    "\n",
    "    # as defined from before\n",
    "    n_epochs = 48\n",
    "    batch_size = 100\n",
    "\n",
    "    # lr_initial = 1e-5\n",
    "    lr_max = 1e-1\n",
    "    \n",
    "    # as defined from before\n",
    "    step_size = 800\n",
    "    lr_schedule = LRCyclingSchedule(lr_initial, lr_max, step_size)\n",
    "    optimizer = SGDOptimizer(lr_schedule=lr_schedule, grad_clipper=GradClipperByNothing())\n",
    "\n",
    "    metrics = [AccuracyMetrics()]\n",
    "\n",
    "    model.compile_model(optimizer, loss, metrics)\n",
    "    print(model)\n",
    "\n",
    "    # verbosity level of fit\n",
    "    verbose = 1\n",
    "    history = model.fit(x_train, y_train, x_val, y_val, n_epochs, batch_size, verbose, aug_func=None)\n",
    "\n",
    "    params_val = {\"mode\": \"test\"}\n",
    "    scores_val = model.forward(x_val, **params_val)\n",
    "    y_hat_val = np.argmax(scores_val, axis=1)\n",
    "    metrics_val = model.compute_metrics(y_val, scores_val)\n",
    "    print(f\"val acc: {metrics_val}, with lr_initial={lr_initial}, reg_rate_l2={reg_rate_l2}\")\n",
    "    val_acc = metrics_val['accuracy']\n",
    "\n",
    "    return {\n",
    "        'loss': -val_acc,\n",
    "        'status': STATUS_OK,\n",
    "        'eval_time': time.time(),\n",
    "        'val_acc': val_acc,\n",
    "        'reg_rate_l2': reg_rate_l2,\n",
    "        'lr_initial': lr_initial\n",
    "    }\n",
    "\n",
    "\n",
    "def run_trials(x_train, y_train, x_val, y_val, pickle_saved_path):\n",
    "    # how many additional trials to do after loading saved trials. 1 = save after iteration\n",
    "    trials_step = 1\n",
    "    # initial max_trials. put something small to not have to wait\n",
    "    max_trials = 2\n",
    "\n",
    "    try:  # try to load an already saved trials object, and increase the max\n",
    "        trials = pickle.load(open(pickle_saved_path, \"rb\"))\n",
    "        print(\"Found saved Trials! Loading...\")\n",
    "        max_trials = len(trials.trials) + trials_step\n",
    "        print(\"Rerunning from {} trials to {} (+{}) trials\".format(len(trials.trials), max_trials, trials_step))\n",
    "    except:  # create a new trials object and start searching\n",
    "        trials = Trials()\n",
    "\n",
    "    objective_lambda = lambda kwargs: objective(x_train, y_train, x_val, y_val, **kwargs)\n",
    "\n",
    "    reg_rate_l2_limits = (1e-3, 0.5)\n",
    "    lr_initial_limits = (1e-6, 0.9*1e-1)\n",
    "\n",
    "    space = {\n",
    "        \"reg_rate_l2\": hp.uniform(\"reg_rate_l2\", reg_rate_l2_limits[0], reg_rate_l2_limits[1]),\n",
    "        \"lr_initial\": hp.uniform(\"lr_initial\", lr_initial_limits[0], lr_initial_limits[1]),\n",
    "    }\n",
    "\n",
    "    # max_evals = 2\n",
    "\n",
    "    best = fmin(objective_lambda,\n",
    "                space=space,\n",
    "                algo=tpe.suggest,\n",
    "                max_evals=max_trials,\n",
    "                trials=trials)\n",
    "\n",
    "    print(\"Best:\", best)\n",
    "\n",
    "    # save the trials object\n",
    "    with open(pickle_saved_path, \"wb\") as f:\n",
    "        pickle.dump(trials, f)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "improved-melbourne",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found saved Trials! Loading...\n",
      "Rerunning from 16 trials to 17 (+1) trials\n",
      "model summary:                                         \n",
      "layer 0: dense: \n",
      "\t shape -- in: 3072, out: 50\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.018042^2), reg: l2 with 1.2513e-01\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: relu\n",
      "\n",
      "layer 1: dense: \n",
      "\t shape -- in: 50, out: 10\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.141421^2), reg: l2 with 1.2513e-01\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: softmax\n",
      "\n",
      "categorical cross-entropy loss with loss smoother constant\n",
      "sgd with cycling lr schedule and clipper who does nothing\n",
      "\n",
      "starting epoch: 1 ...                                  \n",
      "epoch 1/48                                             \n",
      " \t -- train loss = 1.7254762913021027 / train cost = 2.8827959978710385\n",
      "\t -- {\"accuracy_train\": 0.3915} \n",
      "\t -- val loss = 1.8043602922021709 / val cost = 2.961679998771107\n",
      "\t -- {\"accuracy_val\": 0.3558} \n",
      "\n",
      "\n",
      "starting epoch: 2 ...                                  \n",
      "epoch 2/48                                             \n",
      " \t -- train loss = 1.6855382602185058 / train cost = 2.126803024609433\n",
      "\t -- {\"accuracy_train\": 0.4056} \n",
      "\t -- val loss = 1.767770517108494 / val cost = 2.209035281499421\n",
      "\t -- {\"accuracy_val\": 0.3737} \n",
      "\n",
      "\n",
      "starting epoch: 3 ...                                  \n",
      "epoch 3/48                                             \n",
      " \t -- train loss = 1.6701085866862064 / train cost = 1.9297863621585847\n",
      "\t -- {\"accuracy_train\": 0.4186} \n",
      "\t -- val loss = 1.7522838380266232 / val cost = 2.0119616134990013\n",
      "\t -- {\"accuracy_val\": 0.3768} \n",
      "\n",
      "\n",
      "starting epoch: 4 ...                                  \n",
      "epoch 4/48                                             \n",
      " \t -- train loss = 1.681596784881709 / train cost = 1.899669483972348\n",
      "\t -- {\"accuracy_train\": 0.4065} \n",
      "\t -- val loss = 1.7606249556323048 / val cost = 1.9786976547229438\n",
      "\t -- {\"accuracy_val\": 0.373} \n",
      "\n",
      "\n",
      "starting epoch: 5 ...                                  \n",
      "epoch 5/48                                             \n",
      " \t -- train loss = 1.6788524423556752 / train cost = 1.8905577679991474\n",
      "\t -- {\"accuracy_train\": 0.4153} \n",
      "\t -- val loss = 1.7569038420006577 / val cost = 1.96860916764413\n",
      "\t -- {\"accuracy_val\": 0.3795} \n",
      "\n",
      "\n",
      "starting epoch: 6 ...                                  \n",
      "epoch 6/48                                             \n",
      " \t -- train loss = 1.7294349628630912 / train cost = 1.9392727242120913\n",
      "\t -- {\"accuracy_train\": 0.3773} \n",
      "\t -- val loss = 1.803675628137379 / val cost = 2.013513389486379\n",
      "\t -- {\"accuracy_val\": 0.3509} \n",
      "\n",
      "\n",
      "starting epoch: 7 ...                                  \n",
      "epoch 7/48                                             \n",
      " \t -- train loss = 1.7140474304229465 / train cost = 1.9205138968154187\n",
      "\t -- {\"accuracy_train\": 0.3987} \n",
      "\t -- val loss = 1.7836990928758911 / val cost = 1.9901655592683634\n",
      "\t -- {\"accuracy_val\": 0.3667} \n",
      "\n",
      "\n",
      "starting epoch: 8 ...                                  \n",
      "epoch 8/48                                             \n",
      " \t -- train loss = 1.7149132419878315 / train cost = 1.9213460769954638\n",
      "\t -- {\"accuracy_train\": 0.3772} \n",
      "\t -- val loss = 1.7847306245110726 / val cost = 1.9911634595187049\n",
      "\t -- {\"accuracy_val\": 0.3439} \n",
      "\n",
      "\n",
      "starting epoch: 9 ...                                  \n",
      "epoch 9/48                                             \n",
      " \t -- train loss = 1.6839747208720899 / train cost = 1.8874544320695448\n",
      "\t -- {\"accuracy_train\": 0.4123} \n",
      "\t -- val loss = 1.7606569210043992 / val cost = 1.964136632201854\n",
      "\t -- {\"accuracy_val\": 0.3721} \n",
      "\n",
      "\n",
      "starting epoch: 10 ...                                 \n",
      "epoch 10/48                                            \n",
      " \t -- train loss = 1.6844956827760587 / train cost = 1.8909872986994534\n",
      "\t -- {\"accuracy_train\": 0.4066} \n",
      "\t -- val loss = 1.7583135790538371 / val cost = 1.9648051949772318\n",
      "\t -- {\"accuracy_val\": 0.3701} \n",
      "\n",
      "\n",
      "starting epoch: 11 ...                                 \n",
      "epoch 11/48                                            \n",
      " \t -- train loss = 1.6808900070081145 / train cost = 1.8895366469660693\n",
      "\t -- {\"accuracy_train\": 0.4119} \n",
      "\t -- val loss = 1.7504961503103762 / val cost = 1.9591427902683312\n",
      "\t -- {\"accuracy_val\": 0.3702} \n",
      "\n",
      "\n",
      "starting epoch: 12 ...                                 \n",
      "epoch 12/48                                            \n",
      " \t -- train loss = 1.6785347234386787 / train cost = 1.8925717368606902\n",
      "\t -- {\"accuracy_train\": 0.4073} \n",
      "\t -- val loss = 1.7597964118010356 / val cost = 1.9738334252230472\n",
      "\t -- {\"accuracy_val\": 0.37} \n",
      "\n",
      "\n",
      "starting epoch: 13 ...                                 \n",
      "epoch 13/48                                            \n",
      " \t -- train loss = 1.6656415839504177 / train cost = 1.878312209324208\n",
      "\t -- {\"accuracy_train\": 0.4168} \n",
      "\t -- val loss = 1.7443671133098917 / val cost = 1.957037738683682\n",
      "\t -- {\"accuracy_val\": 0.3758} \n",
      "\n",
      "\n",
      "starting epoch: 14 ...                                 \n",
      "epoch 14/48                                            \n",
      " \t -- train loss = 1.6458194947047988 / train cost = 1.86098953587189\n",
      "\t -- {\"accuracy_train\": 0.4249} \n",
      "\t -- val loss = 1.7234574116440469 / val cost = 1.938627452811138\n",
      "\t -- {\"accuracy_val\": 0.3856} \n",
      "\n",
      "\n",
      "starting epoch: 15 ...                                 \n",
      "epoch 15/48                                            \n",
      " \t -- train loss = 1.635649032348798 / train cost = 1.8519616577010445\n",
      "\t -- {\"accuracy_train\": 0.438} \n",
      "\t -- val loss = 1.7127312429777395 / val cost = 1.9290438683299862\n",
      "\t -- {\"accuracy_val\": 0.4025} \n",
      "\n",
      "\n",
      "starting epoch: 16 ...                                 \n",
      "epoch 16/48                                            \n",
      " \t -- train loss = 1.6323648417025587 / train cost = 1.850286151357603\n",
      "\t -- {\"accuracy_train\": 0.4374} \n",
      "\t -- val loss = 1.7193998617792008 / val cost = 1.937321171434245\n",
      "\t -- {\"accuracy_val\": 0.3868} \n",
      "\n",
      "\n",
      "starting epoch: 17 ...                                 \n",
      "epoch 17/48                                            \n",
      " \t -- train loss = 1.6309503257817375 / train cost = 1.855237308459272\n",
      "\t -- {\"accuracy_train\": 0.4387} \n",
      "\t -- val loss = 1.7081322375823835 / val cost = 1.932419220259918\n",
      "\t -- {\"accuracy_val\": 0.4039} \n",
      "\n",
      "\n",
      "starting epoch: 18 ...                                 \n",
      "epoch 18/48                                            \n",
      " \t -- train loss = 1.6328949935417736 / train cost = 1.8546658383694263\n",
      "\t -- {\"accuracy_train\": 0.4272} \n",
      "\t -- val loss = 1.7152108012140161 / val cost = 1.936981646041669\n",
      "\t -- {\"accuracy_val\": 0.3981} \n",
      "\n",
      "\n",
      "starting epoch: 19 ...                                 \n",
      "epoch 19/48                                            \n",
      " \t -- train loss = 1.662921912074616 / train cost = 1.8837420529117554\n",
      "\t -- {\"accuracy_train\": 0.416} \n",
      "\t -- val loss = 1.7451876239643567 / val cost = 1.9660077648014962\n",
      "\t -- {\"accuracy_val\": 0.3752} \n",
      "\n",
      "\n",
      "starting epoch: 20 ...                                 \n",
      "epoch 20/48                                            \n",
      " \t -- train loss = 1.6448649559670419 / train cost = 1.8678528068127538\n",
      "\t -- {\"accuracy_train\": 0.4316} \n",
      "\t -- val loss = 1.726428527245552 / val cost = 1.949416378091264\n",
      "\t -- {\"accuracy_val\": 0.389} \n",
      "\n",
      "\n",
      "starting epoch: 21 ...                                 \n",
      "epoch 21/48                                            \n",
      " \t -- train loss = 1.6899742673634643 / train cost = 1.9048841397776441\n",
      "\t -- {\"accuracy_train\": 0.3998} \n",
      "\t -- val loss = 1.7695163126476015 / val cost = 1.9844261850617813\n",
      "\t -- {\"accuracy_val\": 0.3537} \n",
      "\n",
      "\n",
      "starting epoch: 22 ...                                 \n",
      "epoch 22/48                                            \n",
      " \t -- train loss = 1.699154229809655 / train cost = 1.9157907650795252\n",
      "\t -- {\"accuracy_train\": 0.3959} \n",
      "\t -- val loss = 1.7927726435703157 / val cost = 2.009409178840186\n",
      "\t -- {\"accuracy_val\": 0.3601} \n",
      "\n",
      "\n",
      "starting epoch: 23 ...                                 \n",
      "epoch 23/48                                            \n",
      " \t -- train loss = 1.6944419718013364 / train cost = 1.9107350582784783\n",
      "\t -- {\"accuracy_train\": 0.3961} \n",
      "\t -- val loss = 1.7773169469680288 / val cost = 1.9936100334451705\n",
      "\t -- {\"accuracy_val\": 0.3612} \n",
      "\n",
      "\n",
      "starting epoch: 24 ...                                 \n",
      "epoch 24/48                                            \n",
      " \t -- train loss = 1.6990461023874497 / train cost = 1.9069998596133655\n",
      "\t -- {\"accuracy_train\": 0.4074} \n",
      "\t -- val loss = 1.7647725496197888 / val cost = 1.9727263068457046\n",
      "\t -- {\"accuracy_val\": 0.3788} \n",
      "\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "starting epoch: 25 ...                                 \n",
      "epoch 25/48                                            \n",
      " \t -- train loss = 1.6721950882649643 / train cost = 1.880730375473798\n",
      "\t -- {\"accuracy_train\": 0.4199} \n",
      "\t -- val loss = 1.7429604740811229 / val cost = 1.9514957612899564\n",
      "\t -- {\"accuracy_val\": 0.3854} \n",
      "\n",
      "\n",
      "starting epoch: 26 ...                                 \n",
      "epoch 26/48                                            \n",
      " \t -- train loss = 1.6799536860471294 / train cost = 1.8870248483968481\n",
      "\t -- {\"accuracy_train\": 0.4091} \n",
      "\t -- val loss = 1.7630445281801568 / val cost = 1.9701156905298756\n",
      "\t -- {\"accuracy_val\": 0.3743} \n",
      "\n",
      "\n",
      "starting epoch: 27 ...                                 \n",
      "epoch 27/48                                            \n",
      " \t -- train loss = 1.6839635620921711 / train cost = 1.899467472365975\n",
      "\t -- {\"accuracy_train\": 0.4003} \n",
      "\t -- val loss = 1.7555817002409098 / val cost = 1.9710856105147136\n",
      "\t -- {\"accuracy_val\": 0.3678} \n",
      "\n",
      "\n",
      "starting epoch: 28 ...                                 \n",
      "epoch 28/48                                            \n",
      " \t -- train loss = 1.6547819997246036 / train cost = 1.868678681132447\n",
      "\t -- {\"accuracy_train\": 0.4216} \n",
      "\t -- val loss = 1.732532200141647 / val cost = 1.9464288815494903\n",
      "\t -- {\"accuracy_val\": 0.3825} \n",
      "\n",
      "\n",
      "starting epoch: 29 ...                                 \n",
      "epoch 29/48                                            \n",
      " \t -- train loss = 1.6701684329158073 / train cost = 1.8803080324696018\n",
      "\t -- {\"accuracy_train\": 0.4202} \n",
      "\t -- val loss = 1.7453733335575286 / val cost = 1.955512933111323\n",
      "\t -- {\"accuracy_val\": 0.385} \n",
      "\n",
      "\n",
      "starting epoch: 30 ...                                 \n",
      "epoch 30/48                                            \n",
      " \t -- train loss = 1.6452898779091505 / train cost = 1.863017057659506\n",
      "\t -- {\"accuracy_train\": 0.4218} \n",
      "\t -- val loss = 1.7317081697004666 / val cost = 1.949435349450822\n",
      "\t -- {\"accuracy_val\": 0.3824} \n",
      "\n",
      "\n",
      "starting epoch: 31 ...                                 \n",
      "epoch 31/48                                            \n",
      " \t -- train loss = 1.6367971109753379 / train cost = 1.8577785370472804\n",
      "\t -- {\"accuracy_train\": 0.4242} \n",
      "\t -- val loss = 1.7262209749144426 / val cost = 1.947202400986385\n",
      "\t -- {\"accuracy_val\": 0.3903} \n",
      "\n",
      "\n",
      "starting epoch: 32 ...                                 \n",
      "epoch 32/48                                            \n",
      " \t -- train loss = 1.6188835793659875 / train cost = 1.8418714486654049\n",
      "\t -- {\"accuracy_train\": 0.436} \n",
      "\t -- val loss = 1.7060723032368355 / val cost = 1.9290601725362528\n",
      "\t -- {\"accuracy_val\": 0.3941} \n",
      "\n",
      "\n",
      "starting epoch: 33 ...                                 \n",
      "epoch 33/48                                            \n",
      " \t -- train loss = 1.6328798755304532 / train cost = 1.853239717293393\n",
      "\t -- {\"accuracy_train\": 0.4363} \n",
      "\t -- val loss = 1.7182340063302355 / val cost = 1.9385938480931753\n",
      "\t -- {\"accuracy_val\": 0.3929} \n",
      "\n",
      "\n",
      "starting epoch: 34 ...                                 \n",
      "epoch 34/48                                            \n",
      " \t -- train loss = 1.65651076875301 / train cost = 1.8764226190173454\n",
      "\t -- {\"accuracy_train\": 0.4253} \n",
      "\t -- val loss = 1.740612454443975 / val cost = 1.9605243047083103\n",
      "\t -- {\"accuracy_val\": 0.3902} \n",
      "\n",
      "\n",
      "starting epoch: 35 ...                                 \n",
      "epoch 35/48                                            \n",
      " \t -- train loss = 1.6451933976270272 / train cost = 1.8603625328922029\n",
      "\t -- {\"accuracy_train\": 0.4218} \n",
      "\t -- val loss = 1.7274114224862112 / val cost = 1.9425805577513868\n",
      "\t -- {\"accuracy_val\": 0.3888} \n",
      "\n",
      "\n",
      "starting epoch: 36 ...                                 \n",
      "epoch 36/48                                            \n",
      " \t -- train loss = 1.6416835002651613 / train cost = 1.8600480000651518\n",
      "\t -- {\"accuracy_train\": 0.4243} \n",
      "\t -- val loss = 1.7304111166396432 / val cost = 1.9487756164396335\n",
      "\t -- {\"accuracy_val\": 0.3846} \n",
      "\n",
      "\n",
      "starting epoch: 37 ...                                 \n",
      "epoch 37/48                                            \n",
      " \t -- train loss = 1.6523413113060117 / train cost = 1.8637050520339769\n",
      "\t -- {\"accuracy_train\": 0.4254} \n",
      "\t -- val loss = 1.733668927637384 / val cost = 1.9450326683653492\n",
      "\t -- {\"accuracy_val\": 0.3867} \n",
      "\n",
      "\n",
      "starting epoch: 38 ...                                 \n",
      "epoch 38/48                                            \n",
      " \t -- train loss = 1.6905572738680923 / train cost = 1.9049876087638156\n",
      "\t -- {\"accuracy_train\": 0.381} \n",
      "\t -- val loss = 1.7728983591574583 / val cost = 1.9873286940531814\n",
      "\t -- {\"accuracy_val\": 0.3517} \n",
      "\n",
      "\n",
      "starting epoch: 39 ...                                 \n",
      "epoch 39/48                                            \n",
      " \t -- train loss = 1.692715781452763 / train cost = 1.9022603132699967\n",
      "\t -- {\"accuracy_train\": 0.3999} \n",
      "\t -- val loss = 1.7698956680353497 / val cost = 1.9794401998525832\n",
      "\t -- {\"accuracy_val\": 0.3584} \n",
      "\n",
      "\n",
      "starting epoch: 40 ...                                 \n",
      "epoch 40/48                                            \n",
      " \t -- train loss = 1.6943699446439975 / train cost = 1.8977990578084634\n",
      "\t -- {\"accuracy_train\": 0.3999} \n",
      "\t -- val loss = 1.76823559742804 / val cost = 1.9716647105925058\n",
      "\t -- {\"accuracy_val\": 0.3632} \n",
      "\n",
      "\n",
      "starting epoch: 41 ...                                 \n",
      "epoch 41/48                                            \n",
      " \t -- train loss = 1.6877714176143352 / train cost = 1.890853194207951\n",
      "\t -- {\"accuracy_train\": 0.4004} \n",
      "\t -- val loss = 1.764194629368787 / val cost = 1.9672764059624026\n",
      "\t -- {\"accuracy_val\": 0.3619} \n",
      "\n",
      "\n",
      "starting epoch: 42 ...                                 \n",
      "epoch 42/48                                            \n",
      " \t -- train loss = 1.7050012612915795 / train cost = 1.9154594927617985\n",
      "\t -- {\"accuracy_train\": 0.3828} \n",
      "\t -- val loss = 1.7772721861788108 / val cost = 1.9877304176490298\n",
      "\t -- {\"accuracy_val\": 0.3594} \n",
      "\n",
      "\n",
      "starting epoch: 43 ...                                 \n",
      "epoch 43/48                                            \n",
      " \t -- train loss = 1.6898749801247144 / train cost = 1.9021886925171072\n",
      "\t -- {\"accuracy_train\": 0.3937} \n",
      "\t -- val loss = 1.7687512492195456 / val cost = 1.9810649616119385\n",
      "\t -- {\"accuracy_val\": 0.3542} \n",
      "\n",
      "\n",
      "starting epoch: 44 ...                                 \n",
      "epoch 44/48                                            \n",
      " \t -- train loss = 1.6483401361199024 / train cost = 1.8621181140723273\n",
      "\t -- {\"accuracy_train\": 0.4291} \n",
      "\t -- val loss = 1.7225238784049914 / val cost = 1.9363018563574164\n",
      "\t -- {\"accuracy_val\": 0.3973} \n",
      "\n",
      "\n",
      "starting epoch: 45 ...                                 \n",
      "epoch 45/48                                            \n",
      " \t -- train loss = 1.6467815677489104 / train cost = 1.8591527760581772\n",
      "\t -- {\"accuracy_train\": 0.4261} \n",
      "\t -- val loss = 1.7269888583107535 / val cost = 1.9393600666200204\n",
      "\t -- {\"accuracy_val\": 0.3949} \n",
      "\n",
      "\n",
      "starting epoch: 46 ...                                 \n",
      "epoch 46/48                                            \n",
      " \t -- train loss = 1.6312532925161005 / train cost = 1.8501573906721023\n",
      "\t -- {\"accuracy_train\": 0.4364} \n",
      "\t -- val loss = 1.7129353041025865 / val cost = 1.9318394022585883\n",
      "\t -- {\"accuracy_val\": 0.4035} \n",
      "\n",
      "\n",
      "starting epoch: 47 ...                                 \n",
      "epoch 47/48                                            \n",
      " \t -- train loss = 1.6300176830391837 / train cost = 1.8502467650673076\n",
      "\t -- {\"accuracy_train\": 0.4226} \n",
      "\t -- val loss = 1.711994441453512 / val cost = 1.9322235234816358\n",
      "\t -- {\"accuracy_val\": 0.3873} \n",
      "\n",
      "\n",
      "starting epoch: 48 ...                                 \n",
      "epoch 48/48                                            \n",
      " \t -- train loss = 1.6209399594064888 / train cost = 1.839779860886824\n",
      "\t -- {\"accuracy_train\": 0.4445} \n",
      "\t -- val loss = 1.7038226614459797 / val cost = 1.922662562926315\n",
      "\t -- {\"accuracy_val\": 0.4044} \n",
      "\n",
      "\n",
      "val acc: {'accuracy': 0.4044}, with lr_initial=0.04727819283013913, reg_rate_l2=0.12512923608642768\n",
      "100%|██████████| 17/17 [01:04<00:00, 64.34s/trial, best loss: -0.4615]\n",
      "Best: {'lr_initial': 0.041070795239371384, 'reg_rate_l2': 0.05141531425601553}\n",
      "Found saved Trials! Loading...\n",
      "Rerunning from 17 trials to 18 (+1) trials\n",
      "model summary:                                         \n",
      "layer 0: dense: \n",
      "\t shape -- in: 3072, out: 50\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.018042^2), reg: l2 with 4.9159e-01\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: relu\n",
      "\n",
      "layer 1: dense: \n",
      "\t shape -- in: 50, out: 10\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.141421^2), reg: l2 with 4.9159e-01\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: softmax\n",
      "\n",
      "categorical cross-entropy loss with loss smoother constant\n",
      "sgd with cycling lr schedule and clipper who does nothing\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "starting epoch: 1 ...                                  \n",
      "epoch 1/48                                             \n",
      " \t -- train loss = 1.9512864070094933 / train cost = 2.260535400044547\n",
      "\t -- {\"accuracy_train\": 0.3182} \n",
      "\t -- val loss = 1.9750442401459993 / val cost = 2.284293233181053\n",
      "\t -- {\"accuracy_val\": 0.2976} \n",
      "\n",
      "\n",
      "starting epoch: 2 ...                                  \n",
      "epoch 2/48                                             \n",
      " \t -- train loss = 1.952461508265452 / train cost = 2.155795402148968\n",
      "\t -- {\"accuracy_train\": 0.2991} \n",
      "\t -- val loss = 1.9765942371810967 / val cost = 2.1799281310646124\n",
      "\t -- {\"accuracy_val\": 0.2794} \n",
      "\n",
      "\n",
      "starting epoch: 3 ...                                  \n",
      "epoch 3/48                                             \n",
      " \t -- train loss = 1.9560332440903996 / train cost = 2.145195673496071\n",
      "\t -- {\"accuracy_train\": 0.308} \n",
      "\t -- val loss = 1.9785809069406717 / val cost = 2.167743336346343\n",
      "\t -- {\"accuracy_val\": 0.2937} \n",
      "\n",
      "\n",
      "starting epoch: 4 ...                                  \n",
      "epoch 4/48                                             \n",
      " \t -- train loss = 1.9752749619415353 / train cost = 2.156892339162084\n",
      "\t -- {\"accuracy_train\": 0.263} \n",
      "\t -- val loss = 1.9989464865661404 / val cost = 2.180563863786689\n",
      "\t -- {\"accuracy_val\": 0.2579} \n",
      "\n",
      "\n",
      "starting epoch: 5 ...                                  \n",
      "epoch 5/48                                             \n",
      " \t -- train loss = 1.9699472469319916 / train cost = 2.1462162769259887\n",
      "\t -- {\"accuracy_train\": 0.2801} \n",
      "\t -- val loss = 1.9923236463421057 / val cost = 2.168592676336103\n",
      "\t -- {\"accuracy_val\": 0.2658} \n",
      "\n",
      "\n",
      "starting epoch: 6 ...                                  \n",
      "epoch 6/48                                             \n",
      " \t -- train loss = 1.9773398308767158 / train cost = 2.1654403333165253\n",
      "\t -- {\"accuracy_train\": 0.2738} \n",
      "\t -- val loss = 1.9982024868940036 / val cost = 2.186302989333813\n",
      "\t -- {\"accuracy_val\": 0.2647} \n",
      "\n",
      "\n",
      "starting epoch: 7 ...                                  \n",
      "epoch 7/48                                             \n",
      " \t -- train loss = 1.9897410771405528 / train cost = 2.159329702180038\n",
      "\t -- {\"accuracy_train\": 0.2712} \n",
      "\t -- val loss = 2.0061318315492302 / val cost = 2.175720456588716\n",
      "\t -- {\"accuracy_val\": 0.2638} \n",
      "\n",
      "\n",
      "starting epoch: 8 ...                                  \n",
      "epoch 8/48                                             \n",
      " \t -- train loss = 1.9845489416988629 / train cost = 2.1569492074747108\n",
      "\t -- {\"accuracy_train\": 0.2614} \n",
      "\t -- val loss = 2.003464053208217 / val cost = 2.175864318984065\n",
      "\t -- {\"accuracy_val\": 0.2559} \n",
      "\n",
      "\n",
      "starting epoch: 9 ...                                  \n",
      "epoch 9/48                                             \n",
      " \t -- train loss = 1.9628880812650769 / train cost = 2.1324910423439767\n",
      "\t -- {\"accuracy_train\": 0.294} \n",
      "\t -- val loss = 1.9862065218121596 / val cost = 2.1558094828910592\n",
      "\t -- {\"accuracy_val\": 0.282} \n",
      "\n",
      "\n",
      "starting epoch: 10 ...                                 \n",
      "epoch 10/48                                            \n",
      " \t -- train loss = 1.9602009558813025 / train cost = 2.1315306905213056\n",
      "\t -- {\"accuracy_train\": 0.2891} \n",
      "\t -- val loss = 1.9809166317843336 / val cost = 2.1522463664243365\n",
      "\t -- {\"accuracy_val\": 0.276} \n",
      "\n",
      "\n",
      "starting epoch: 11 ...                                 \n",
      "epoch 11/48                                            \n",
      " \t -- train loss = 1.9728640434036075 / train cost = 2.141192615098548\n",
      "\t -- {\"accuracy_train\": 0.2829} \n",
      "\t -- val loss = 1.9887353837834856 / val cost = 2.1570639554784257\n",
      "\t -- {\"accuracy_val\": 0.2753} \n",
      "\n",
      "\n",
      "starting epoch: 12 ...                                 \n",
      "epoch 12/48                                            \n",
      " \t -- train loss = 1.9605853820430559 / train cost = 2.133454378937425\n",
      "\t -- {\"accuracy_train\": 0.2898} \n",
      "\t -- val loss = 1.9844115601004149 / val cost = 2.1572805569947837\n",
      "\t -- {\"accuracy_val\": 0.2721} \n",
      "\n",
      "\n",
      "starting epoch: 13 ...                                 \n",
      "epoch 13/48                                            \n",
      " \t -- train loss = 1.9626215165469352 / train cost = 2.1287845635465734\n",
      "\t -- {\"accuracy_train\": 0.3046} \n",
      "\t -- val loss = 1.9802756324352624 / val cost = 2.1464386794349006\n",
      "\t -- {\"accuracy_val\": 0.2927} \n",
      "\n",
      "\n",
      "starting epoch: 14 ...                                 \n",
      "epoch 14/48                                            \n",
      " \t -- train loss = 1.9538382875999927 / train cost = 2.121992826110273\n",
      "\t -- {\"accuracy_train\": 0.3011} \n",
      "\t -- val loss = 1.9724114437954594 / val cost = 2.14056598230574\n",
      "\t -- {\"accuracy_val\": 0.2908} \n",
      "\n",
      "\n",
      "starting epoch: 15 ...                                 \n",
      "epoch 15/48                                            \n",
      " \t -- train loss = 1.953655982414123 / train cost = 2.1144454610249235\n",
      "\t -- {\"accuracy_train\": 0.3222} \n",
      "\t -- val loss = 1.970749752010006 / val cost = 2.1315392306208065\n",
      "\t -- {\"accuracy_val\": 0.3097} \n",
      "\n",
      "\n",
      "starting epoch: 16 ...                                 \n",
      "epoch 16/48                                            \n",
      " \t -- train loss = 1.951745886942736 / train cost = 2.1149401906950773\n",
      "\t -- {\"accuracy_train\": 0.3251} \n",
      "\t -- val loss = 1.9737498198415575 / val cost = 2.1369441235938984\n",
      "\t -- {\"accuracy_val\": 0.3013} \n",
      "\n",
      "\n",
      "starting epoch: 17 ...                                 \n",
      "epoch 17/48                                            \n",
      " \t -- train loss = 1.9473950703912295 / train cost = 2.1170163326643254\n",
      "\t -- {\"accuracy_train\": 0.3066} \n",
      "\t -- val loss = 1.964780592271881 / val cost = 2.134401854544977\n",
      "\t -- {\"accuracy_val\": 0.3008} \n",
      "\n",
      "\n",
      "starting epoch: 18 ...                                 \n",
      "epoch 18/48                                            \n",
      " \t -- train loss = 1.9523863890740012 / train cost = 2.1204749992837826\n",
      "\t -- {\"accuracy_train\": 0.3133} \n",
      "\t -- val loss = 1.9700834241041658 / val cost = 2.138172034313947\n",
      "\t -- {\"accuracy_val\": 0.3071} \n",
      "\n",
      "\n",
      "starting epoch: 19 ...                                 \n",
      "epoch 19/48                                            \n",
      " \t -- train loss = 1.9599567013083303 / train cost = 2.1309763536624144\n",
      "\t -- {\"accuracy_train\": 0.3027} \n",
      "\t -- val loss = 1.980605019806049 / val cost = 2.151624672160133\n",
      "\t -- {\"accuracy_val\": 0.2858} \n",
      "\n",
      "\n",
      "starting epoch: 20 ...                                 \n",
      "epoch 20/48                                            \n",
      " \t -- train loss = 1.9520651595276242 / train cost = 2.1206973887788965\n",
      "\t -- {\"accuracy_train\": 0.3001} \n",
      "\t -- val loss = 1.9725716854750643 / val cost = 2.1412039147263364\n",
      "\t -- {\"accuracy_val\": 0.2881} \n",
      "\n",
      "\n",
      "starting epoch: 21 ...                                 \n",
      "epoch 21/48                                            \n",
      " \t -- train loss = 1.9725276676947543 / train cost = 2.13458249391371\n",
      "\t -- {\"accuracy_train\": 0.2853} \n",
      "\t -- val loss = 1.9945487901646963 / val cost = 2.1566036163836517\n",
      "\t -- {\"accuracy_val\": 0.271} \n",
      "\n",
      "\n",
      "starting epoch: 22 ...                                 \n",
      "epoch 22/48                                            \n",
      " \t -- train loss = 1.9816204083458404 / train cost = 2.1570845826509695\n",
      "\t -- {\"accuracy_train\": 0.2695} \n",
      "\t -- val loss = 2.011553553225026 / val cost = 2.1870177275301548\n",
      "\t -- {\"accuracy_val\": 0.2541} \n",
      "\n",
      "\n",
      "starting epoch: 23 ...                                 \n",
      "epoch 23/48                                            \n",
      " \t -- train loss = 1.9478432494730946 / train cost = 2.1296049220377116\n",
      "\t -- {\"accuracy_train\": 0.2981} \n",
      "\t -- val loss = 1.9687487234541896 / val cost = 2.150510396018807\n",
      "\t -- {\"accuracy_val\": 0.2789} \n",
      "\n",
      "\n",
      "starting epoch: 24 ...                                 \n",
      "epoch 24/48                                            \n",
      " \t -- train loss = 1.9724933229434567 / train cost = 2.1326685003843977\n",
      "\t -- {\"accuracy_train\": 0.3045} \n",
      "\t -- val loss = 1.987844538094637 / val cost = 2.148019715535578\n",
      "\t -- {\"accuracy_val\": 0.2927} \n",
      "\n",
      "\n",
      "starting epoch: 25 ...                                 \n",
      "epoch 25/48                                            \n",
      " \t -- train loss = 1.9640646599086093 / train cost = 2.1240687837535215\n",
      "\t -- {\"accuracy_train\": 0.2926} \n",
      "\t -- val loss = 1.9813460332827701 / val cost = 2.141350157127682\n",
      "\t -- {\"accuracy_val\": 0.2833} \n",
      "\n",
      "\n",
      "starting epoch: 26 ...                                 \n",
      "epoch 26/48                                            \n",
      " \t -- train loss = 1.9650906950197793 / train cost = 2.129750610748985\n",
      "\t -- {\"accuracy_train\": 0.276} \n",
      "\t -- val loss = 1.9885815432419032 / val cost = 2.153241458971109\n",
      "\t -- {\"accuracy_val\": 0.2622} \n",
      "\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "starting epoch: 27 ...                                 \n",
      "epoch 27/48                                            \n",
      " \t -- train loss = 1.9537256469982611 / train cost = 2.130714675142792\n",
      "\t -- {\"accuracy_train\": 0.3061} \n",
      "\t -- val loss = 1.975131678397928 / val cost = 2.1521207065424592\n",
      "\t -- {\"accuracy_val\": 0.2904} \n",
      "\n",
      "\n",
      "starting epoch: 28 ...                                 \n",
      "epoch 28/48                                            \n",
      " \t -- train loss = 1.9595899091686304 / train cost = 2.1212706432872213\n",
      "\t -- {\"accuracy_train\": 0.3059} \n",
      "\t -- val loss = 1.9802434915059361 / val cost = 2.141924225624527\n",
      "\t -- {\"accuracy_val\": 0.29} \n",
      "\n",
      "\n",
      "starting epoch: 29 ...                                 \n",
      "epoch 29/48                                            \n",
      " \t -- train loss = 1.970840935999388 / train cost = 2.124734484163626\n",
      "\t -- {\"accuracy_train\": 0.2756} \n",
      "\t -- val loss = 1.988811407787101 / val cost = 2.1427049559513396\n",
      "\t -- {\"accuracy_val\": 0.2662} \n",
      "\n",
      "\n",
      "starting epoch: 30 ...                                 \n",
      "epoch 30/48                                            \n",
      " \t -- train loss = 1.9443402085264716 / train cost = 2.113937693596836\n",
      "\t -- {\"accuracy_train\": 0.3103} \n",
      "\t -- val loss = 1.967904476244476 / val cost = 2.1375019613148405\n",
      "\t -- {\"accuracy_val\": 0.2928} \n",
      "\n",
      "\n",
      "starting epoch: 31 ...                                 \n",
      "epoch 31/48                                            \n",
      " \t -- train loss = 1.9441593844045362 / train cost = 2.1168017514615016\n",
      "\t -- {\"accuracy_train\": 0.2957} \n",
      "\t -- val loss = 1.9690332798797954 / val cost = 2.141675646936761\n",
      "\t -- {\"accuracy_val\": 0.2859} \n",
      "\n",
      "\n",
      "starting epoch: 32 ...                                 \n",
      "epoch 32/48                                            \n",
      " \t -- train loss = 1.9405941055225666 / train cost = 2.1078410544606387\n",
      "\t -- {\"accuracy_train\": 0.3184} \n",
      "\t -- val loss = 1.9623303732401978 / val cost = 2.1295773221782697\n",
      "\t -- {\"accuracy_val\": 0.302} \n",
      "\n",
      "\n",
      "starting epoch: 33 ...                                 \n",
      "epoch 33/48                                            \n",
      " \t -- train loss = 1.9519954335671086 / train cost = 2.114760419351328\n",
      "\t -- {\"accuracy_train\": 0.3122} \n",
      "\t -- val loss = 1.974229269778137 / val cost = 2.1369942555623562\n",
      "\t -- {\"accuracy_val\": 0.2921} \n",
      "\n",
      "\n",
      "starting epoch: 34 ...                                 \n",
      "epoch 34/48                                            \n",
      " \t -- train loss = 1.9614063189450721 / train cost = 2.1321318667759748\n",
      "\t -- {\"accuracy_train\": 0.311} \n",
      "\t -- val loss = 1.9800879438202816 / val cost = 2.150813491651184\n",
      "\t -- {\"accuracy_val\": 0.2981} \n",
      "\n",
      "\n",
      "starting epoch: 35 ...                                 \n",
      "epoch 35/48                                            \n",
      " \t -- train loss = 1.9554285578649044 / train cost = 2.1147255929080835\n",
      "\t -- {\"accuracy_train\": 0.3071} \n",
      "\t -- val loss = 1.9751845043853897 / val cost = 2.134481539428569\n",
      "\t -- {\"accuracy_val\": 0.2934} \n",
      "\n",
      "\n",
      "starting epoch: 36 ...                                 \n",
      "epoch 36/48                                            \n",
      " \t -- train loss = 1.9533404504501966 / train cost = 2.1226486239440674\n",
      "\t -- {\"accuracy_train\": 0.306} \n",
      "\t -- val loss = 1.9777665931868664 / val cost = 2.1470747666807375\n",
      "\t -- {\"accuracy_val\": 0.2891} \n",
      "\n",
      "\n",
      "starting epoch: 37 ...                                 \n",
      "epoch 37/48                                            \n",
      " \t -- train loss = 1.9624746922848466 / train cost = 2.1260152540516954\n",
      "\t -- {\"accuracy_train\": 0.2981} \n",
      "\t -- val loss = 1.9823905388259284 / val cost = 2.1459311005927773\n",
      "\t -- {\"accuracy_val\": 0.2838} \n",
      "\n",
      "\n",
      "starting epoch: 38 ...                                 \n",
      "epoch 38/48                                            \n",
      " \t -- train loss = 1.950269669972263 / train cost = 2.1297061169414473\n",
      "\t -- {\"accuracy_train\": 0.274} \n",
      "\t -- val loss = 1.9743605873453187 / val cost = 2.153797034314503\n",
      "\t -- {\"accuracy_val\": 0.2607} \n",
      "\n",
      "\n",
      "starting epoch: 39 ...                                 \n",
      "epoch 39/48                                            \n",
      " \t -- train loss = 1.9872316144487232 / train cost = 2.145595514641364\n",
      "\t -- {\"accuracy_train\": 0.2718} \n",
      "\t -- val loss = 2.0064900800010284 / val cost = 2.164853980193669\n",
      "\t -- {\"accuracy_val\": 0.2623} \n",
      "\n",
      "\n",
      "starting epoch: 40 ...                                 \n",
      "epoch 40/48                                            \n",
      " \t -- train loss = 1.9714364637130748 / train cost = 2.134048158566812\n",
      "\t -- {\"accuracy_train\": 0.2818} \n",
      "\t -- val loss = 1.9894202021981546 / val cost = 2.152031897051892\n",
      "\t -- {\"accuracy_val\": 0.2635} \n",
      "\n",
      "\n",
      "starting epoch: 41 ...                                 \n",
      "epoch 41/48                                            \n",
      " \t -- train loss = 1.970003888974272 / train cost = 2.13835021084999\n",
      "\t -- {\"accuracy_train\": 0.3045} \n",
      "\t -- val loss = 1.994716784649555 / val cost = 2.1630631065252732\n",
      "\t -- {\"accuracy_val\": 0.2844} \n",
      "\n",
      "\n",
      "starting epoch: 42 ...                                 \n",
      "epoch 42/48                                            \n",
      " \t -- train loss = 1.9969272511389442 / train cost = 2.154404941919767\n",
      "\t -- {\"accuracy_train\": 0.256} \n",
      "\t -- val loss = 2.0100677111916507 / val cost = 2.1675454019724736\n",
      "\t -- {\"accuracy_val\": 0.2531} \n",
      "\n",
      "\n",
      "starting epoch: 43 ...                                 \n",
      "epoch 43/48                                            \n",
      " \t -- train loss = 1.968882669661193 / train cost = 2.1352682008265216\n",
      "\t -- {\"accuracy_train\": 0.2896} \n",
      "\t -- val loss = 1.9884243502640981 / val cost = 2.154809881429427\n",
      "\t -- {\"accuracy_val\": 0.2776} \n",
      "\n",
      "\n",
      "starting epoch: 44 ...                                 \n",
      "epoch 44/48                                            \n",
      " \t -- train loss = 1.9495337422713503 / train cost = 2.11554427389213\n",
      "\t -- {\"accuracy_train\": 0.3136} \n",
      "\t -- val loss = 1.9685263169102332 / val cost = 2.1345368485310128\n",
      "\t -- {\"accuracy_val\": 0.3003} \n",
      "\n",
      "\n",
      "starting epoch: 45 ...                                 \n",
      "epoch 45/48                                            \n",
      " \t -- train loss = 1.9504869970357452 / train cost = 2.1129592684730114\n",
      "\t -- {\"accuracy_train\": 0.3047} \n",
      "\t -- val loss = 1.9686455409537251 / val cost = 2.1311178123909915\n",
      "\t -- {\"accuracy_val\": 0.299} \n",
      "\n",
      "\n",
      "starting epoch: 46 ...                                 \n",
      "epoch 46/48                                            \n",
      " \t -- train loss = 1.9566776415151588 / train cost = 2.113339168832824\n",
      "\t -- {\"accuracy_train\": 0.308} \n",
      "\t -- val loss = 1.977769611852857 / val cost = 2.1344311391705224\n",
      "\t -- {\"accuracy_val\": 0.2929} \n",
      "\n",
      "\n",
      "starting epoch: 47 ...                                 \n",
      "epoch 47/48                                            \n",
      " \t -- train loss = 1.94644394703559 / train cost = 2.114904746271725\n",
      "\t -- {\"accuracy_train\": 0.3004} \n",
      "\t -- val loss = 1.966989076963425 / val cost = 2.13544987619956\n",
      "\t -- {\"accuracy_val\": 0.2875} \n",
      "\n",
      "\n",
      "starting epoch: 48 ...                                 \n",
      "epoch 48/48                                            \n",
      " \t -- train loss = 1.948832577921755 / train cost = 2.108987899769654\n",
      "\t -- {\"accuracy_train\": 0.3054} \n",
      "\t -- val loss = 1.970681093987501 / val cost = 2.1308364158354\n",
      "\t -- {\"accuracy_val\": 0.29} \n",
      "\n",
      "\n",
      "val acc: {'accuracy': 0.29}, with lr_initial=0.04773420966448636, reg_rate_l2=0.4915881306629891\n",
      "100%|██████████| 18/18 [01:03<00:00, 63.12s/trial, best loss: -0.4615]\n",
      "Best: {'lr_initial': 0.041070795239371384, 'reg_rate_l2': 0.05141531425601553}\n",
      "Found saved Trials! Loading...\n",
      "Rerunning from 18 trials to 19 (+1) trials\n",
      "model summary:                                         \n",
      "layer 0: dense: \n",
      "\t shape -- in: 3072, out: 50\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.018042^2), reg: l2 with 2.3856e-01\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: relu\n",
      "\n",
      "layer 1: dense: \n",
      "\t shape -- in: 50, out: 10\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.141421^2), reg: l2 with 2.3856e-01\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: softmax\n",
      "\n",
      "categorical cross-entropy loss with loss smoother constant\n",
      "sgd with cycling lr schedule and clipper who does nothing\n",
      "\n",
      "starting epoch: 1 ...                                  \n",
      "epoch 1/48                                             \n",
      " \t -- train loss = 1.7854518028230775 / train cost = 4.378344347113941\n",
      "\t -- {\"accuracy_train\": 0.382} \n",
      "\t -- val loss = 1.8468029158604473 / val cost = 4.439695460151311\n",
      "\t -- {\"accuracy_val\": 0.3498} \n",
      "\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "starting epoch: 2 ...                                  \n",
      "epoch 2/48                                             \n",
      " \t -- train loss = 1.7770752349021095 / train cost = 2.5076907790393688\n",
      "\t -- {\"accuracy_train\": 0.3889} \n",
      "\t -- val loss = 1.829465789064883 / val cost = 2.5600813332021426\n",
      "\t -- {\"accuracy_val\": 0.356} \n",
      "\n",
      "\n",
      "starting epoch: 3 ...                                  \n",
      "epoch 3/48                                             \n",
      " \t -- train loss = 1.7861579580654732 / train cost = 2.0739542653551695\n",
      "\t -- {\"accuracy_train\": 0.3796} \n",
      "\t -- val loss = 1.8323928924667023 / val cost = 2.120189199756399\n",
      "\t -- {\"accuracy_val\": 0.3564} \n",
      "\n",
      "\n",
      "starting epoch: 4 ...                                  \n",
      "epoch 4/48                                             \n",
      " \t -- train loss = 1.8070702330025044 / train cost = 2.024691118654411\n",
      "\t -- {\"accuracy_train\": 0.3632} \n",
      "\t -- val loss = 1.8537258532450607 / val cost = 2.0713467388969673\n",
      "\t -- {\"accuracy_val\": 0.3367} \n",
      "\n",
      "\n",
      "starting epoch: 5 ...                                  \n",
      "epoch 5/48                                             \n",
      " \t -- train loss = 1.803220690775587 / train cost = 2.0097875667267\n",
      "\t -- {\"accuracy_train\": 0.3573} \n",
      "\t -- val loss = 1.8483926237048067 / val cost = 2.0549594996559195\n",
      "\t -- {\"accuracy_val\": 0.3422} \n",
      "\n",
      "\n",
      "starting epoch: 6 ...                                  \n",
      "epoch 6/48                                             \n",
      " \t -- train loss = 1.8409587343386935 / train cost = 2.0481581992954716\n",
      "\t -- {\"accuracy_train\": 0.333} \n",
      "\t -- val loss = 1.8828028954690181 / val cost = 2.0900023604257965\n",
      "\t -- {\"accuracy_val\": 0.3172} \n",
      "\n",
      "\n",
      "starting epoch: 7 ...                                  \n",
      "epoch 7/48                                             \n",
      " \t -- train loss = 1.8452123364080166 / train cost = 2.0422605416607476\n",
      "\t -- {\"accuracy_train\": 0.343} \n",
      "\t -- val loss = 1.8809382318964705 / val cost = 2.0779864371492014\n",
      "\t -- {\"accuracy_val\": 0.3302} \n",
      "\n",
      "\n",
      "starting epoch: 8 ...                                  \n",
      "epoch 8/48                                             \n",
      " \t -- train loss = 1.8431157184545663 / train cost = 2.0414076110189656\n",
      "\t -- {\"accuracy_train\": 0.3228} \n",
      "\t -- val loss = 1.88076869618859 / val cost = 2.0790605887529896\n",
      "\t -- {\"accuracy_val\": 0.3075} \n",
      "\n",
      "\n",
      "starting epoch: 9 ...                                  \n",
      "epoch 9/48                                             \n",
      " \t -- train loss = 1.8166143239667791 / train cost = 2.0099491379553793\n",
      "\t -- {\"accuracy_train\": 0.357} \n",
      "\t -- val loss = 1.862330064688367 / val cost = 2.055664878676967\n",
      "\t -- {\"accuracy_val\": 0.3374} \n",
      "\n",
      "\n",
      "starting epoch: 10 ...                                 \n",
      "epoch 10/48                                            \n",
      " \t -- train loss = 1.8134222560844548 / train cost = 2.007833098943733\n",
      "\t -- {\"accuracy_train\": 0.361} \n",
      "\t -- val loss = 1.853861556466527 / val cost = 2.048272399325805\n",
      "\t -- {\"accuracy_val\": 0.336} \n",
      "\n",
      "\n",
      "starting epoch: 11 ...                                 \n",
      "epoch 11/48                                            \n",
      " \t -- train loss = 1.8209546868265256 / train cost = 2.015317651178279\n",
      "\t -- {\"accuracy_train\": 0.3541} \n",
      "\t -- val loss = 1.8578321994533722 / val cost = 2.0521951638051252\n",
      "\t -- {\"accuracy_val\": 0.3311} \n",
      "\n",
      "\n",
      "starting epoch: 12 ...                                 \n",
      "epoch 12/48                                            \n",
      " \t -- train loss = 1.7954904722547016 / train cost = 1.9974846460598703\n",
      "\t -- {\"accuracy_train\": 0.3657} \n",
      "\t -- val loss = 1.8416832524546212 / val cost = 2.04367742625979\n",
      "\t -- {\"accuracy_val\": 0.3428} \n",
      "\n",
      "\n",
      "starting epoch: 13 ...                                 \n",
      "epoch 13/48                                            \n",
      " \t -- train loss = 1.7945434090155064 / train cost = 1.9939317910715024\n",
      "\t -- {\"accuracy_train\": 0.365} \n",
      "\t -- val loss = 1.8363824086525107 / val cost = 2.0357707907085065\n",
      "\t -- {\"accuracy_val\": 0.344} \n",
      "\n",
      "\n",
      "starting epoch: 14 ...                                 \n",
      "epoch 14/48                                            \n",
      " \t -- train loss = 1.7831908139463937 / train cost = 1.984805350662975\n",
      "\t -- {\"accuracy_train\": 0.3751} \n",
      "\t -- val loss = 1.8255136675196204 / val cost = 2.0271282042362015\n",
      "\t -- {\"accuracy_val\": 0.3459} \n",
      "\n",
      "\n",
      "starting epoch: 15 ...                                 \n",
      "epoch 15/48                                            \n",
      " \t -- train loss = 1.774466021412702 / train cost = 1.9748786236428406\n",
      "\t -- {\"accuracy_train\": 0.3864} \n",
      "\t -- val loss = 1.8176921010065943 / val cost = 2.018104703236733\n",
      "\t -- {\"accuracy_val\": 0.3644} \n",
      "\n",
      "\n",
      "starting epoch: 16 ...                                 \n",
      "epoch 16/48                                            \n",
      " \t -- train loss = 1.7664721402043815 / train cost = 1.971472231427584\n",
      "\t -- {\"accuracy_train\": 0.3882} \n",
      "\t -- val loss = 1.8143748385807463 / val cost = 2.019374929803949\n",
      "\t -- {\"accuracy_val\": 0.3582} \n",
      "\n",
      "\n",
      "starting epoch: 17 ...                                 \n",
      "epoch 17/48                                            \n",
      " \t -- train loss = 1.7659057336636412 / train cost = 1.9747980304749988\n",
      "\t -- {\"accuracy_train\": 0.3869} \n",
      "\t -- val loss = 1.8090697987712119 / val cost = 2.0179620955825697\n",
      "\t -- {\"accuracy_val\": 0.3711} \n",
      "\n",
      "\n",
      "starting epoch: 18 ...                                 \n",
      "epoch 18/48                                            \n",
      " \t -- train loss = 1.7737169005855438 / train cost = 1.9818269948036065\n",
      "\t -- {\"accuracy_train\": 0.3791} \n",
      "\t -- val loss = 1.8183824396141812 / val cost = 2.026492533832244\n",
      "\t -- {\"accuracy_val\": 0.363} \n",
      "\n",
      "\n",
      "starting epoch: 19 ...                                 \n",
      "epoch 19/48                                            \n",
      " \t -- train loss = 1.7878550164768632 / train cost = 1.9937901889201322\n",
      "\t -- {\"accuracy_train\": 0.3754} \n",
      "\t -- val loss = 1.8325941972076885 / val cost = 2.0385293696509574\n",
      "\t -- {\"accuracy_val\": 0.3506} \n",
      "\n",
      "\n",
      "starting epoch: 20 ...                                 \n",
      "epoch 20/48                                            \n",
      " \t -- train loss = 1.7803205785390375 / train cost = 1.9877717291516723\n",
      "\t -- {\"accuracy_train\": 0.3758} \n",
      "\t -- val loss = 1.8252809469449625 / val cost = 2.0327320975575973\n",
      "\t -- {\"accuracy_val\": 0.3596} \n",
      "\n",
      "\n",
      "starting epoch: 21 ...                                 \n",
      "epoch 21/48                                            \n",
      " \t -- train loss = 1.8129081784151735 / train cost = 2.0101084608543753\n",
      "\t -- {\"accuracy_train\": 0.3561} \n",
      "\t -- val loss = 1.8555630631181175 / val cost = 2.0527633455573193\n",
      "\t -- {\"accuracy_val\": 0.3308} \n",
      "\n",
      "\n",
      "starting epoch: 22 ...                                 \n",
      "epoch 22/48                                            \n",
      " \t -- train loss = 1.8333385666542878 / train cost = 2.0365801615106895\n",
      "\t -- {\"accuracy_train\": 0.3312} \n",
      "\t -- val loss = 1.888068386806308 / val cost = 2.0913099816627096\n",
      "\t -- {\"accuracy_val\": 0.3135} \n",
      "\n",
      "\n",
      "starting epoch: 23 ...                                 \n",
      "epoch 23/48                                            \n",
      " \t -- train loss = 1.8073602251485514 / train cost = 2.0120665586318105\n",
      "\t -- {\"accuracy_train\": 0.3575} \n",
      "\t -- val loss = 1.85205692042921 / val cost = 2.056763253912469\n",
      "\t -- {\"accuracy_val\": 0.3365} \n",
      "\n",
      "\n",
      "starting epoch: 24 ...                                 \n",
      "epoch 24/48                                            \n",
      " \t -- train loss = 1.8370108860949006 / train cost = 2.0269333741097815\n",
      "\t -- {\"accuracy_train\": 0.3529} \n",
      "\t -- val loss = 1.8714005478963702 / val cost = 2.061323035911251\n",
      "\t -- {\"accuracy_val\": 0.3389} \n",
      "\n",
      "\n",
      "starting epoch: 25 ...                                 \n",
      "epoch 25/48                                            \n",
      " \t -- train loss = 1.8152071576032238 / train cost = 2.0060671212751533\n",
      "\t -- {\"accuracy_train\": 0.3605} \n",
      "\t -- val loss = 1.8525983760616684 / val cost = 2.043458339733598\n",
      "\t -- {\"accuracy_val\": 0.3392} \n",
      "\n",
      "\n",
      "starting epoch: 26 ...                                 \n",
      "epoch 26/48                                            \n",
      " \t -- train loss = 1.81465809909562 / train cost = 2.0042688488466625\n",
      "\t -- {\"accuracy_train\": 0.3474} \n",
      "\t -- val loss = 1.8603858976769387 / val cost = 2.049996647427981\n",
      "\t -- {\"accuracy_val\": 0.3306} \n",
      "\n",
      "\n",
      "starting epoch: 27 ...                                 \n",
      "epoch 27/48                                            \n",
      " \t -- train loss = 1.8037371265968989 / train cost = 2.005755343368988\n",
      "\t -- {\"accuracy_train\": 0.3633} \n",
      "\t -- val loss = 1.8429368741900045 / val cost = 2.0449550909620937\n",
      "\t -- {\"accuracy_val\": 0.3425} \n",
      "\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "starting epoch: 28 ...                                 \n",
      "epoch 28/48                                            \n",
      " \t -- train loss = 1.7923981535620657 / train cost = 1.989236527950943\n",
      "\t -- {\"accuracy_train\": 0.3764} \n",
      "\t -- val loss = 1.8329921965029852 / val cost = 2.0298305708918627\n",
      "\t -- {\"accuracy_val\": 0.3526} \n",
      "\n",
      "\n",
      "starting epoch: 29 ...                                 \n",
      "epoch 29/48                                            \n",
      " \t -- train loss = 1.7951860780499003 / train cost = 1.9883430701054106\n",
      "\t -- {\"accuracy_train\": 0.3766} \n",
      "\t -- val loss = 1.8359405327084481 / val cost = 2.0290975247639587\n",
      "\t -- {\"accuracy_val\": 0.3564} \n",
      "\n",
      "\n",
      "starting epoch: 30 ...                                 \n",
      "epoch 30/48                                            \n",
      " \t -- train loss = 1.7762361074226891 / train cost = 1.9777458589125345\n",
      "\t -- {\"accuracy_train\": 0.3831} \n",
      "\t -- val loss = 1.8230377672979379 / val cost = 2.024547518787783\n",
      "\t -- {\"accuracy_val\": 0.355} \n",
      "\n",
      "\n",
      "starting epoch: 31 ...                                 \n",
      "epoch 31/48                                            \n",
      " \t -- train loss = 1.7698937930240959 / train cost = 1.9750751438906342\n",
      "\t -- {\"accuracy_train\": 0.3747} \n",
      "\t -- val loss = 1.8168138945306451 / val cost = 2.0219952453971834\n",
      "\t -- {\"accuracy_val\": 0.3592} \n",
      "\n",
      "\n",
      "starting epoch: 32 ...                                 \n",
      "epoch 32/48                                            \n",
      " \t -- train loss = 1.7634279062281577 / train cost = 1.967116719341369\n",
      "\t -- {\"accuracy_train\": 0.3889} \n",
      "\t -- val loss = 1.809478483453771 / val cost = 2.0131672965669822\n",
      "\t -- {\"accuracy_val\": 0.3661} \n",
      "\n",
      "\n",
      "starting epoch: 33 ...                                 \n",
      "epoch 33/48                                            \n",
      " \t -- train loss = 1.772946627450323 / train cost = 1.9761295386411475\n",
      "\t -- {\"accuracy_train\": 0.3873} \n",
      "\t -- val loss = 1.819931548230605 / val cost = 2.02311445942143\n",
      "\t -- {\"accuracy_val\": 0.3609} \n",
      "\n",
      "\n",
      "starting epoch: 34 ...                                 \n",
      "epoch 34/48                                            \n",
      " \t -- train loss = 1.7856067882758841 / train cost = 1.9909324877535675\n",
      "\t -- {\"accuracy_train\": 0.3784} \n",
      "\t -- val loss = 1.8300604501437818 / val cost = 2.0353861496214654\n",
      "\t -- {\"accuracy_val\": 0.3544} \n",
      "\n",
      "\n",
      "starting epoch: 35 ...                                 \n",
      "epoch 35/48                                            \n",
      " \t -- train loss = 1.7828485895882553 / train cost = 1.9812093799405146\n",
      "\t -- {\"accuracy_train\": 0.3744} \n",
      "\t -- val loss = 1.82656189331603 / val cost = 2.024922683668289\n",
      "\t -- {\"accuracy_val\": 0.3529} \n",
      "\n",
      "\n",
      "starting epoch: 36 ...                                 \n",
      "epoch 36/48                                            \n",
      " \t -- train loss = 1.7864102317370307 / train cost = 1.9900683081256345\n",
      "\t -- {\"accuracy_train\": 0.3669} \n",
      "\t -- val loss = 1.835103814449902 / val cost = 2.0387618908385057\n",
      "\t -- {\"accuracy_val\": 0.3455} \n",
      "\n",
      "\n",
      "starting epoch: 37 ...                                 \n",
      "epoch 37/48                                            \n",
      " \t -- train loss = 1.8020958204239153 / train cost = 1.9948430367287153\n",
      "\t -- {\"accuracy_train\": 0.3639} \n",
      "\t -- val loss = 1.845304579177903 / val cost = 2.038051795482703\n",
      "\t -- {\"accuracy_val\": 0.3474} \n",
      "\n",
      "\n",
      "starting epoch: 38 ...                                 \n",
      "epoch 38/48                                            \n",
      " \t -- train loss = 1.8122906566758026 / train cost = 2.013046345579317\n",
      "\t -- {\"accuracy_train\": 0.332} \n",
      "\t -- val loss = 1.8581905899606048 / val cost = 2.058946278864119\n",
      "\t -- {\"accuracy_val\": 0.3144} \n",
      "\n",
      "\n",
      "starting epoch: 39 ...                                 \n",
      "epoch 39/48                                            \n",
      " \t -- train loss = 1.8382346602313189 / train cost = 2.028576053784598\n",
      "\t -- {\"accuracy_train\": 0.3418} \n",
      "\t -- val loss = 1.8803484473804746 / val cost = 2.0706898409337535\n",
      "\t -- {\"accuracy_val\": 0.3123} \n",
      "\n",
      "\n",
      "starting epoch: 40 ...                                 \n",
      "epoch 40/48                                            \n",
      " \t -- train loss = 1.827126959386828 / train cost = 2.0143113458885695\n",
      "\t -- {\"accuracy_train\": 0.3404} \n",
      "\t -- val loss = 1.864165122370688 / val cost = 2.0513495088724296\n",
      "\t -- {\"accuracy_val\": 0.3187} \n",
      "\n",
      "\n",
      "starting epoch: 41 ...                                 \n",
      "epoch 41/48                                            \n",
      " \t -- train loss = 1.826469261446475 / train cost = 2.0147131541954106\n",
      "\t -- {\"accuracy_train\": 0.3545} \n",
      "\t -- val loss = 1.8694167506134163 / val cost = 2.057660643362352\n",
      "\t -- {\"accuracy_val\": 0.3361} \n",
      "\n",
      "\n",
      "starting epoch: 42 ...                                 \n",
      "epoch 42/48                                            \n",
      " \t -- train loss = 1.84000911927454 / train cost = 2.0313617198353797\n",
      "\t -- {\"accuracy_train\": 0.3311} \n",
      "\t -- val loss = 1.8736480943446465 / val cost = 2.0650006949054864\n",
      "\t -- {\"accuracy_val\": 0.3204} \n",
      "\n",
      "\n",
      "starting epoch: 43 ...                                 \n",
      "epoch 43/48                                            \n",
      " \t -- train loss = 1.8119153685222213 / train cost = 2.006633388721493\n",
      "\t -- {\"accuracy_train\": 0.3453} \n",
      "\t -- val loss = 1.8535726895627926 / val cost = 2.0482907097620644\n",
      "\t -- {\"accuracy_val\": 0.3274} \n",
      "\n",
      "\n",
      "starting epoch: 44 ...                                 \n",
      "epoch 44/48                                            \n",
      " \t -- train loss = 1.790218130264145 / train cost = 1.9872670618019783\n",
      "\t -- {\"accuracy_train\": 0.3762} \n",
      "\t -- val loss = 1.8290799439732415 / val cost = 2.0261288755110747\n",
      "\t -- {\"accuracy_val\": 0.3575} \n",
      "\n",
      "\n",
      "starting epoch: 45 ...                                 \n",
      "epoch 45/48                                            \n",
      " \t -- train loss = 1.7840956761887392 / train cost = 1.9779334665736248\n",
      "\t -- {\"accuracy_train\": 0.3795} \n",
      "\t -- val loss = 1.8261563524935966 / val cost = 2.0199941428784824\n",
      "\t -- {\"accuracy_val\": 0.3603} \n",
      "\n",
      "\n",
      "starting epoch: 46 ...                                 \n",
      "epoch 46/48                                            \n",
      " \t -- train loss = 1.7795336746152934 / train cost = 1.975437935184691\n",
      "\t -- {\"accuracy_train\": 0.3832} \n",
      "\t -- val loss = 1.823979529864098 / val cost = 2.019883790433495\n",
      "\t -- {\"accuracy_val\": 0.3655} \n",
      "\n",
      "\n",
      "starting epoch: 47 ...                                 \n",
      "epoch 47/48                                            \n",
      " \t -- train loss = 1.7710500552848703 / train cost = 1.9717089639047005\n",
      "\t -- {\"accuracy_train\": 0.3757} \n",
      "\t -- val loss = 1.814814428300666 / val cost = 2.015473336920496\n",
      "\t -- {\"accuracy_val\": 0.3558} \n",
      "\n",
      "\n",
      "starting epoch: 48 ...                                 \n",
      "epoch 48/48                                            \n",
      " \t -- train loss = 1.767886963500237 / train cost = 1.9681186251602836\n",
      "\t -- {\"accuracy_train\": 0.3778} \n",
      "\t -- val loss = 1.8146765306875408 / val cost = 2.0149081923475873\n",
      "\t -- {\"accuracy_val\": 0.3587} \n",
      "\n",
      "\n",
      "val acc: {'accuracy': 0.3587}, with lr_initial=0.0170348370499877, reg_rate_l2=0.23856390596380986\n",
      "100%|██████████| 19/19 [01:00<00:00, 60.43s/trial, best loss: -0.4615]\n",
      "Best: {'lr_initial': 0.041070795239371384, 'reg_rate_l2': 0.05141531425601553}\n",
      "Found saved Trials! Loading...\n",
      "Rerunning from 19 trials to 20 (+1) trials\n",
      "model summary:                                         \n",
      "layer 0: dense: \n",
      "\t shape -- in: 3072, out: 50\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.018042^2), reg: l2 with 1.5251e-01\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: relu\n",
      "\n",
      "layer 1: dense: \n",
      "\t shape -- in: 50, out: 10\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.141421^2), reg: l2 with 1.5251e-01\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: softmax\n",
      "\n",
      "categorical cross-entropy loss with loss smoother constant\n",
      "sgd with cycling lr schedule and clipper who does nothing\n",
      "\n",
      "starting epoch: 1 ...                                  \n",
      "epoch 1/48                                             \n",
      " \t -- train loss = 1.7936158553846127 / train cost = 4.949689874350621\n",
      "\t -- {\"accuracy_train\": 0.3747} \n",
      "\t -- val loss = 1.8634523450068536 / val cost = 5.019526363972862\n",
      "\t -- {\"accuracy_val\": 0.3408} \n",
      "\n",
      "\n",
      "starting epoch: 2 ...                                  \n",
      "epoch 2/48                                             \n",
      " \t -- train loss = 1.7073904944455747 / train cost = 3.3180785855024872\n",
      "\t -- {\"accuracy_train\": 0.4121} \n",
      "\t -- val loss = 1.7888837469567687 / val cost = 3.3995718380136815\n",
      "\t -- {\"accuracy_val\": 0.3708} \n",
      "\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "starting epoch: 3 ...                                  \n",
      "epoch 3/48                                             \n",
      " \t -- train loss = 1.6842135253726938 / train cost = 2.36916183549781\n",
      "\t -- {\"accuracy_train\": 0.4206} \n",
      "\t -- val loss = 1.760411819629067 / val cost = 2.4453601297541834\n",
      "\t -- {\"accuracy_val\": 0.3792} \n",
      "\n",
      "\n",
      "starting epoch: 4 ...                                  \n",
      "epoch 4/48                                             \n",
      " \t -- train loss = 1.7047309329357017 / train cost = 2.0351439806139267\n",
      "\t -- {\"accuracy_train\": 0.4063} \n",
      "\t -- val loss = 1.7779054845011113 / val cost = 2.108318532179336\n",
      "\t -- {\"accuracy_val\": 0.3698} \n",
      "\n",
      "\n",
      "starting epoch: 5 ...                                  \n",
      "epoch 5/48                                             \n",
      " \t -- train loss = 1.7013606105886818 / train cost = 1.9383687528124953\n",
      "\t -- {\"accuracy_train\": 0.4081} \n",
      "\t -- val loss = 1.7702965904622958 / val cost = 2.0073047326861095\n",
      "\t -- {\"accuracy_val\": 0.378} \n",
      "\n",
      "\n",
      "starting epoch: 6 ...                                  \n",
      "epoch 6/48                                             \n",
      " \t -- train loss = 1.74963427133921 / train cost = 1.9670681325426072\n",
      "\t -- {\"accuracy_train\": 0.3733} \n",
      "\t -- val loss = 1.8138745212230598 / val cost = 2.031308382426457\n",
      "\t -- {\"accuracy_val\": 0.348} \n",
      "\n",
      "\n",
      "starting epoch: 7 ...                                  \n",
      "epoch 7/48                                             \n",
      " \t -- train loss = 1.7532467733119808 / train cost = 1.9610118769372078\n",
      "\t -- {\"accuracy_train\": 0.3783} \n",
      "\t -- val loss = 1.8097475776604965 / val cost = 2.0175126812857234\n",
      "\t -- {\"accuracy_val\": 0.3538} \n",
      "\n",
      "\n",
      "starting epoch: 8 ...                                  \n",
      "epoch 8/48                                             \n",
      " \t -- train loss = 1.7547299656645068 / train cost = 1.9615706580495167\n",
      "\t -- {\"accuracy_train\": 0.3555} \n",
      "\t -- val loss = 1.8119442964316441 / val cost = 2.018784988816654\n",
      "\t -- {\"accuracy_val\": 0.331} \n",
      "\n",
      "\n",
      "starting epoch: 9 ...                                  \n",
      "epoch 9/48                                             \n",
      " \t -- train loss = 1.7199049984596118 / train cost = 1.9220801596847426\n",
      "\t -- {\"accuracy_train\": 0.3952} \n",
      "\t -- val loss = 1.7853103652724036 / val cost = 1.9874855264975344\n",
      "\t -- {\"accuracy_val\": 0.3638} \n",
      "\n",
      "\n",
      "starting epoch: 10 ...                                 \n",
      "epoch 10/48                                            \n",
      " \t -- train loss = 1.716096756872117 / train cost = 1.9210245082750943\n",
      "\t -- {\"accuracy_train\": 0.3972} \n",
      "\t -- val loss = 1.778916392110278 / val cost = 1.9838441435132552\n",
      "\t -- {\"accuracy_val\": 0.3637} \n",
      "\n",
      "\n",
      "starting epoch: 11 ...                                 \n",
      "epoch 11/48                                            \n",
      " \t -- train loss = 1.7118144572396228 / train cost = 1.919334029800617\n",
      "\t -- {\"accuracy_train\": 0.4022} \n",
      "\t -- val loss = 1.771054049219349 / val cost = 1.9785736217803431\n",
      "\t -- {\"accuracy_val\": 0.3692} \n",
      "\n",
      "\n",
      "starting epoch: 12 ...                                 \n",
      "epoch 12/48                                            \n",
      " \t -- train loss = 1.6874724846325309 / train cost = 1.9025401023102604\n",
      "\t -- {\"accuracy_train\": 0.4073} \n",
      "\t -- val loss = 1.7566674003779932 / val cost = 1.9717350180557227\n",
      "\t -- {\"accuracy_val\": 0.3747} \n",
      "\n",
      "\n",
      "starting epoch: 13 ...                                 \n",
      "epoch 13/48                                            \n",
      " \t -- train loss = 1.6830154873767984 / train cost = 1.897008215567774\n",
      "\t -- {\"accuracy_train\": 0.4128} \n",
      "\t -- val loss = 1.7517196685140257 / val cost = 1.9657123967050014\n",
      "\t -- {\"accuracy_val\": 0.3774} \n",
      "\n",
      "\n",
      "starting epoch: 14 ...                                 \n",
      "epoch 14/48                                            \n",
      " \t -- train loss = 1.6684880859196354 / train cost = 1.886313125023402\n",
      "\t -- {\"accuracy_train\": 0.4254} \n",
      "\t -- val loss = 1.7364926495687232 / val cost = 1.9543176886724898\n",
      "\t -- {\"accuracy_val\": 0.386} \n",
      "\n",
      "\n",
      "starting epoch: 15 ...                                 \n",
      "epoch 15/48                                            \n",
      " \t -- train loss = 1.6555592134043469 / train cost = 1.8748516305688105\n",
      "\t -- {\"accuracy_train\": 0.4343} \n",
      "\t -- val loss = 1.7259583015481774 / val cost = 1.945250718712641\n",
      "\t -- {\"accuracy_val\": 0.3972} \n",
      "\n",
      "\n",
      "starting epoch: 16 ...                                 \n",
      "epoch 16/48                                            \n",
      " \t -- train loss = 1.6481029912218779 / train cost = 1.8708108812912652\n",
      "\t -- {\"accuracy_train\": 0.4374} \n",
      "\t -- val loss = 1.722179482742377 / val cost = 1.9448873728117644\n",
      "\t -- {\"accuracy_val\": 0.3968} \n",
      "\n",
      "\n",
      "starting epoch: 17 ...                                 \n",
      "epoch 17/48                                            \n",
      " \t -- train loss = 1.6482672356011425 / train cost = 1.874065463342141\n",
      "\t -- {\"accuracy_train\": 0.4389} \n",
      "\t -- val loss = 1.7186469744784019 / val cost = 1.9444452022194003\n",
      "\t -- {\"accuracy_val\": 0.4034} \n",
      "\n",
      "\n",
      "starting epoch: 18 ...                                 \n",
      "epoch 18/48                                            \n",
      " \t -- train loss = 1.655505820503262 / train cost = 1.8823046428714958\n",
      "\t -- {\"accuracy_train\": 0.4258} \n",
      "\t -- val loss = 1.7295733459756375 / val cost = 1.9563721683438713\n",
      "\t -- {\"accuracy_val\": 0.3946} \n",
      "\n",
      "\n",
      "starting epoch: 19 ...                                 \n",
      "epoch 19/48                                            \n",
      " \t -- train loss = 1.6731713443953802 / train cost = 1.8984916148712725\n",
      "\t -- {\"accuracy_train\": 0.4218} \n",
      "\t -- val loss = 1.7450740410257894 / val cost = 1.9703943115016815\n",
      "\t -- {\"accuracy_val\": 0.3848} \n",
      "\n",
      "\n",
      "starting epoch: 20 ...                                 \n",
      "epoch 20/48                                            \n",
      " \t -- train loss = 1.6666653964595295 / train cost = 1.8923114744036749\n",
      "\t -- {\"accuracy_train\": 0.4237} \n",
      "\t -- val loss = 1.7384575685907753 / val cost = 1.9641036465349206\n",
      "\t -- {\"accuracy_val\": 0.3887} \n",
      "\n",
      "\n",
      "starting epoch: 21 ...                                 \n",
      "epoch 21/48                                            \n",
      " \t -- train loss = 1.702802496092382 / train cost = 1.9188990092314453\n",
      "\t -- {\"accuracy_train\": 0.3971} \n",
      "\t -- val loss = 1.7694785659157486 / val cost = 1.985575079054812\n",
      "\t -- {\"accuracy_val\": 0.3625} \n",
      "\n",
      "\n",
      "starting epoch: 22 ...                                 \n",
      "epoch 22/48                                            \n",
      " \t -- train loss = 1.7334562962307827 / train cost = 1.951074649360002\n",
      "\t -- {\"accuracy_train\": 0.3797} \n",
      "\t -- val loss = 1.8148867801468782 / val cost = 2.0325051332760977\n",
      "\t -- {\"accuracy_val\": 0.35} \n",
      "\n",
      "\n",
      "starting epoch: 23 ...                                 \n",
      "epoch 23/48                                            \n",
      " \t -- train loss = 1.720360862748114 / train cost = 1.9362847724565069\n",
      "\t -- {\"accuracy_train\": 0.3919} \n",
      "\t -- val loss = 1.7886167722552666 / val cost = 2.0045406819636598\n",
      "\t -- {\"accuracy_val\": 0.3595} \n",
      "\n",
      "\n",
      "starting epoch: 24 ...                                 \n",
      "epoch 24/48                                            \n",
      " \t -- train loss = 1.7446813321541008 / train cost = 1.948947815190979\n",
      "\t -- {\"accuracy_train\": 0.3881} \n",
      "\t -- val loss = 1.799589204779308 / val cost = 2.003855687816186\n",
      "\t -- {\"accuracy_val\": 0.3672} \n",
      "\n",
      "\n",
      "starting epoch: 25 ...                                 \n",
      "epoch 25/48                                            \n",
      " \t -- train loss = 1.7137511930727338 / train cost = 1.9190909410151513\n",
      "\t -- {\"accuracy_train\": 0.4025} \n",
      "\t -- val loss = 1.7720802637391655 / val cost = 1.977420011681583\n",
      "\t -- {\"accuracy_val\": 0.3759} \n",
      "\n",
      "\n",
      "starting epoch: 26 ...                                 \n",
      "epoch 26/48                                            \n",
      " \t -- train loss = 1.715222488755902 / train cost = 1.9198564441235617\n",
      "\t -- {\"accuracy_train\": 0.3894} \n",
      "\t -- val loss = 1.7855811583038186 / val cost = 1.990215113671478\n",
      "\t -- {\"accuracy_val\": 0.3643} \n",
      "\n",
      "\n",
      "starting epoch: 27 ...                                 \n",
      "epoch 27/48                                            \n",
      " \t -- train loss = 1.6988013824440251 / train cost = 1.914164168346844\n",
      "\t -- {\"accuracy_train\": 0.4022} \n",
      "\t -- val loss = 1.761880190508907 / val cost = 1.977242976411726\n",
      "\t -- {\"accuracy_val\": 0.3688} \n",
      "\n",
      "\n",
      "starting epoch: 28 ...                                 \n",
      "epoch 28/48                                            \n",
      " \t -- train loss = 1.6801539704492683 / train cost = 1.8944953894371652\n",
      "\t -- {\"accuracy_train\": 0.4176} \n",
      "\t -- val loss = 1.7463488467520543 / val cost = 1.9606902657399512\n",
      "\t -- {\"accuracy_val\": 0.3851} \n",
      "\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "starting epoch: 29 ...                                 \n",
      "epoch 29/48                                            \n",
      " \t -- train loss = 1.679491501473625 / train cost = 1.8924226810705957\n",
      "\t -- {\"accuracy_train\": 0.424} \n",
      "\t -- val loss = 1.7469832613914011 / val cost = 1.959914440988372\n",
      "\t -- {\"accuracy_val\": 0.3924} \n",
      "\n",
      "\n",
      "starting epoch: 30 ...                                 \n",
      "epoch 30/48                                            \n",
      " \t -- train loss = 1.6603965468518664 / train cost = 1.8801522531304853\n",
      "\t -- {\"accuracy_train\": 0.4266} \n",
      "\t -- val loss = 1.7345505413468967 / val cost = 1.9543062476255155\n",
      "\t -- {\"accuracy_val\": 0.387} \n",
      "\n",
      "\n",
      "starting epoch: 31 ...                                 \n",
      "epoch 31/48                                            \n",
      " \t -- train loss = 1.6501594365097823 / train cost = 1.873687530340832\n",
      "\t -- {\"accuracy_train\": 0.4324} \n",
      "\t -- val loss = 1.725342640264268 / val cost = 1.9488707340953177\n",
      "\t -- {\"accuracy_val\": 0.3924} \n",
      "\n",
      "\n",
      "starting epoch: 32 ...                                 \n",
      "epoch 32/48                                            \n",
      " \t -- train loss = 1.64246170870801 / train cost = 1.8663316288496299\n",
      "\t -- {\"accuracy_train\": 0.4407} \n",
      "\t -- val loss = 1.7176968963898598 / val cost = 1.9415668165314797\n",
      "\t -- {\"accuracy_val\": 0.4004} \n",
      "\n",
      "\n",
      "starting epoch: 33 ...                                 \n",
      "epoch 33/48                                            \n",
      " \t -- train loss = 1.649202338123467 / train cost = 1.8737772267386767\n",
      "\t -- {\"accuracy_train\": 0.4354} \n",
      "\t -- val loss = 1.7246031114898501 / val cost = 1.9491780001050598\n",
      "\t -- {\"accuracy_val\": 0.397} \n",
      "\n",
      "\n",
      "starting epoch: 34 ...                                 \n",
      "epoch 34/48                                            \n",
      " \t -- train loss = 1.664874622558234 / train cost = 1.8907447557089958\n",
      "\t -- {\"accuracy_train\": 0.425} \n",
      "\t -- val loss = 1.7399275263033596 / val cost = 1.9657976594541213\n",
      "\t -- {\"accuracy_val\": 0.3927} \n",
      "\n",
      "\n",
      "starting epoch: 35 ...                                 \n",
      "epoch 35/48                                            \n",
      " \t -- train loss = 1.6613824559870438 / train cost = 1.8832723591069944\n",
      "\t -- {\"accuracy_train\": 0.4206} \n",
      "\t -- val loss = 1.735137354637602 / val cost = 1.9570272577575527\n",
      "\t -- {\"accuracy_val\": 0.3891} \n",
      "\n",
      "\n",
      "starting epoch: 36 ...                                 \n",
      "epoch 36/48                                            \n",
      " \t -- train loss = 1.6701089255231796 / train cost = 1.893679546784095\n",
      "\t -- {\"accuracy_train\": 0.4144} \n",
      "\t -- val loss = 1.7462512161953474 / val cost = 1.9698218374562628\n",
      "\t -- {\"accuracy_val\": 0.38} \n",
      "\n",
      "\n",
      "starting epoch: 37 ...                                 \n",
      "epoch 37/48                                            \n",
      " \t -- train loss = 1.6845955771916756 / train cost = 1.8974746788642394\n",
      "\t -- {\"accuracy_train\": 0.41} \n",
      "\t -- val loss = 1.7549830611971595 / val cost = 1.9678621628697233\n",
      "\t -- {\"accuracy_val\": 0.3824} \n",
      "\n",
      "\n",
      "starting epoch: 38 ...                                 \n",
      "epoch 38/48                                            \n",
      " \t -- train loss = 1.716872255004878 / train cost = 1.9323650807267134\n",
      "\t -- {\"accuracy_train\": 0.3743} \n",
      "\t -- val loss = 1.7875041351592502 / val cost = 2.0029969608810854\n",
      "\t -- {\"accuracy_val\": 0.3471} \n",
      "\n",
      "\n",
      "starting epoch: 39 ...                                 \n",
      "epoch 39/48                                            \n",
      " \t -- train loss = 1.7315121597709324 / train cost = 1.939419532167454\n",
      "\t -- {\"accuracy_train\": 0.3879} \n",
      "\t -- val loss = 1.7985490159721713 / val cost = 2.006456388368693\n",
      "\t -- {\"accuracy_val\": 0.346} \n",
      "\n",
      "\n",
      "starting epoch: 40 ...                                 \n",
      "epoch 40/48                                            \n",
      " \t -- train loss = 1.7302786318608174 / train cost = 1.9311408744751084\n",
      "\t -- {\"accuracy_train\": 0.383} \n",
      "\t -- val loss = 1.7903127788285023 / val cost = 1.9911750214427932\n",
      "\t -- {\"accuracy_val\": 0.3499} \n",
      "\n",
      "\n",
      "starting epoch: 41 ...                                 \n",
      "epoch 41/48                                            \n",
      " \t -- train loss = 1.7253294129612546 / train cost = 1.9257510999176723\n",
      "\t -- {\"accuracy_train\": 0.3875} \n",
      "\t -- val loss = 1.7908904928162432 / val cost = 1.991312179772661\n",
      "\t -- {\"accuracy_val\": 0.3565} \n",
      "\n",
      "\n",
      "starting epoch: 42 ...                                 \n",
      "epoch 42/48                                            \n",
      " \t -- train loss = 1.7350499655438993 / train cost = 1.9430052609955706\n",
      "\t -- {\"accuracy_train\": 0.3781} \n",
      "\t -- val loss = 1.7942375806383184 / val cost = 2.0021928760899894\n",
      "\t -- {\"accuracy_val\": 0.353} \n",
      "\n",
      "\n",
      "starting epoch: 43 ...                                 \n",
      "epoch 43/48                                            \n",
      " \t -- train loss = 1.7109763070170323 / train cost = 1.921786237146408\n",
      "\t -- {\"accuracy_train\": 0.385} \n",
      "\t -- val loss = 1.77886515006507 / val cost = 1.9896750801944458\n",
      "\t -- {\"accuracy_val\": 0.3485} \n",
      "\n",
      "\n",
      "starting epoch: 44 ...                                 \n",
      "epoch 44/48                                            \n",
      " \t -- train loss = 1.678466202690232 / train cost = 1.8921624427055042\n",
      "\t -- {\"accuracy_train\": 0.4201} \n",
      "\t -- val loss = 1.7415479996670975 / val cost = 1.9552442396823697\n",
      "\t -- {\"accuracy_val\": 0.3908} \n",
      "\n",
      "\n",
      "starting epoch: 45 ...                                 \n",
      "epoch 45/48                                            \n",
      " \t -- train loss = 1.6675553976616895 / train cost = 1.8808054654444653\n",
      "\t -- {\"accuracy_train\": 0.4286} \n",
      "\t -- val loss = 1.7373889552243282 / val cost = 1.950639023007104\n",
      "\t -- {\"accuracy_val\": 0.3939} \n",
      "\n",
      "\n",
      "starting epoch: 46 ...                                 \n",
      "epoch 46/48                                            \n",
      " \t -- train loss = 1.6582775035582782 / train cost = 1.874912810733645\n",
      "\t -- {\"accuracy_train\": 0.4322} \n",
      "\t -- val loss = 1.7299379993526418 / val cost = 1.9465733065280086\n",
      "\t -- {\"accuracy_val\": 0.4012} \n",
      "\n",
      "\n",
      "starting epoch: 47 ...                                 \n",
      "epoch 47/48                                            \n",
      " \t -- train loss = 1.6492507938070682 / train cost = 1.869607752858934\n",
      "\t -- {\"accuracy_train\": 0.4286} \n",
      "\t -- val loss = 1.720831887159121 / val cost = 1.941188846210987\n",
      "\t -- {\"accuracy_val\": 0.3939} \n",
      "\n",
      "\n",
      "starting epoch: 48 ...                                 \n",
      "epoch 48/48                                            \n",
      " \t -- train loss = 1.6432890153699995 / train cost = 1.8657153638121433\n",
      "\t -- {\"accuracy_train\": 0.434} \n",
      "\t -- val loss = 1.718681619454336 / val cost = 1.9411079678964798\n",
      "\t -- {\"accuracy_val\": 0.3966} \n",
      "\n",
      "\n",
      "val acc: {'accuracy': 0.3966}, with lr_initial=0.006851949625181092, reg_rate_l2=0.15251201880931006\n",
      "100%|██████████| 20/20 [01:00<00:00, 60.38s/trial, best loss: -0.4615]\n",
      "Best: {'lr_initial': 0.041070795239371384, 'reg_rate_l2': 0.05141531425601553}\n",
      "Found saved Trials! Loading...\n",
      "Rerunning from 20 trials to 21 (+1) trials\n",
      "model summary:                                         \n",
      "layer 0: dense: \n",
      "\t shape -- in: 3072, out: 50\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.018042^2), reg: l2 with 1.7056e-02\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: relu\n",
      "\n",
      "layer 1: dense: \n",
      "\t shape -- in: 50, out: 10\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.141421^2), reg: l2 with 1.7056e-02\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: softmax\n",
      "\n",
      "categorical cross-entropy loss with loss smoother constant\n",
      "sgd with cycling lr schedule and clipper who does nothing\n",
      "\n",
      "starting epoch: 1 ...                                  \n",
      "epoch 1/48                                             \n",
      " \t -- train loss = 1.810768671572909 / train cost = 2.3082157477250793\n",
      "\t -- {\"accuracy_train\": 0.3649} \n",
      "\t -- val loss = 1.8880910141547589 / val cost = 2.3855380903069294\n",
      "\t -- {\"accuracy_val\": 0.333} \n",
      "\n",
      "\n",
      "starting epoch: 2 ...                                  \n",
      "epoch 2/48                                             \n",
      " \t -- train loss = 1.6628676998383605 / train cost = 2.1324771756537575\n",
      "\t -- {\"accuracy_train\": 0.4218} \n",
      "\t -- val loss = 1.79872731938018 / val cost = 2.268336795195577\n",
      "\t -- {\"accuracy_val\": 0.369} \n",
      "\n",
      "\n",
      "starting epoch: 3 ...                                  \n",
      "epoch 3/48                                             \n",
      " \t -- train loss = 1.5302423591286174 / train cost = 1.9600265528494345\n",
      "\t -- {\"accuracy_train\": 0.4653} \n",
      "\t -- val loss = 1.7179093429887164 / val cost = 2.1476935367095336\n",
      "\t -- {\"accuracy_val\": 0.3965} \n",
      "\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "starting epoch: 4 ...                                  \n",
      "epoch 4/48                                             \n",
      " \t -- train loss = 1.5084558294472816 / train cost = 1.8939312102686776\n",
      "\t -- {\"accuracy_train\": 0.4707} \n",
      "\t -- val loss = 1.7245036215381448 / val cost = 2.109979002359541\n",
      "\t -- {\"accuracy_val\": 0.4001} \n",
      "\n",
      "\n",
      "starting epoch: 5 ...                                  \n",
      "epoch 5/48                                             \n",
      " \t -- train loss = 1.4652307968689897 / train cost = 1.8074676191058985\n",
      "\t -- {\"accuracy_train\": 0.4881} \n",
      "\t -- val loss = 1.7105334553623697 / val cost = 2.0527702775992784\n",
      "\t -- {\"accuracy_val\": 0.4004} \n",
      "\n",
      "\n",
      "starting epoch: 6 ...                                  \n",
      "epoch 6/48                                             \n",
      " \t -- train loss = 1.509888827114606 / train cost = 1.8129280920257231\n",
      "\t -- {\"accuracy_train\": 0.4667} \n",
      "\t -- val loss = 1.7714992277505872 / val cost = 2.0745384926617043\n",
      "\t -- {\"accuracy_val\": 0.3793} \n",
      "\n",
      "\n",
      "starting epoch: 7 ...                                  \n",
      "epoch 7/48                                             \n",
      " \t -- train loss = 1.456475288014532 / train cost = 1.7286132986983325\n",
      "\t -- {\"accuracy_train\": 0.4901} \n",
      "\t -- val loss = 1.7375002806921995 / val cost = 2.009638291376\n",
      "\t -- {\"accuracy_val\": 0.3868} \n",
      "\n",
      "\n",
      "starting epoch: 8 ...                                  \n",
      "epoch 8/48                                             \n",
      " \t -- train loss = 1.3977445177783732 / train cost = 1.6450727355494574\n",
      "\t -- {\"accuracy_train\": 0.5039} \n",
      "\t -- val loss = 1.6892627130189812 / val cost = 1.9365909307900655\n",
      "\t -- {\"accuracy_val\": 0.4008} \n",
      "\n",
      "\n",
      "starting epoch: 9 ...                                  \n",
      "epoch 9/48                                             \n",
      " \t -- train loss = 1.3500352989708557 / train cost = 1.5824499451024656\n",
      "\t -- {\"accuracy_train\": 0.5285} \n",
      "\t -- val loss = 1.6658591166063874 / val cost = 1.8982737627379973\n",
      "\t -- {\"accuracy_val\": 0.4194} \n",
      "\n",
      "\n",
      "starting epoch: 10 ...                                 \n",
      "epoch 10/48                                            \n",
      " \t -- train loss = 1.3011849138562068 / train cost = 1.5280478040740118\n",
      "\t -- {\"accuracy_train\": 0.5418} \n",
      "\t -- val loss = 1.6568913190214254 / val cost = 1.8837542092392303\n",
      "\t -- {\"accuracy_val\": 0.4244} \n",
      "\n",
      "\n",
      "starting epoch: 11 ...                                 \n",
      "epoch 11/48                                            \n",
      " \t -- train loss = 1.2452026967915863 / train cost = 1.4724302472819255\n",
      "\t -- {\"accuracy_train\": 0.5738} \n",
      "\t -- val loss = 1.6260338844104207 / val cost = 1.85326143490076\n",
      "\t -- {\"accuracy_val\": 0.4305} \n",
      "\n",
      "\n",
      "starting epoch: 12 ...                                 \n",
      "epoch 12/48                                            \n",
      " \t -- train loss = 1.2130082782175935 / train cost = 1.442639113613386\n",
      "\t -- {\"accuracy_train\": 0.5786} \n",
      "\t -- val loss = 1.633138313112121 / val cost = 1.8627691485079136\n",
      "\t -- {\"accuracy_val\": 0.4321} \n",
      "\n",
      "\n",
      "starting epoch: 13 ...                                 \n",
      "epoch 13/48                                            \n",
      " \t -- train loss = 1.1525607556812525 / train cost = 1.3854270333036993\n",
      "\t -- {\"accuracy_train\": 0.6124} \n",
      "\t -- val loss = 1.6006894352050693 / val cost = 1.8335557128275162\n",
      "\t -- {\"accuracy_val\": 0.446} \n",
      "\n",
      "\n",
      "starting epoch: 14 ...                                 \n",
      "epoch 14/48                                            \n",
      " \t -- train loss = 1.1102017417910404 / train cost = 1.3467097422520493\n",
      "\t -- {\"accuracy_train\": 0.6273} \n",
      "\t -- val loss = 1.5779256436547031 / val cost = 1.814433644115712\n",
      "\t -- {\"accuracy_val\": 0.4552} \n",
      "\n",
      "\n",
      "starting epoch: 15 ...                                 \n",
      "epoch 15/48                                            \n",
      " \t -- train loss = 1.0770646730987303 / train cost = 1.316846228718996\n",
      "\t -- {\"accuracy_train\": 0.6461} \n",
      "\t -- val loss = 1.567812977476932 / val cost = 1.8075945330971976\n",
      "\t -- {\"accuracy_val\": 0.4565} \n",
      "\n",
      "\n",
      "starting epoch: 16 ...                                 \n",
      "epoch 16/48                                            \n",
      " \t -- train loss = 1.0607375964219175 / train cost = 1.302209286442792\n",
      "\t -- {\"accuracy_train\": 0.6522} \n",
      "\t -- val loss = 1.5640618482140496 / val cost = 1.8055335382349242\n",
      "\t -- {\"accuracy_val\": 0.456} \n",
      "\n",
      "\n",
      "starting epoch: 17 ...                                 \n",
      "epoch 17/48                                            \n",
      " \t -- train loss = 1.0637988799685614 / train cost = 1.3070646283069827\n",
      "\t -- {\"accuracy_train\": 0.6483} \n",
      "\t -- val loss = 1.567836248654184 / val cost = 1.8111019969926052\n",
      "\t -- {\"accuracy_val\": 0.4573} \n",
      "\n",
      "\n",
      "starting epoch: 18 ...                                 \n",
      "epoch 18/48                                            \n",
      " \t -- train loss = 1.0868960322899235 / train cost = 1.333234370617086\n",
      "\t -- {\"accuracy_train\": 0.6337} \n",
      "\t -- val loss = 1.6053654778741155 / val cost = 1.851703816201278\n",
      "\t -- {\"accuracy_val\": 0.4448} \n",
      "\n",
      "\n",
      "starting epoch: 19 ...                                 \n",
      "epoch 19/48                                            \n",
      " \t -- train loss = 1.1066723162277308 / train cost = 1.3571250555396615\n",
      "\t -- {\"accuracy_train\": 0.6201} \n",
      "\t -- val loss = 1.6377516147997706 / val cost = 1.8882043541117013\n",
      "\t -- {\"accuracy_val\": 0.4381} \n",
      "\n",
      "\n",
      "starting epoch: 20 ...                                 \n",
      "epoch 20/48                                            \n",
      " \t -- train loss = 1.1052683490121218 / train cost = 1.3589602490899078\n",
      "\t -- {\"accuracy_train\": 0.6283} \n",
      "\t -- val loss = 1.6383671133291313 / val cost = 1.8920590134069173\n",
      "\t -- {\"accuracy_val\": 0.4443} \n",
      "\n",
      "\n",
      "starting epoch: 21 ...                                 \n",
      "epoch 21/48                                            \n",
      " \t -- train loss = 1.3421474567684615 / train cost = 1.5976516313069011\n",
      "\t -- {\"accuracy_train\": 0.5341} \n",
      "\t -- val loss = 1.8634793654263697 / val cost = 2.1189835399648094\n",
      "\t -- {\"accuracy_val\": 0.382} \n",
      "\n",
      "\n",
      "starting epoch: 22 ...                                 \n",
      "epoch 22/48                                            \n",
      " \t -- train loss = 1.2208786624060692 / train cost = 1.4751974178273326\n",
      "\t -- {\"accuracy_train\": 0.5763} \n",
      "\t -- val loss = 1.733112704548875 / val cost = 1.9874314599701384\n",
      "\t -- {\"accuracy_val\": 0.4106} \n",
      "\n",
      "\n",
      "starting epoch: 23 ...                                 \n",
      "epoch 23/48                                            \n",
      " \t -- train loss = 1.3194073456217648 / train cost = 1.5729862872049352\n",
      "\t -- {\"accuracy_train\": 0.5466} \n",
      "\t -- val loss = 1.7938668834360654 / val cost = 2.0474458250192358\n",
      "\t -- {\"accuracy_val\": 0.3923} \n",
      "\n",
      "\n",
      "starting epoch: 24 ...                                 \n",
      "epoch 24/48                                            \n",
      " \t -- train loss = 1.2760689507840308 / train cost = 1.527061181761654\n",
      "\t -- {\"accuracy_train\": 0.5567} \n",
      "\t -- val loss = 1.7231121413803692 / val cost = 1.9741043723579923\n",
      "\t -- {\"accuracy_val\": 0.4157} \n",
      "\n",
      "\n",
      "starting epoch: 25 ...                                 \n",
      "epoch 25/48                                            \n",
      " \t -- train loss = 1.181568220584053 / train cost = 1.4339318102609986\n",
      "\t -- {\"accuracy_train\": 0.5954} \n",
      "\t -- val loss = 1.6528906693688616 / val cost = 1.9052542590458073\n",
      "\t -- {\"accuracy_val\": 0.4306} \n",
      "\n",
      "\n",
      "starting epoch: 26 ...                                 \n",
      "epoch 26/48                                            \n",
      " \t -- train loss = 1.1643207007912046 / train cost = 1.4220635123151029\n",
      "\t -- {\"accuracy_train\": 0.599} \n",
      "\t -- val loss = 1.6602266690285397 / val cost = 1.9179694805524379\n",
      "\t -- {\"accuracy_val\": 0.4297} \n",
      "\n",
      "\n",
      "starting epoch: 27 ...                                 \n",
      "epoch 27/48                                            \n",
      " \t -- train loss = 1.1574069351265388 / train cost = 1.421255794314519\n",
      "\t -- {\"accuracy_train\": 0.5983} \n",
      "\t -- val loss = 1.6619003178910259 / val cost = 1.925749177079006\n",
      "\t -- {\"accuracy_val\": 0.4336} \n",
      "\n",
      "\n",
      "starting epoch: 28 ...                                 \n",
      "epoch 28/48                                            \n",
      " \t -- train loss = 1.0643540217642176 / train cost = 1.3327360891489273\n",
      "\t -- {\"accuracy_train\": 0.6441} \n",
      "\t -- val loss = 1.603614277859449 / val cost = 1.8719963452441588\n",
      "\t -- {\"accuracy_val\": 0.4508} \n",
      "\n",
      "\n",
      "starting epoch: 29 ...                                 \n",
      "epoch 29/48                                            \n",
      " \t -- train loss = 1.029473087146619 / train cost = 1.3025347463864545\n",
      "\t -- {\"accuracy_train\": 0.6654} \n",
      "\t -- val loss = 1.593297970817042 / val cost = 1.8663596300568777\n",
      "\t -- {\"accuracy_val\": 0.4497} \n",
      "\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "starting epoch: 30 ...                                 \n",
      "epoch 30/48                                            \n",
      " \t -- train loss = 0.9979409477824541 / train cost = 1.2757572801110926\n",
      "\t -- {\"accuracy_train\": 0.6718} \n",
      "\t -- val loss = 1.5981623754625258 / val cost = 1.8759787077911643\n",
      "\t -- {\"accuracy_val\": 0.4504} \n",
      "\n",
      "\n",
      "starting epoch: 31 ...                                 \n",
      "epoch 31/48                                            \n",
      " \t -- train loss = 0.9656196764186534 / train cost = 1.2471930198600933\n",
      "\t -- {\"accuracy_train\": 0.6886} \n",
      "\t -- val loss = 1.5791974634261505 / val cost = 1.8607708068675906\n",
      "\t -- {\"accuracy_val\": 0.4512} \n",
      "\n",
      "\n",
      "starting epoch: 32 ...                                 \n",
      "epoch 32/48                                            \n",
      " \t -- train loss = 0.9392161916482433 / train cost = 1.2224977633432204\n",
      "\t -- {\"accuracy_train\": 0.7036} \n",
      "\t -- val loss = 1.5638178669009146 / val cost = 1.8470994385958917\n",
      "\t -- {\"accuracy_val\": 0.4593} \n",
      "\n",
      "\n",
      "starting epoch: 33 ...                                 \n",
      "epoch 33/48                                            \n",
      " \t -- train loss = 0.9493759302278874 / train cost = 1.2343450134598997\n",
      "\t -- {\"accuracy_train\": 0.6971} \n",
      "\t -- val loss = 1.574516232392564 / val cost = 1.8594853156245765\n",
      "\t -- {\"accuracy_val\": 0.4565} \n",
      "\n",
      "\n",
      "starting epoch: 34 ...                                 \n",
      "epoch 34/48                                            \n",
      " \t -- train loss = 0.9761839500337194 / train cost = 1.2648940812639846\n",
      "\t -- {\"accuracy_train\": 0.6901} \n",
      "\t -- val loss = 1.6099496132524838 / val cost = 1.898659744482749\n",
      "\t -- {\"accuracy_val\": 0.4493} \n",
      "\n",
      "\n",
      "starting epoch: 35 ...                                 \n",
      "epoch 35/48                                            \n",
      " \t -- train loss = 0.9930772217978591 / train cost = 1.2861395967746079\n",
      "\t -- {\"accuracy_train\": 0.6755} \n",
      "\t -- val loss = 1.6328761818005335 / val cost = 1.9259385567772822\n",
      "\t -- {\"accuracy_val\": 0.4442} \n",
      "\n",
      "\n",
      "starting epoch: 36 ...                                 \n",
      "epoch 36/48                                            \n",
      " \t -- train loss = 1.0321208153675039 / train cost = 1.328772515911963\n",
      "\t -- {\"accuracy_train\": 0.6483} \n",
      "\t -- val loss = 1.686987034769019 / val cost = 1.9836387353134781\n",
      "\t -- {\"accuracy_val\": 0.4338} \n",
      "\n",
      "\n",
      "starting epoch: 37 ...                                 \n",
      "epoch 37/48                                            \n",
      " \t -- train loss = 1.0790173472206466 / train cost = 1.376052826368869\n",
      "\t -- {\"accuracy_train\": 0.6306} \n",
      "\t -- val loss = 1.708015982988235 / val cost = 2.0050514621364575\n",
      "\t -- {\"accuracy_val\": 0.4234} \n",
      "\n",
      "\n",
      "starting epoch: 38 ...                                 \n",
      "epoch 38/48                                            \n",
      " \t -- train loss = 1.145091236786799 / train cost = 1.4398692893391798\n",
      "\t -- {\"accuracy_train\": 0.6087} \n",
      "\t -- val loss = 1.7288404391368393 / val cost = 2.02361849168922\n",
      "\t -- {\"accuracy_val\": 0.4267} \n",
      "\n",
      "\n",
      "starting epoch: 39 ...                                 \n",
      "epoch 39/48                                            \n",
      " \t -- train loss = 1.141221225942407 / train cost = 1.4309938814445249\n",
      "\t -- {\"accuracy_train\": 0.6013} \n",
      "\t -- val loss = 1.7020631595229074 / val cost = 1.9918358150250253\n",
      "\t -- {\"accuracy_val\": 0.4163} \n",
      "\n",
      "\n",
      "starting epoch: 40 ...                                 \n",
      "epoch 40/48                                            \n",
      " \t -- train loss = 1.2316718366594728 / train cost = 1.5162686188170915\n",
      "\t -- {\"accuracy_train\": 0.5714} \n",
      "\t -- val loss = 1.7347544283285252 / val cost = 2.019351210486144\n",
      "\t -- {\"accuracy_val\": 0.4102} \n",
      "\n",
      "\n",
      "starting epoch: 41 ...                                 \n",
      "epoch 41/48                                            \n",
      " \t -- train loss = 1.1251030157278958 / train cost = 1.4074718017922014\n",
      "\t -- {\"accuracy_train\": 0.6162} \n",
      "\t -- val loss = 1.6406237114961688 / val cost = 1.9229924975604744\n",
      "\t -- {\"accuracy_val\": 0.4327} \n",
      "\n",
      "\n",
      "starting epoch: 42 ...                                 \n",
      "epoch 42/48                                            \n",
      " \t -- train loss = 1.1764696916337163 / train cost = 1.463068079460377\n",
      "\t -- {\"accuracy_train\": 0.5876} \n",
      "\t -- val loss = 1.7173440284978758 / val cost = 2.0039424163245365\n",
      "\t -- {\"accuracy_val\": 0.4107} \n",
      "\n",
      "\n",
      "starting epoch: 43 ...                                 \n",
      "epoch 43/48                                            \n",
      " \t -- train loss = 1.10411464081715 / train cost = 1.3946715833610694\n",
      "\t -- {\"accuracy_train\": 0.6208} \n",
      "\t -- val loss = 1.655936688445785 / val cost = 1.9464936309897043\n",
      "\t -- {\"accuracy_val\": 0.4346} \n",
      "\n",
      "\n",
      "starting epoch: 44 ...                                 \n",
      "epoch 44/48                                            \n",
      " \t -- train loss = 1.008450980873492 / train cost = 1.3041282710634523\n",
      "\t -- {\"accuracy_train\": 0.6677} \n",
      "\t -- val loss = 1.5921230590661362 / val cost = 1.8878003492560964\n",
      "\t -- {\"accuracy_val\": 0.4475} \n",
      "\n",
      "\n",
      "starting epoch: 45 ...                                 \n",
      "epoch 45/48                                            \n",
      " \t -- train loss = 0.9732322411047919 / train cost = 1.2736395453418639\n",
      "\t -- {\"accuracy_train\": 0.6888} \n",
      "\t -- val loss = 1.5946913874927229 / val cost = 1.8950986917297947\n",
      "\t -- {\"accuracy_val\": 0.455} \n",
      "\n",
      "\n",
      "starting epoch: 46 ...                                 \n",
      "epoch 46/48                                            \n",
      " \t -- train loss = 0.9385982530133935 / train cost = 1.2430220825887726\n",
      "\t -- {\"accuracy_train\": 0.7027} \n",
      "\t -- val loss = 1.5811083539599793 / val cost = 1.8855321835353585\n",
      "\t -- {\"accuracy_val\": 0.4598} \n",
      "\n",
      "\n",
      "starting epoch: 47 ...                                 \n",
      "epoch 47/48                                            \n",
      " \t -- train loss = 0.9089033945397772 / train cost = 1.2165604704924395\n",
      "\t -- {\"accuracy_train\": 0.7153} \n",
      "\t -- val loss = 1.5681173979112697 / val cost = 1.875774473863932\n",
      "\t -- {\"accuracy_val\": 0.4658} \n",
      "\n",
      "\n",
      "starting epoch: 48 ...                                 \n",
      "epoch 48/48                                            \n",
      " \t -- train loss = 0.8903909321886687 / train cost = 1.1997404054411855\n",
      "\t -- {\"accuracy_train\": 0.7269} \n",
      "\t -- val loss = 1.5600950644843037 / val cost = 1.8694445377368205\n",
      "\t -- {\"accuracy_val\": 0.4635} \n",
      "\n",
      "\n",
      "val acc: {'accuracy': 0.4635}, with lr_initial=0.003418750782037748, reg_rate_l2=0.017056334940157236\n",
      "100%|██████████| 21/21 [01:02<00:00, 62.96s/trial, best loss: -0.4635]\n",
      "Best: {'lr_initial': 0.003418750782037748, 'reg_rate_l2': 0.017056334940157236}\n"
     ]
    }
   ],
   "source": [
    "pickle_saved_path = \"assets/two_layer/two_layer_hyperopt.hyperopt\"\n",
    "# loop indefinitely and stop whenever you like\n",
    "run_for = 5\n",
    "for i in range(run_for):\n",
    "    run_trials(x_train, y_train, x_val, y_val, pickle_saved_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "editorial-principal",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "21\n",
      "[{'loss': -0.2275, 'status': 'ok', 'eval_time': 1628167898.3786826, 'val_acc': 0.2275, 'reg_rate_l2': 0.8733541898810822, 'lr_initial': 0.03531479235905652}, {'loss': -0.3007, 'status': 'ok', 'eval_time': 1628167953.9175081, 'val_acc': 0.3007, 'reg_rate_l2': 0.44957888390078576, 'lr_initial': 0.03757552007448616}, {'loss': -0.2952, 'status': 'ok', 'eval_time': 1628168009.819987, 'val_acc': 0.2952, 'reg_rate_l2': 0.4439174008841718, 'lr_initial': 0.08976089086332134}, {'loss': -0.2287, 'status': 'ok', 'eval_time': 1628168068.0323248, 'val_acc': 0.2287, 'reg_rate_l2': 0.867269908972918, 'lr_initial': 0.07094561279439587}, {'loss': -0.3414, 'status': 'ok', 'eval_time': 1628168124.995717, 'val_acc': 0.3414, 'reg_rate_l2': 0.2495046751966928, 'lr_initial': 0.08324555623250757}, {'loss': -0.283, 'status': 'ok', 'eval_time': 1628168184.4977756, 'val_acc': 0.283, 'reg_rate_l2': 0.5916647953685226, 'lr_initial': 0.009071932769960366}, {'loss': -0.3676, 'status': 'ok', 'eval_time': 1628168780.8982017, 'val_acc': 0.3676, 'reg_rate_l2': 0.20563957547472028, 'lr_initial': 0.025455458972727146}, {'loss': -0.3638, 'status': 'ok', 'eval_time': 1628168842.717839, 'val_acc': 0.3638, 'reg_rate_l2': 0.19773765626340353, 'lr_initial': 0.05567154057341354}, {'loss': -0.3051, 'status': 'ok', 'eval_time': 1628168902.1343253, 'val_acc': 0.3051, 'reg_rate_l2': 0.39566331098540874, 'lr_initial': 0.06987096618963154}, {'loss': -0.4447, 'status': 'ok', 'eval_time': 1628168965.7817807, 'val_acc': 0.4447, 'reg_rate_l2': 0.07779194111312751, 'lr_initial': 0.00904874029242342}, {'loss': -0.373, 'status': 'ok', 'eval_time': 1628169027.943557, 'val_acc': 0.373, 'reg_rate_l2': 0.1802207115342738, 'lr_initial': 0.05587434065110954}, {'loss': -0.4615, 'status': 'ok', 'eval_time': 1628169874.241304, 'val_acc': 0.4615, 'reg_rate_l2': 0.05141531425601553, 'lr_initial': 0.041070795239371384}, {'loss': -0.439, 'status': 'ok', 'eval_time': 1628169930.4812906, 'val_acc': 0.439, 'reg_rate_l2': 0.06380905695939129, 'lr_initial': 0.06842479935255724}, {'loss': -0.367, 'status': 'ok', 'eval_time': 1628169986.2973452, 'val_acc': 0.367, 'reg_rate_l2': 0.21287170399119315, 'lr_initial': 0.022872345264861062}, {'loss': -0.3051, 'status': 'ok', 'eval_time': 1628170042.017691, 'val_acc': 0.3051, 'reg_rate_l2': 0.4088758457004862, 'lr_initial': 0.0740982919356081}, {'loss': -0.3655, 'status': 'ok', 'eval_time': 1628170097.8019488, 'val_acc': 0.3655, 'reg_rate_l2': 0.1988790408574803, 'lr_initial': 0.04376397246929965}, {'loss': -0.4044, 'status': 'ok', 'eval_time': 1631283045.5381234, 'val_acc': 0.4044, 'reg_rate_l2': 0.12512923608642768, 'lr_initial': 0.04727819283013913}, {'loss': -0.29, 'status': 'ok', 'eval_time': 1631283108.6598809, 'val_acc': 0.29, 'reg_rate_l2': 0.4915881306629891, 'lr_initial': 0.04773420966448636}, {'loss': -0.3587, 'status': 'ok', 'eval_time': 1631283169.093296, 'val_acc': 0.3587, 'reg_rate_l2': 0.23856390596380986, 'lr_initial': 0.0170348370499877}, {'loss': -0.3966, 'status': 'ok', 'eval_time': 1631283229.4813895, 'val_acc': 0.3966, 'reg_rate_l2': 0.15251201880931006, 'lr_initial': 0.006851949625181092}, {'loss': -0.4635, 'status': 'ok', 'eval_time': 1631283292.4526088, 'val_acc': 0.4635, 'reg_rate_l2': 0.017056334940157236, 'lr_initial': 0.003418750782037748}]\n",
      "best_trial\n",
      "{'loss': -0.4635, 'status': 'ok', 'eval_time': 1631283292.4526088, 'val_acc': 0.4635, 'reg_rate_l2': 0.017056334940157236, 'lr_initial': 0.003418750782037748}\n"
     ]
    }
   ],
   "source": [
    "trials = pickle.load(open(pickle_saved_path, \"rb\"))\n",
    "print(len(trials.trials))\n",
    "print(trials.results)\n",
    "\n",
    "best_trial = max(trials.results, key=lambda x: x['val_acc'])\n",
    "print(\"best_trial\")\n",
    "print(best_trial)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "fallen-anaheim",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>loss</th>\n",
       "      <th>val_acc</th>\n",
       "      <th>reg_rate_l2</th>\n",
       "      <th>lr_initial</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>-0.2275</td>\n",
       "      <td>0.2275</td>\n",
       "      <td>0.873354</td>\n",
       "      <td>0.035315</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>-0.3007</td>\n",
       "      <td>0.3007</td>\n",
       "      <td>0.449579</td>\n",
       "      <td>0.037576</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>-0.2952</td>\n",
       "      <td>0.2952</td>\n",
       "      <td>0.443917</td>\n",
       "      <td>0.089761</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>-0.2287</td>\n",
       "      <td>0.2287</td>\n",
       "      <td>0.867270</td>\n",
       "      <td>0.070946</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>-0.3414</td>\n",
       "      <td>0.3414</td>\n",
       "      <td>0.249505</td>\n",
       "      <td>0.083246</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>-0.2830</td>\n",
       "      <td>0.2830</td>\n",
       "      <td>0.591665</td>\n",
       "      <td>0.009072</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>-0.3676</td>\n",
       "      <td>0.3676</td>\n",
       "      <td>0.205640</td>\n",
       "      <td>0.025455</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>-0.3638</td>\n",
       "      <td>0.3638</td>\n",
       "      <td>0.197738</td>\n",
       "      <td>0.055672</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>-0.3051</td>\n",
       "      <td>0.3051</td>\n",
       "      <td>0.395663</td>\n",
       "      <td>0.069871</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>-0.4447</td>\n",
       "      <td>0.4447</td>\n",
       "      <td>0.077792</td>\n",
       "      <td>0.009049</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>-0.3730</td>\n",
       "      <td>0.3730</td>\n",
       "      <td>0.180221</td>\n",
       "      <td>0.055874</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>-0.4615</td>\n",
       "      <td>0.4615</td>\n",
       "      <td>0.051415</td>\n",
       "      <td>0.041071</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>-0.4390</td>\n",
       "      <td>0.4390</td>\n",
       "      <td>0.063809</td>\n",
       "      <td>0.068425</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>-0.3670</td>\n",
       "      <td>0.3670</td>\n",
       "      <td>0.212872</td>\n",
       "      <td>0.022872</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>-0.3051</td>\n",
       "      <td>0.3051</td>\n",
       "      <td>0.408876</td>\n",
       "      <td>0.074098</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>-0.3655</td>\n",
       "      <td>0.3655</td>\n",
       "      <td>0.198879</td>\n",
       "      <td>0.043764</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>-0.4044</td>\n",
       "      <td>0.4044</td>\n",
       "      <td>0.125129</td>\n",
       "      <td>0.047278</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>-0.2900</td>\n",
       "      <td>0.2900</td>\n",
       "      <td>0.491588</td>\n",
       "      <td>0.047734</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>-0.3587</td>\n",
       "      <td>0.3587</td>\n",
       "      <td>0.238564</td>\n",
       "      <td>0.017035</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>-0.3966</td>\n",
       "      <td>0.3966</td>\n",
       "      <td>0.152512</td>\n",
       "      <td>0.006852</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>-0.4635</td>\n",
       "      <td>0.4635</td>\n",
       "      <td>0.017056</td>\n",
       "      <td>0.003419</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      loss  val_acc  reg_rate_l2  lr_initial\n",
       "0  -0.2275   0.2275     0.873354    0.035315\n",
       "1  -0.3007   0.3007     0.449579    0.037576\n",
       "2  -0.2952   0.2952     0.443917    0.089761\n",
       "3  -0.2287   0.2287     0.867270    0.070946\n",
       "4  -0.3414   0.3414     0.249505    0.083246\n",
       "5  -0.2830   0.2830     0.591665    0.009072\n",
       "6  -0.3676   0.3676     0.205640    0.025455\n",
       "7  -0.3638   0.3638     0.197738    0.055672\n",
       "8  -0.3051   0.3051     0.395663    0.069871\n",
       "9  -0.4447   0.4447     0.077792    0.009049\n",
       "10 -0.3730   0.3730     0.180221    0.055874\n",
       "11 -0.4615   0.4615     0.051415    0.041071\n",
       "12 -0.4390   0.4390     0.063809    0.068425\n",
       "13 -0.3670   0.3670     0.212872    0.022872\n",
       "14 -0.3051   0.3051     0.408876    0.074098\n",
       "15 -0.3655   0.3655     0.198879    0.043764\n",
       "16 -0.4044   0.4044     0.125129    0.047278\n",
       "17 -0.2900   0.2900     0.491588    0.047734\n",
       "18 -0.3587   0.3587     0.238564    0.017035\n",
       "19 -0.3966   0.3966     0.152512    0.006852\n",
       "20 -0.4635   0.4635     0.017056    0.003419"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.DataFrame(trials.results)\n",
    "df[\"val_acc\"] = df[\"loss\"] * -1\n",
    "df = df.drop(['status', 'eval_time'], 1)\n",
    "df.to_csv(\"assets/two_layer/two_layer_hyperopt.csv\")\n",
    "\n",
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "valued-peace",
   "metadata": {},
   "source": [
    "Having found the best hyperparameter combination of the L2 regularization rate and the initial learning rate of the cyclical learning rate schedule, let's fit again the model, but this time with the combination of the training and the validation sets, and then evaluate on the test set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "standing-ottawa",
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train_val = np.vstack([x_train, x_val])\n",
    "y_train_val = np.hstack([y_train, y_val])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "constant-unknown",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "step size of cyc. lr: 400 update steps\n",
      "full cycle of cyc.lr : 800 update steps\n",
      "4.0 epochs = 1 full cycle = 800 update steps\n",
      "18 cycle = 72.0 epochs = 14400 update steps\n"
     ]
    }
   ],
   "source": [
    "batch_size = 100\n",
    "\n",
    "n_s = int(2*np.floor(x_train_val.shape[0] / batch_size))\n",
    "print(f\"step size of cyc. lr: {n_s} update steps\")\n",
    "\n",
    "cycle_steps = 2*n_s\n",
    "print(f\"full cycle of cyc.lr : {cycle_steps} update steps\")\n",
    "\n",
    "epochs_one_full_cycle = (cycle_steps * batch_size) / x_train_val.shape[0]\n",
    "print(f\"{epochs_one_full_cycle} epochs = 1 full cycle = {cycle_steps} update steps\")\n",
    "\n",
    "n_cycle = 18\n",
    "print(f\"{n_cycle} cycle = {n_cycle*epochs_one_full_cycle} epochs = {n_cycle*cycle_steps} update steps\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "informal-evidence",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "model summary: \n",
      "layer 0: dense: \n",
      "\t shape -- in: 3072, out: 50\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.018042^2), reg: l2 with 1.7056e-02\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: relu\n",
      "\n",
      "layer 1: dense: \n",
      "\t shape -- in: 50, out: 10\n",
      "\t w -- init: Xavier ~ 1.000000 x N(0.000000, 0.141421^2), reg: l2 with 1.7056e-02\n",
      "\t b -- init: Xavier ~ 1.000000 x N(0.000000, 1.000000^2)\n",
      "\t activation: softmax\n",
      "\n",
      "categorical cross-entropy loss with loss smoother constant\n",
      "sgd with cycling lr schedule and clipper who does nothing\n",
      "\n",
      "starting epoch: 1 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 78.73it/s]\n",
      "epoch 1/72 \n",
      " \t -- train loss = 1.7170605133922865 / train cost = 2.1842552357518734\n",
      "\t -- {\"accuracy_train\": 0.39075} \n",
      "\t -- val loss = 1.9971381466793243 / val cost = 2.4643328690389112\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 2 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 84.27it/s]\n",
      "epoch 2/72 \n",
      " \t -- train loss = 1.6386475903376376 / train cost = 2.014277029481502\n",
      "\t -- {\"accuracy_train\": 0.41965} \n",
      "\t -- val loss = 1.9120457341063892 / val cost = 2.2876751732502534\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 3 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 80.43it/s]\n",
      "epoch 3/72 \n",
      " \t -- train loss = 1.6874168893926345 / train cost = 1.9700276710164166\n",
      "\t -- {\"accuracy_train\": 0.41325} \n",
      "\t -- val loss = 2.417166835641563 / val cost = 2.699777617265345\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 4 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 78.60it/s]\n",
      "epoch 4/72 \n",
      " \t -- train loss = 1.696087475600928 / train cost = 1.911155945111678\n",
      "\t -- {\"accuracy_train\": 0.4081} \n",
      "\t -- val loss = 1.8202002140367126 / val cost = 2.0352686835474625\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 5 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 88.98it/s]\n",
      "epoch 5/72 \n",
      " \t -- train loss = 1.4476749268109605 / train cost = 1.6316785850811804\n",
      "\t -- {\"accuracy_train\": 0.4974} \n",
      "\t -- val loss = 1.2996106793095064 / val cost = 1.4836143375797262\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 6 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 89.59it/s]\n",
      "epoch 6/72 \n",
      " \t -- train loss = 1.3804856869054913 / train cost = 1.5560318554265475\n",
      "\t -- {\"accuracy_train\": 0.5221} \n",
      "\t -- val loss = 1.5939272849698747 / val cost = 1.7694734534909309\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 7 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 82.12it/s]\n",
      "epoch 7/72 \n",
      " \t -- train loss = 1.3283278083329242 / train cost = 1.5034137202312046\n",
      "\t -- {\"accuracy_train\": 0.5454} \n",
      "\t -- val loss = 1.2339459317018246 / val cost = 1.409031843600105\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 8 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 83.73it/s]\n",
      "epoch 8/72 \n",
      " \t -- train loss = 1.284388608428324 / train cost = 1.4613696926595476\n",
      "\t -- {\"accuracy_train\": 0.5622} \n",
      "\t -- val loss = 1.4473043291143033 / val cost = 1.624285413345527\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 9 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 74.41it/s]\n",
      "epoch 9/72 \n",
      " \t -- train loss = 1.3003149327777603 / train cost = 1.479511153395336\n",
      "\t -- {\"accuracy_train\": 0.55085} \n",
      "\t -- val loss = 1.3261438697342096 / val cost = 1.5053400903517853\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 10 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 77.27it/s]\n",
      "epoch 10/72 \n",
      " \t -- train loss = 1.3465158139491582 / train cost = 1.526374022277952\n",
      "\t -- {\"accuracy_train\": 0.5317} \n",
      "\t -- val loss = 1.623311579798632 / val cost = 1.8031697881274258\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 11 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 84.53it/s]\n",
      "epoch 11/72 \n",
      " \t -- train loss = 1.432196055181596 / train cost = 1.610524166946365\n",
      "\t -- {\"accuracy_train\": 0.5003} \n",
      "\t -- val loss = 1.923499796606503 / val cost = 2.101827908371272\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 12 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:03<00:00, 66.41it/s]\n",
      "epoch 12/72 \n",
      " \t -- train loss = 1.4410483524538753 / train cost = 1.6156611211272844\n",
      "\t -- {\"accuracy_train\": 0.48965} \n",
      "\t -- val loss = 1.284910473453846 / val cost = 1.459523242127255\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 13 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 71.22it/s]\n",
      "epoch 13/72 \n",
      " \t -- train loss = 1.4159519615129745 / train cost = 1.5921666089781699\n",
      "\t -- {\"accuracy_train\": 0.50525} \n",
      "\t -- val loss = 1.3055281068775906 / val cost = 1.481742754342786\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 14 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.93it/s]\n",
      "epoch 14/72 \n",
      " \t -- train loss = 1.3027147022952725 / train cost = 1.4846542575718336\n",
      "\t -- {\"accuracy_train\": 0.5447} \n",
      "\t -- val loss = 0.6418838521910529 / val cost = 0.823823407467614\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 15 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 67.36it/s]\n",
      "epoch 15/72 \n",
      " \t -- train loss = 1.252771655335529 / train cost = 1.4393224895515344\n",
      "\t -- {\"accuracy_train\": 0.568} \n",
      "\t -- val loss = 1.0927653587948767 / val cost = 1.279316193010882\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 16 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 67.00it/s]\n",
      "epoch 16/72 \n",
      " \t -- train loss = 1.2078047392256344 / train cost = 1.3973761450508777\n",
      "\t -- {\"accuracy_train\": 0.59235} \n",
      "\t -- val loss = 1.3295433315700538 / val cost = 1.5191147373952971\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 17 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 70.67it/s]\n",
      "epoch 17/72 \n",
      " \t -- train loss = 1.2224190899633034 / train cost = 1.4152386213870305\n",
      "\t -- {\"accuracy_train\": 0.58035} \n",
      "\t -- val loss = 1.0368086764287836 / val cost = 1.2296282078525107\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 18 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.98it/s]\n",
      "epoch 18/72 \n",
      " \t -- train loss = 1.3196679023474474 / train cost = 1.516896102078313\n",
      "\t -- {\"accuracy_train\": 0.53675} \n",
      "\t -- val loss = 2.4210377538498684 / val cost = 2.618265953580734\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 19 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 70.14it/s]\n",
      "epoch 19/72 \n",
      " \t -- train loss = 1.393899483755197 / train cost = 1.5889588237152386\n",
      "\t -- {\"accuracy_train\": 0.502} \n",
      "\t -- val loss = 1.296233013598766 / val cost = 1.4912923535588076\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 20 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 70.74it/s]\n",
      "epoch 20/72 \n",
      " \t -- train loss = 1.4419687346691223 / train cost = 1.6334387495121925\n",
      "\t -- {\"accuracy_train\": 0.4845} \n",
      "\t -- val loss = 1.8392340566633676 / val cost = 2.0307040715064377\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 21 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 67.95it/s]\n",
      "epoch 21/72 \n",
      " \t -- train loss = 1.3312424250599588 / train cost = 1.5210926401960467\n",
      "\t -- {\"accuracy_train\": 0.53565} \n",
      "\t -- val loss = 1.002200400725689 / val cost = 1.1920506158617767\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 22 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.26it/s]\n",
      "epoch 22/72 \n",
      " \t -- train loss = 1.2616688320418559 / train cost = 1.4575103245786032\n",
      "\t -- {\"accuracy_train\": 0.56755} \n",
      "\t -- val loss = 1.2751202593439455 / val cost = 1.4709617518806928\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 23 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.08it/s]\n",
      "epoch 23/72 \n",
      " \t -- train loss = 1.2205972526975202 / train cost = 1.4209277348722176\n",
      "\t -- {\"accuracy_train\": 0.58025} \n",
      "\t -- val loss = 1.3094381412337581 / val cost = 1.5097686234084555\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 24 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.14it/s]\n",
      "epoch 24/72 \n",
      " \t -- train loss = 1.1793224638478623 / train cost = 1.3827409609325865\n",
      "\t -- {\"accuracy_train\": 0.6003} \n",
      "\t -- val loss = 1.10638918427227 / val cost = 1.309807681356994\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 25 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:03<00:00, 66.45it/s]\n",
      "epoch 25/72 \n",
      " \t -- train loss = 1.192735806659562 / train cost = 1.3999001745516817\n",
      "\t -- {\"accuracy_train\": 0.59135} \n",
      "\t -- val loss = 1.0081653727568565 / val cost = 1.2153297406489763\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 26 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.34it/s]\n",
      "epoch 26/72 \n",
      " \t -- train loss = 1.3344156427746978 / train cost = 1.5442554094805665\n",
      "\t -- {\"accuracy_train\": 0.53035} \n",
      "\t -- val loss = 2.0089778106242493 / val cost = 2.218817577330118\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 27 ...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 71.37it/s]\n",
      "epoch 27/72 \n",
      " \t -- train loss = 1.307951260383667 / train cost = 1.5145627510949238\n",
      "\t -- {\"accuracy_train\": 0.541} \n",
      "\t -- val loss = 1.5873526988435895 / val cost = 1.7939641895548464\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 28 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 71.19it/s]\n",
      "epoch 28/72 \n",
      " \t -- train loss = 1.4812501541848146 / train cost = 1.6821233790986074\n",
      "\t -- {\"accuracy_train\": 0.47855} \n",
      "\t -- val loss = 1.496060304491451 / val cost = 1.6969335294052437\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 29 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 72.01it/s]\n",
      "epoch 29/72 \n",
      " \t -- train loss = 1.310379247360369 / train cost = 1.5103167123128298\n",
      "\t -- {\"accuracy_train\": 0.5423} \n",
      "\t -- val loss = 1.4307564495720286 / val cost = 1.6306939145244894\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 30 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 72.53it/s]\n",
      "epoch 30/72 \n",
      " \t -- train loss = 1.2694295040106016 / train cost = 1.4725999824840266\n",
      "\t -- {\"accuracy_train\": 0.5588} \n",
      "\t -- val loss = 1.62022250158867 / val cost = 1.823392980062095\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 31 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 69.86it/s]\n",
      "epoch 31/72 \n",
      " \t -- train loss = 1.1973863265246416 / train cost = 1.4051353702739189\n",
      "\t -- {\"accuracy_train\": 0.59465} \n",
      "\t -- val loss = 1.375883444941103 / val cost = 1.5836324886903803\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 32 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 69.92it/s]\n",
      "epoch 32/72 \n",
      " \t -- train loss = 1.163485199214415 / train cost = 1.37393265004706\n",
      "\t -- {\"accuracy_train\": 0.6092} \n",
      "\t -- val loss = 1.11733686326711 / val cost = 1.3277843140997547\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 33 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 69.39it/s]\n",
      "epoch 33/72 \n",
      " \t -- train loss = 1.1899012949558125 / train cost = 1.4036993892040166\n",
      "\t -- {\"accuracy_train\": 0.5934} \n",
      "\t -- val loss = 1.3800731918240614 / val cost = 1.5938712860722655\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 34 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.31it/s]\n",
      "epoch 34/72 \n",
      " \t -- train loss = 1.2290586935715806 / train cost = 1.4464386272337644\n",
      "\t -- {\"accuracy_train\": 0.57365} \n",
      "\t -- val loss = 1.4975248315431617 / val cost = 1.7149047652053455\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 35 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.92it/s]\n",
      "epoch 35/72 \n",
      " \t -- train loss = 1.341134987409775 / train cost = 1.5534611913968166\n",
      "\t -- {\"accuracy_train\": 0.5278} \n",
      "\t -- val loss = 1.0254525218678496 / val cost = 1.237778725854891\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 36 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.61it/s]\n",
      "epoch 36/72 \n",
      " \t -- train loss = 1.5044047830068183 / train cost = 1.7103714356935196\n",
      "\t -- {\"accuracy_train\": 0.47725} \n",
      "\t -- val loss = 2.060395937650181 / val cost = 2.266362590336882\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 37 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:03<00:00, 66.57it/s]\n",
      "epoch 37/72 \n",
      " \t -- train loss = 1.3194896950306814 / train cost = 1.5240135705946458\n",
      "\t -- {\"accuracy_train\": 0.5419} \n",
      "\t -- val loss = 1.377574404506539 / val cost = 1.5820982800705035\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 38 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.84it/s]\n",
      "epoch 38/72 \n",
      " \t -- train loss = 1.2744832065435556 / train cost = 1.4818564241279253\n",
      "\t -- {\"accuracy_train\": 0.5657} \n",
      "\t -- val loss = 0.8629653946203686 / val cost = 1.0703386122047382\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 39 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.22it/s]\n",
      "epoch 39/72 \n",
      " \t -- train loss = 1.1916219045646534 / train cost = 1.4039050536361148\n",
      "\t -- {\"accuracy_train\": 0.59} \n",
      "\t -- val loss = 0.7449786752073513 / val cost = 0.9572618242788127\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 40 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 67.92it/s]\n",
      "epoch 40/72 \n",
      " \t -- train loss = 1.1523874303379047 / train cost = 1.3675709017675335\n",
      "\t -- {\"accuracy_train\": 0.61505} \n",
      "\t -- val loss = 1.1444176481371469 / val cost = 1.3596011195667757\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 41 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 69.00it/s]\n",
      "epoch 41/72 \n",
      " \t -- train loss = 1.1772300247964347 / train cost = 1.395718727047116\n",
      "\t -- {\"accuracy_train\": 0.6003} \n",
      "\t -- val loss = 1.3186114502756792 / val cost = 1.5371001525263606\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 42 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 67.55it/s]\n",
      "epoch 42/72 \n",
      " \t -- train loss = 1.2246184368070918 / train cost = 1.4453158359859073\n",
      "\t -- {\"accuracy_train\": 0.57535} \n",
      "\t -- val loss = 0.9873521525690954 / val cost = 1.2080495517479108\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 43 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 67.28it/s]\n",
      "epoch 43/72 \n",
      " \t -- train loss = 1.4486481370217659 / train cost = 1.6659968955925535\n",
      "\t -- {\"accuracy_train\": 0.49695} \n",
      "\t -- val loss = 1.770389052619662 / val cost = 1.9877378111904496\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 44 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 69.77it/s]\n",
      "epoch 44/72 \n",
      " \t -- train loss = 1.4584482507951535 / train cost = 1.668356546564489\n",
      "\t -- {\"accuracy_train\": 0.48065} \n",
      "\t -- val loss = 0.6812926373257515 / val cost = 0.8912009330950869\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 45 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 66.81it/s]\n",
      "epoch 45/72 \n",
      " \t -- train loss = 1.3018563144938453 / train cost = 1.5113981578826396\n",
      "\t -- {\"accuracy_train\": 0.54585} \n",
      "\t -- val loss = 2.0873715471054384 / val cost = 2.2969133904942325\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 46 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 69.60it/s]\n",
      "epoch 46/72 \n",
      " \t -- train loss = 1.2543473904459108 / train cost = 1.4676523068541776\n",
      "\t -- {\"accuracy_train\": 0.56595} \n",
      "\t -- val loss = 1.3265587522454687 / val cost = 1.5398636686537355\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 47 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 67.71it/s]\n",
      "epoch 47/72 \n",
      " \t -- train loss = 1.1822952372102529 / train cost = 1.3995981631065253\n",
      "\t -- {\"accuracy_train\": 0.60095} \n",
      "\t -- val loss = 1.2504634099550014 / val cost = 1.467766335851274\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 48 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.74it/s]\n",
      "epoch 48/72 \n",
      " \t -- train loss = 1.1424284391859074 / train cost = 1.3622510052892143\n",
      "\t -- {\"accuracy_train\": 0.6176} \n",
      "\t -- val loss = 1.0849498179532615 / val cost = 1.3047723840565684\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 49 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:03<00:00, 66.51it/s]\n",
      "epoch 49/72 \n",
      " \t -- train loss = 1.16162631549544 / train cost = 1.3845317444460656\n",
      "\t -- {\"accuracy_train\": 0.60935} \n",
      "\t -- val loss = 0.9440927819050993 / val cost = 1.1669982108557249\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 50 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:03<00:00, 66.33it/s]\n",
      "epoch 50/72 \n",
      " \t -- train loss = 1.189115414782414 / train cost = 1.4149641509185011\n",
      "\t -- {\"accuracy_train\": 0.59305} \n",
      "\t -- val loss = 1.5326811210202456 / val cost = 1.7585298571563328\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 51 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.61it/s]\n",
      "epoch 51/72 \n",
      " \t -- train loss = 1.387653866098671 / train cost = 1.6088855046755652\n",
      "\t -- {\"accuracy_train\": 0.5183} \n",
      "\t -- val loss = 1.721644049615744 / val cost = 1.9428756881926381\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 52 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 69.37it/s]\n",
      "epoch 52/72 \n",
      " \t -- train loss = 1.4027107101812275 / train cost = 1.6184358988279566\n",
      "\t -- {\"accuracy_train\": 0.5126} \n",
      "\t -- val loss = 0.9055043006676045 / val cost = 1.1212294893143335\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 53 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 69.71it/s]\n",
      "epoch 53/72 \n",
      " \t -- train loss = 1.2792119008325962 / train cost = 1.4917185246968263\n",
      "\t -- {\"accuracy_train\": 0.5528} \n",
      "\t -- val loss = 1.378853552984806 / val cost = 1.5913601768490362\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 54 ...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 73.05it/s]\n",
      "epoch 54/72 \n",
      " \t -- train loss = 1.2708960758758334 / train cost = 1.4872538735717185\n",
      "\t -- {\"accuracy_train\": 0.5537} \n",
      "\t -- val loss = 0.8859095951659021 / val cost = 1.102267392861787\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 55 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 72.74it/s]\n",
      "epoch 55/72 \n",
      " \t -- train loss = 1.1898251903076755 / train cost = 1.4109224967819396\n",
      "\t -- {\"accuracy_train\": 0.58945} \n",
      "\t -- val loss = 0.674790622899452 / val cost = 0.8958879293737161\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 56 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 72.89it/s]\n",
      "epoch 56/72 \n",
      " \t -- train loss = 1.13786071217578 / train cost = 1.36126525980595\n",
      "\t -- {\"accuracy_train\": 0.6189} \n",
      "\t -- val loss = 1.0036918695154484 / val cost = 1.2270964171456185\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 57 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 70.15it/s]\n",
      "epoch 57/72 \n",
      " \t -- train loss = 1.1653880814109239 / train cost = 1.3915523299082369\n",
      "\t -- {\"accuracy_train\": 0.60515} \n",
      "\t -- val loss = 1.3652933768558448 / val cost = 1.5914576253531578\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 58 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 68.71it/s]\n",
      "epoch 58/72 \n",
      " \t -- train loss = 1.2441667083196604 / train cost = 1.4725071825152298\n",
      "\t -- {\"accuracy_train\": 0.568} \n",
      "\t -- val loss = 1.2242601858023017 / val cost = 1.4526006599978711\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 59 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 71.12it/s]\n",
      "epoch 59/72 \n",
      " \t -- train loss = 1.2773397309091483 / train cost = 1.5010580777587625\n",
      "\t -- {\"accuracy_train\": 0.5511} \n",
      "\t -- val loss = 1.738844440846683 / val cost = 1.9625627876962972\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 60 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 71.69it/s]\n",
      "epoch 60/72 \n",
      " \t -- train loss = 1.441064150835044 / train cost = 1.6575029178774632\n",
      "\t -- {\"accuracy_train\": 0.48615} \n",
      "\t -- val loss = 1.6530458110179072 / val cost = 1.8694845780603264\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 61 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 70.26it/s]\n",
      "epoch 61/72 \n",
      " \t -- train loss = 1.2671004842143863 / train cost = 1.4821097186323735\n",
      "\t -- {\"accuracy_train\": 0.5615} \n",
      "\t -- val loss = 1.0821221016247229 / val cost = 1.2971313360427101\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 62 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 73.36it/s]\n",
      "epoch 62/72 \n",
      " \t -- train loss = 1.2183219789657254 / train cost = 1.4362952699257288\n",
      "\t -- {\"accuracy_train\": 0.58785} \n",
      "\t -- val loss = 1.0630822791433863 / val cost = 1.2810555701033897\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 63 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 71.21it/s]\n",
      "epoch 63/72 \n",
      " \t -- train loss = 1.17113157228411 / train cost = 1.393421468928929\n",
      "\t -- {\"accuracy_train\": 0.60525} \n",
      "\t -- val loss = 1.4005422542021266 / val cost = 1.6228321508469457\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 64 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 71.89it/s]\n",
      "epoch 64/72 \n",
      " \t -- train loss = 1.133914900618784 / train cost = 1.3590409351732253\n",
      "\t -- {\"accuracy_train\": 0.6183} \n",
      "\t -- val loss = 1.1372277113244065 / val cost = 1.3623537458788477\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 65 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 71.64it/s]\n",
      "epoch 65/72 \n",
      " \t -- train loss = 1.1494328316384483 / train cost = 1.3774924343077841\n",
      "\t -- {\"accuracy_train\": 0.6102} \n",
      "\t -- val loss = 0.8939665729922417 / val cost = 1.1220261756615773\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 66 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 71.03it/s]\n",
      "epoch 66/72 \n",
      " \t -- train loss = 1.2278018958368206 / train cost = 1.4577068780253666\n",
      "\t -- {\"accuracy_train\": 0.5757} \n",
      "\t -- val loss = 1.306662394510412 / val cost = 1.536567376698958\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 67 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 71.71it/s]\n",
      "epoch 67/72 \n",
      " \t -- train loss = 1.269458792147238 / train cost = 1.4958167859986586\n",
      "\t -- {\"accuracy_train\": 0.5561} \n",
      "\t -- val loss = 1.178852354456917 / val cost = 1.4052103483083376\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 68 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 72.85it/s]\n",
      "epoch 68/72 \n",
      " \t -- train loss = 1.3036176688442178 / train cost = 1.5222574838761538\n",
      "\t -- {\"accuracy_train\": 0.5428} \n",
      "\t -- val loss = 1.3243152622485472 / val cost = 1.5429550772804832\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 69 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 72.50it/s]\n",
      "epoch 69/72 \n",
      " \t -- train loss = 1.261634799976203 / train cost = 1.4785628510120672\n",
      "\t -- {\"accuracy_train\": 0.5632} \n",
      "\t -- val loss = 2.072387474472796 / val cost = 2.2893155255086604\n",
      "\t -- {\"accuracy_val\": 0.0} \n",
      "\n",
      "\n",
      "starting epoch: 70 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 73.42it/s]\n",
      "epoch 70/72 \n",
      " \t -- train loss = 1.2240203398841034 / train cost = 1.4433770828384702\n",
      "\t -- {\"accuracy_train\": 0.5783} \n",
      "\t -- val loss = 1.5162646995747666 / val cost = 1.7356214425291334\n",
      "\t -- {\"accuracy_val\": 0.5} \n",
      "\n",
      "\n",
      "starting epoch: 71 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 72.42it/s]\n",
      "epoch 71/72 \n",
      " \t -- train loss = 1.1644147930171056 / train cost = 1.3879688051385461\n",
      "\t -- {\"accuracy_train\": 0.6019} \n",
      "\t -- val loss = 0.8282245233955772 / val cost = 1.0517785355170177\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "starting epoch: 72 ...\n",
      "batch 200/200: 100%|██████████| 200/200 [00:02<00:00, 70.28it/s]\n",
      "epoch 72/72 \n",
      " \t -- train loss = 1.1290914200855473 / train cost = 1.35550883848091\n",
      "\t -- {\"accuracy_train\": 0.62295} \n",
      "\t -- val loss = 0.993250925102908 / val cost = 1.2196683434982707\n",
      "\t -- {\"accuracy_val\": 1.0} \n",
      "\n",
      "\n",
      "The test metrics are: {'accuracy': 0.508}\n"
     ]
    }
   ],
   "source": [
    "reg_rate_l2 = best_trial[\"reg_rate_l2\"]\n",
    "lr_initial = best_trial[\"lr_initial\"]\n",
    "\n",
    "params = {\"coeff\": 1.0, \"mean\": 0.0, \"std\":None}\n",
    "\n",
    "in_dim = x_train_val.shape[1]\n",
    "out_dim = np.unique(y_train_val).size\n",
    "mid_dim = 50\n",
    "\n",
    "seed = 200\n",
    "\n",
    "dense_1 = \\\n",
    "    Dense(in_dim=in_dim, out_dim=mid_dim, \n",
    "          kernel_initializer=XavierInitializer(seed=seed, **params), \n",
    "          bias_initializer=XavierInitializer(seed=seed+1, **params), \n",
    "          kernel_regularizer=L2Regularizer(reg_rate=reg_rate_l2), \n",
    "          activation=ReLUActivation()\n",
    "         )\n",
    "\n",
    "dense_2 = \\\n",
    "    Dense(in_dim=mid_dim, out_dim=out_dim,\n",
    "          kernel_initializer=XavierInitializer(seed=seed+2, **params), \n",
    "          bias_initializer=XavierInitializer(seed=seed+3, **params), \n",
    "          kernel_regularizer=L2Regularizer(reg_rate=reg_rate_l2), \n",
    "          activation=SoftmaxActivation()\n",
    "         )\n",
    "\n",
    "layers = [\n",
    "    dense_1,\n",
    "    dense_2\n",
    "]\n",
    "\n",
    "model = Model(layers)\n",
    "\n",
    "loss = CategoricalCrossEntropyLoss(loss_smoother=LossSmootherConstant())\n",
    "\n",
    "# train longer\n",
    "n_epochs = 72\n",
    "batch_size = 100\n",
    "\n",
    "lr_max = 1e-1\n",
    "\n",
    "# as defined from before\n",
    "step_size = 800\n",
    "lr_schedule = LRCyclingSchedule(lr_initial, lr_max, step_size)\n",
    "optimizer = SGDOptimizer(lr_schedule=lr_schedule, grad_clipper=GradClipperByNothing())\n",
    "\n",
    "metrics = [AccuracyMetrics()]\n",
    "\n",
    "model.compile_model(optimizer, loss, metrics)\n",
    "print(model)\n",
    "\n",
    "# verbosity level of fit\n",
    "verbose = 2\n",
    "# x_val y_val here dont matter\n",
    "history = model.fit(x_train_val, y_train_val, x_val[:2], y_val[:2], n_epochs, batch_size, verbose, aug_func=None)\n",
    "\n",
    "params_test = {\"mode\": \"test\"}\n",
    "scores_test = model.forward(x_test, **params_test)\n",
    "y_hat_test = np.argmax(scores_test, axis=1)\n",
    "metrics_test = model.compute_metrics(y_test, scores_test)\n",
    "\n",
    "print(f\"The test metrics are: {metrics_test}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "necessary-professor",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_lrs(history, None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "straight-tragedy",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}