Taxi/taxi_v3@QLearning#numpy.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Q-Learning"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Import"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import gym\n",
    "import numpy as np\n",
    "from time import sleep\n",
    "from IPython.display import clear_output"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Environment"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "env = gym.make('Taxi-v3').env"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Discrete(6)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "env.action_space"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Discrete(500)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "env.observation_space"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Random Agent"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+---------+\n",
      "|\u001b[35m\u001b[34;1m\u001b[43mR\u001b[0m\u001b[0m\u001b[0m: | : :G|\n",
      "| : | : : |\n",
      "| : : : : |\n",
      "| | : | : |\n",
      "|Y| : |B: |\n",
      "+---------+\n",
      "  (Dropoff)\n",
      "Steps taken: 3131, penalties incurred: 1065\n",
      "Wall time: 4.81 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "done = False\n",
    "steps, penalties = 0, 0\n",
    "while not done:\n",
    "    clear_output(wait=True)\n",
    "    state, reward, done, info = env.step(env.action_space.sample())\n",
    "    env.render()\n",
    "    if reward==-10:\n",
    "        penalties += 1\n",
    "    steps += 1\n",
    "print(f'Steps taken: {steps}, penalties incurred: {penalties}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Q-Learning"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "q_table = np.zeros([env.observation_space.n, env.action_space.n])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0., 0., 0., 0., 0., 0.],\n",
       "       [0., 0., 0., 0., 0., 0.],\n",
       "       [0., 0., 0., 0., 0., 0.],\n",
       "       ...,\n",
       "       [0., 0., 0., 0., 0., 0.],\n",
       "       [0., 0., 0., 0., 0., 0.],\n",
       "       [0., 0., 0., 0., 0., 0.]])"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "q_table"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Hyperparameters\n",
    "alpha = 0.1\n",
    "gamma = 0.6\n",
    "epsilon = 0.1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Episode: 1000\n",
      "Episode: 2000\n",
      "Episode: 3000\n",
      "Episode: 4000\n",
      "Episode: 5000\n",
      "Episode: 6000\n",
      "Episode: 7000\n",
      "Episode: 8000\n",
      "Episode: 9000\n",
      "Episode: 10000\n",
      "Wall time: 12.8 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "for i in range(1,10001):\n",
    "    state = env.reset()\n",
    "    done = False\n",
    "    while not done:\n",
    "        if np.random.uniform() < epsilon:\n",
    "            # exploration\n",
    "            action = env.action_space.sample()\n",
    "        else:\n",
    "            # exploitation\n",
    "            action = np.argmax(q_table[state])\n",
    "        next_state, reward, done, info = env.step(action)\n",
    "        old_value = q_table[state, action]\n",
    "        next_max = np.max(q_table[next_state])\n",
    "        new_value = (1-alpha)*old_value + alpha*(reward+gamma*next_max)\n",
    "        q_table[state, action] = new_value\n",
    "        state = next_state\n",
    "    if i%1000==0:\n",
    "        print(f'Episode: {i}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.        ,  0.        ,  0.        ,  0.        ,  0.        ,\n",
       "         0.        ],\n",
       "       [-2.3930052 , -2.33616699, -2.37527098, -2.35314108, -2.27325184,\n",
       "        -9.16157496],\n",
       "       [-1.84909945, -1.49348636, -1.84927848, -1.50895688, -0.7504    ,\n",
       "        -8.90552491],\n",
       "       ...,\n",
       "       [-1.08831672, -0.02591116, -1.08831672, -1.11746961, -2.82266006,\n",
       "        -1.96      ],\n",
       "       [-2.08280307, -2.07093967, -2.08379147, -2.07584331, -2.88086489,\n",
       "        -3.62559153],\n",
       "       [-0.196     , -0.196     ,  0.7058308 , 10.85889867, -1.93574214,\n",
       "        -1.        ]])"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "q_table"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "+---------+\n",
      "|R: | : :G|\n",
      "| : | : : |\n",
      "| : : : : |\n",
      "| | : | : |\n",
      "|Y| : |\u001b[35m\u001b[34;1m\u001b[43mB\u001b[0m\u001b[0m\u001b[0m: |\n",
      "+---------+\n",
      "  (Dropoff)\n",
      "Steps taken: 16, penalties incurred: 0\n"
     ]
    }
   ],
   "source": [
    "state = env.reset()\n",
    "done = False\n",
    "steps, penalties = 0, 0\n",
    "while not done:\n",
    "    clear_output(wait=True)\n",
    "    action = np.argmax(q_table[state])\n",
    "    next_state, reward, done, info = env.step(action)\n",
    "    env.render()\n",
    "    if reward==-10:\n",
    "        penalties += 1\n",
    "    steps += 1\n",
    "    state = next_state\n",
    "    sleep(0.1)\n",
    "print(f'Steps taken: {steps}, penalties incurred: {penalties}')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Comment"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Because this environment has finite state space and finite action space, it could be solved very quick and easily by using a simple Q-table. For any environment having small finite state and action space, Q-Learning may be the most preferred way to try first."
   ]
  }
 ],
 "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": "3.8.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}