data_augmentation.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:34.316915100Z",
     "start_time": "2023-11-07T19:02:34.238362500Z"
    }
   },
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "Data augmentation for aphasic text generation\n",
    "authors:\n",
    "    Damiano Pasquini damiano23@ru.is\n",
    "    Andrea Terenziani andrea23@ru.is\n",
    "    Aurora Ayn Wickstrom aurora23@ru.is\n",
    "course: T-725-MALV, Málvinnsla\n",
    "\"\"\"\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import numpy.random as rnd\n",
    "import matplotlib.pyplot as plt\n",
    "import string\n",
    "import nltk\n",
    "from nltk.corpus import stopwords, gutenberg"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:34.532676500Z",
     "start_time": "2023-11-07T19:02:34.254002600Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "nltk.download('gutenberg', quiet=True)\n",
    "nltk.download('punkt', quiet=True)\n",
    "nltk.download('stopwords', quiet=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:34.532676500Z",
     "start_time": "2023-11-07T19:02:34.527446400Z"
    }
   },
   "outputs": [],
   "source": [
    "random_state = 111111\n",
    "rnd.seed(random_state)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:34.543443600Z",
     "start_time": "2023-11-07T19:02:34.540684900Z"
    }
   },
   "outputs": [],
   "source": [
    "def histplot(df: pd.DataFrame, bins=60, column=None, xlabel=None, ylabel=\"Frequency\", title=None, **plot_args):\n",
    "    \"\"\"\n",
    "    Plot a histogram of the given column of the given dataframe\n",
    "    \"\"\"\n",
    "    if column is None:\n",
    "        column = df.columns[0]\n",
    "\n",
    "    if xlabel is None:\n",
    "        xlabel = column\n",
    "\n",
    "    if title is None:\n",
    "        title = f'Distribution of {column}'\n",
    "\n",
    "    # Create a histogram to visualize the distribution of sentence lengths\n",
    "    plt.hist(df[column], bins=bins, **plot_args)\n",
    "    plt.xlabel(xlabel)\n",
    "    plt.ylabel(ylabel)\n",
    "    plt.title(title)\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:34.635680700Z",
     "start_time": "2023-11-07T19:02:34.543443600Z"
    }
   },
   "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>sentences</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>28806</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>unique</th>\n",
       "      <td>28806</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>top</th>\n",
       "      <td>Children who spend more time outdoors have a l...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>freq</th>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                sentences\n",
       "count                                               28806\n",
       "unique                                              28806\n",
       "top     Children who spend more time outdoors have a l...\n",
       "freq                                                    1"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.read_csv(\"eng_sents.csv\", names=[\"sentences\"])\n",
    "df.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preprocessing"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The longest (by number of words) sentences are unrealistic so we do a two-step filter:\n",
    "\n",
    "1. Remove sentences longer than some fixed threshold $L_{max} := 400$ \n",
    "2. Of those remaining, keep only those who are in the $p := 0.8$ fraction closest to the median\n",
    "3. Of those remaining, keep a randomly selected subset of size $s := 4000$"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:34.653913300Z",
     "start_time": "2023-11-07T19:02:34.635680700Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    28806.000000\n",
       "mean        12.917830\n",
       "std         10.673157\n",
       "min          1.000000\n",
       "25%          6.000000\n",
       "50%          9.000000\n",
       "75%         17.000000\n",
       "max        146.000000\n",
       "Name: sent-len, dtype: float64"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df[\"sent-len\"] = df[\"sentences\"].apply(lambda s : len(s.split()))\n",
    "df[\"sent-len\"].describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:34.870537200Z",
     "start_time": "2023-11-07T19:02:34.653913300Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "histplot(df, column=\"sent-len\", xlabel=\"Length\", edgecolor=\"k\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:34.870537200Z",
     "start_time": "2023-11-07T19:02:34.855805800Z"
    }
   },
   "outputs": [],
   "source": [
    "Lmin = 2\n",
    "Lmax = 30\n",
    "df1 = df[df[\"sent-len\"].between(Lmin, Lmax, inclusive=\"both\")]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.058401300Z",
     "start_time": "2023-11-07T19:02:34.870537200Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    26745.000000\n",
       "mean        10.817798\n",
       "std          7.039617\n",
       "min          2.000000\n",
       "25%          5.000000\n",
       "50%          8.000000\n",
       "75%         15.000000\n",
       "max         30.000000\n",
       "Name: sent-len, dtype: float64"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df1[\"sent-len\"].describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.074135Z",
     "start_time": "2023-11-07T19:02:34.964749400Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "histplot(df1, bins=100, column=\"sent-len\", xlabel=\"Length\", edgecolor=\"k\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.105788800Z",
     "start_time": "2023-11-07T19:02:35.074135Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\pasqu\\AppData\\Local\\Temp\\ipykernel_23708\\4082126998.py:6: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  df1['len-diff'] = abs(df1['sent-len'] - med_len)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "7580"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "p=1.0 # previously 0.8\n",
    "total_left = len(df1)\n",
    "to_extract = int(p * total_left)\n",
    "\n",
    "med_len = df1[\"sent-len\"].median()\n",
    "df1['len-diff'] = abs(df1['sent-len'] - med_len)\n",
    "df2 = df1.sort_values(by=\"len-diff\").head(n=to_extract)\n",
    "\n",
    "len(df2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.169317900Z",
     "start_time": "2023-11-07T19:02:35.085078500Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "count    7580.000000\n",
       "mean       22.861873\n",
       "std         4.953943\n",
       "min         3.000000\n",
       "25%        19.000000\n",
       "50%        23.000000\n",
       "75%        27.000000\n",
       "max        30.000000\n",
       "Name: sent-len, dtype: float64"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df2[\"sent-len\"].describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.285098500Z",
     "start_time": "2023-11-07T19:02:35.093262200Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "histplot(df2, bins=100, column=\"sent-len\", xlabel=\"Length\", edgecolor=\"k\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.285098500Z",
     "start_time": "2023-11-07T19:02:35.267336900Z"
    }
   },
   "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>sentences</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>4000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>unique</th>\n",
       "      <td>4000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>top</th>\n",
       "      <td>I may need to see the booty.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>freq</th>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                           sentences\n",
       "count                           4000\n",
       "unique                          4000\n",
       "top     I may need to see the booty.\n",
       "freq                               1"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = df2[\"sentences\"].sample(n=4000, random_state=random_state).to_frame()\n",
    "# data = df2[\"sentences\"].sample(n=4000, random_state=random_state).to_frame()\n",
    "data.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.285098500Z",
     "start_time": "2023-11-07T19:02:35.273363900Z"
    }
   },
   "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>sentences</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>16610</th>\n",
       "      <td>I may need to see the booty.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7473</th>\n",
       "      <td>Tom isn't Australian.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15390</th>\n",
       "      <td>I cry every time.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1708</th>\n",
       "      <td>He's probably exaggerating.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17674</th>\n",
       "      <td>Give me your credit card.</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                          sentences\n",
       "16610  I may need to see the booty.\n",
       "7473          Tom isn't Australian.\n",
       "15390             I cry every time.\n",
       "1708    He's probably exaggerating.\n",
       "17674     Give me your credit card."
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## (pseudo-)Aphasic text generation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### N-gram model\n",
    "\n",
    "Creation of an n-gram character LM from text for word variant scoring"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.285098500Z",
     "start_time": "2023-11-07T19:02:35.283410500Z"
    }
   },
   "outputs": [],
   "source": [
    "src_title = \"melville-moby_dick.txt\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.622884500Z",
     "start_time": "2023-11-07T19:02:35.285098500Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['[', 'M', 'o', 'b', 'y', 'D', 'i', 'c', 'k', 'b', 'y', 'H', 'e', 'r', 'm', 'a', 'n', 'M', 'e', 'l', 'v', 'i', 'l', 'l', 'e', '1', '8', '5', '1', ']']\n",
      "['E', 'T', 'Y', 'M', 'O', 'L', 'O', 'G', 'Y', '.']\n",
      "['(', 'S', 'u', 'p', 'p', 'l', 'i', 'e', 'd', 'b', 'y', 'a', 'L', 'a', 't', 'e', 'C', 'o', 'n', 's', 'u', 'm', 'p', 't', 'i', 'v', 'e', 'U', 's', 'h', 'e', 'r', 't', 'o', 'a', 'G', 'r', 'a', 'm', 'm', 'a', 'r', 'S', 'c', 'h', 'o', 'o', 'l', ')']\n",
      "['T', 'h', 'e', 'p', 'a', 'l', 'e', 'U', 's', 'h', 'e', 'r', '-', '-', 't', 'h', 'r', 'e', 'a', 'd', 'b', 'a', 'r', 'e', 'i', 'n', 'c', 'o', 'a', 't', ',', 'h', 'e', 'a', 'r', 't', ',', 'b', 'o', 'd', 'y', ',', 'a', 'n', 'd', 'b', 'r', 'a', 'i', 'n', ';', 'I', 's', 'e', 'e', 'h', 'i', 'm', 'n', 'o', 'w', '.']\n",
      "['H', 'e', 'w', 'a', 's', 'e', 'v', 'e', 'r', 'd', 'u', 's', 't', 'i', 'n', 'g', 'h', 'i', 's', 'o', 'l', 'd', 'l', 'e', 'x', 'i', 'c', 'o', 'n', 's', 'a', 'n', 'd', 'g', 'r', 'a', 'm', 'm', 'a', 'r', 's', ',', 'w', 'i', 't', 'h', 'a', 'q', 'u', 'e', 'e', 'r', 'h', 'a', 'n', 'd', 'k', 'e', 'r', 'c', 'h', 'i', 'e', 'f', ',', 'm', 'o', 'c', 'k', 'i', 'n', 'g', 'l', 'y', 'e', 'm', 'b', 'e', 'l', 'l', 'i', 's', 'h', 'e', 'd', 'w', 'i', 't', 'h', 'a', 'l', 'l', 't', 'h', 'e', 'g', 'a', 'y', 'f', 'l', 'a', 'g', 's', 'o', 'f', 'a', 'l', 'l', 't', 'h', 'e', 'k', 'n', 'o', 'w', 'n', 'n', 'a', 't', 'i', 'o', 'n', 's', 'o', 'f', 't', 'h', 'e', 'w', 'o', 'r', 'l', 'd', '.']\n",
      "['H', 'e', 'l', 'o', 'v', 'e', 'd', 't', 'o', 'd', 'u', 's', 't', 'h', 'i', 's', 'o', 'l', 'd', 'g', 'r', 'a', 'm', 'm', 'a', 'r', 's', ';', 'i', 't', 's', 'o', 'm', 'e', 'h', 'o', 'w', 'm', 'i', 'l', 'd', 'l', 'y', 'r', 'e', 'm', 'i', 'n', 'd', 'e', 'd', 'h', 'i', 'm', 'o', 'f', 'h', 'i', 's', 'm', 'o', 'r', 't', 'a', 'l', 'i', 't', 'y', '.']\n",
      "['\"', 'W', 'h', 'i', 'l', 'e', 'y', 'o', 'u', 't', 'a', 'k', 'e', 'i', 'n', 'h', 'a', 'n', 'd', 't', 'o', 's', 'c', 'h', 'o', 'o', 'l', 'o', 't', 'h', 'e', 'r', 's', ',', 'a', 'n', 'd', 't', 'o', 't', 'e', 'a', 'c', 'h', 't', 'h', 'e', 'm', 'b', 'y', 'w', 'h', 'a', 't', 'n', 'a', 'm', 'e', 'a', 'w', 'h', 'a', 'l', 'e', '-', 'f', 'i', 's', 'h', 'i', 's', 't', 'o', 'b', 'e', 'c', 'a', 'l', 'l', 'e', 'd', 'i', 'n', 'o', 'u', 'r', 't', 'o', 'n', 'g', 'u', 'e', 'l', 'e', 'a', 'v', 'i', 'n', 'g', 'o', 'u', 't', ',', 't', 'h', 'r', 'o', 'u', 'g', 'h', 'i', 'g', 'n', 'o', 'r', 'a', 'n', 'c', 'e', ',', 't', 'h', 'e', 'l', 'e', 't', 't', 'e', 'r', 'H', ',', 'w', 'h', 'i', 'c', 'h', 'a', 'l', 'm', 'o', 's', 't', 'a', 'l', 'o', 'n', 'e', 'm', 'a', 'k', 'e', 't', 'h', 't', 'h', 'e', 's', 'i', 'g', 'n', 'i', 'f', 'i', 'c', 'a', 't', 'i', 'o', 'n', 'o', 'f', 't', 'h', 'e', 'w', 'o', 'r', 'd', ',', 'y', 'o', 'u', 'd', 'e', 'l', 'i', 'v', 'e', 'r', 't', 'h', 'a', 't', 'w', 'h', 'i', 'c', 'h', 'i', 's', 'n', 'o', 't', 't', 'r', 'u', 'e', '.', '\"']\n",
      "['-', '-', 'H', 'A', 'C', 'K', 'L', 'U', 'Y', 'T']\n",
      "['\"', 'W', 'H', 'A', 'L', 'E', '.']\n",
      "['.', '.', '.', 'S', 'w', '.', 'a', 'n', 'd', 'D', 'a', 'n', '.']\n"
     ]
    }
   ],
   "source": [
    "source_sents = gutenberg.sents(src_title)\n",
    "source_sents = [[c for w in s for c in w] for s in source_sents]\n",
    "\n",
    "print(*source_sents[:10], sep=\"\\n\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.622884500Z",
     "start_time": "2023-11-07T19:02:35.618212300Z"
    }
   },
   "outputs": [],
   "source": [
    "# Create an N-gram model of the source text\n",
    "from nltk.lm.preprocessing import padded_everygram_pipeline\n",
    "\n",
    "n_gram_size = 4\n",
    "\n",
    "train, vocab = padded_everygram_pipeline(n_gram_size, source_sents)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.632554100Z",
     "start_time": "2023-11-07T19:02:35.622884500Z"
    }
   },
   "outputs": [],
   "source": [
    "from nltk.lm.smoothing import Smoothing\n",
    "\n",
    "#TODO: Actually implement Katz smoothing (somehow)\n",
    "class KatzSmoothing(Smoothing):\n",
    "    def __init__(self, vocabulary, count, gamma=0.5):\n",
    "        super().__init__(vocabulary, count)\n",
    "        self.gamma = gamma\n",
    "\n",
    "    def unigram_score(self, word, context=None):\n",
    "        return self.gamma / self.counts[0]\n",
    "\n",
    "    def alpha_gamma(self, word, context):\n",
    "        return self.gamma\n",
    "\n",
    "    def discount(self):\n",
    "        return self.gamma\n",
    "\n",
    "    def __str__(self):\n",
    "        return f\"KatzSmoothing(gamma={self.gamma})\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:35.679798200Z",
     "start_time": "2023-11-07T19:02:35.629046100Z"
    }
   },
   "outputs": [],
   "source": [
    "from nltk.lm import MLE\n",
    "\n",
    "mle = MLE(n_gram_size)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:47.464627500Z",
     "start_time": "2023-11-07T19:02:35.655120Z"
    }
   },
   "outputs": [],
   "source": [
    "mle.fit(text=train, vocabulary_text=vocab)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:47.471264400Z",
     "start_time": "2023-11-07T19:02:47.466636200Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['c', 'a', 'l', 'l', 'd']"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mle.generate(5)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Data generation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:47.482298500Z",
     "start_time": "2023-11-07T19:02:47.471264400Z"
    }
   },
   "outputs": [],
   "source": [
    "def insert_rand_char(word, charset=string.ascii_lowercase):\n",
    "    l = len(word)\n",
    "    idx = rnd.randint(l)\n",
    "    char_idx = rnd.randint(len(charset))\n",
    "    char = charset[char_idx]\n",
    "\n",
    "    return f\"{word[:idx]}{char}{word[idx:]}\"\n",
    "\n",
    "def delete_rand_char(word):\n",
    "    l = len(word)\n",
    "    idx = rnd.randint(l)\n",
    "\n",
    "    return f\"{word[:idx]}{word[idx+1:]}\"\n",
    "\n",
    "def swap_rand_chars(word):\n",
    "    word = word.removesuffix(\".\").removesuffix(\",\").removesuffix(\";\").removesuffix(\":\").removesuffix(\"!\").removesuffix(\"?\").removesuffix(\"'\").removesuffix('\"').removeprefix(\"'\").removeprefix('\"')\n",
    "    \n",
    "    l = len(word)\n",
    "    if l < 2:\n",
    "        return word\n",
    "\n",
    "    idx1, idx2 = sorted(rnd.choice(range(l), size=2, replace=False))\n",
    "    \n",
    "    c1, c2 = word[idx1], word[idx2]\n",
    "\n",
    "    return f\"{word[:idx1]}{c2}{word[idx1+1:idx2]}{c1}{word[idx2+1:]}\"\n",
    "\n",
    "def alter_word(word, n_changes=1):\n",
    "    changes = rnd.choice([\"I\", \"D\", \"S\"], size=n_changes)\n",
    "    iscapitalized = word[0].isupper()\n",
    "\n",
    "    output = word\n",
    "    for i in range(n_changes):\n",
    "        if changes[i] == \"I\":\n",
    "            output = insert_rand_char(output)\n",
    "        elif changes[i] == \"D\":\n",
    "            output = delete_rand_char(output)\n",
    "        elif changes[i] == \"S\":\n",
    "            output = swap_rand_chars(output)\n",
    "            \n",
    "    if iscapitalized:\n",
    "        output = output.capitalize()\n",
    "\n",
    "    return output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:47.492591700Z",
     "start_time": "2023-11-07T19:02:47.482298500Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'ciaon'"
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "alter_word(\"ciaone\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:47.541986100Z",
     "start_time": "2023-11-07T19:02:47.492591700Z"
    }
   },
   "outputs": [],
   "source": [
    "def score_variant(variant, context):\n",
    "    variant_chars = [c for c in variant]\n",
    "    context_chars = [c for c in context]\n",
    "\n",
    "    \"\"\"scores = [mle.logscore(vc, cc) for vc, cc in zip(variant_chars, context_chars)]\n",
    "\n",
    "    return sum(scores) / len(variant)\"\"\"\n",
    "\n",
    "    return mle.logscore(variant_chars, context_chars)\n",
    "\n",
    "def get_best_variant(word, context, var_count=10, max_iter=20):\n",
    "    variants = [\"\"] * var_count\n",
    "    scores = [0.] * var_count\n",
    "\n",
    "    for i in range(var_count):\n",
    "        for _ in range(max_iter):\n",
    "            new_var = alter_word(word)\n",
    "            if not new_var in variants:\n",
    "                variants[i] = new_var\n",
    "                scores[i] = score_variant(new_var, context)\n",
    "                break\n",
    "\n",
    "    selected = np.argmax(scores)\n",
    "\n",
    "    return variants[selected]\n",
    "\n",
    "def alter_sentence(sent, p=0.4, ng_size = n_gram_size, var_count=10):\n",
    "    output = \"\"\n",
    "    #TODO: Should probably tokenize better with nltk instead of just splitting on spaces\n",
    "    words = sent.split(\" \")\n",
    "    current_idx = 0\n",
    "\n",
    "    for idx, w in enumerate(words):\n",
    "        current_idx += len(w)\n",
    "        # Unless it's the last word, add a space\n",
    "        if idx < len(words) - 1:\n",
    "            current_idx += 1\n",
    "\n",
    "        w2 = w\n",
    "\n",
    "        if rnd.rand() < p:\n",
    "            # Pass as context the previous 2 characters in sent\n",
    "            start = max(0, current_idx - ng_size)\n",
    "            context = sent[start:current_idx]\n",
    "\n",
    "            w2 = get_best_variant(w, context=context, var_count=var_count)\n",
    "\n",
    "        output += f\"{w2} \"\n",
    "\n",
    "    return output.strip() # remove trailing whitespace"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:47.557611500Z",
     "start_time": "2023-11-07T19:02:47.499680600Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'how iqs it ggino lbig guy'"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "alter_sentence(\"how is it going big guy\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:47.557611500Z",
     "start_time": "2023-11-07T19:02:47.510250500Z"
    }
   },
   "outputs": [],
   "source": [
    "data_altered = data.copy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:49.847336600Z",
     "start_time": "2023-11-07T19:02:47.512328200Z"
    }
   },
   "outputs": [],
   "source": [
    "data_altered[\"sents-alt\"] = data_altered[\"sentences\"].apply(alter_sentence)\n",
    "with open('eng_sents_altered.txt', 'w', encoding='utf-8') as file:\n",
    "    for index, row in data_altered.iterrows():\n",
    "        file.write(f\"{row['sents-alt']}\\t{row['sentences']}\\n\")\n",
    "# # save to csv\n",
    "# data_altered.to_csv(\"eng_sents_altered.txt\", index=False, header=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2023-11-07T19:02:49.854180Z",
     "start_time": "2023-11-07T19:02:49.847336600Z"
    }
   },
   "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>sentences</th>\n",
       "      <th>sents-alt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>16610</th>\n",
       "      <td>I may need to see the booty.</td>\n",
       "      <td>I may tneed to see vthe booty.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7473</th>\n",
       "      <td>Tom isn't Australian.</td>\n",
       "      <td>To is't Australian.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15390</th>\n",
       "      <td>I cry every time.</td>\n",
       "      <td>I cry very time.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1708</th>\n",
       "      <td>He's probably exaggerating.</td>\n",
       "      <td>He's probably exaggerating.</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17674</th>\n",
       "      <td>Give me your credit card.</td>\n",
       "      <td>Lgive em your credgit card.</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                          sentences                       sents-alt\n",
       "16610  I may need to see the booty.  I may tneed to see vthe booty.\n",
       "7473          Tom isn't Australian.             To is't Australian.\n",
       "15390             I cry every time.                I cry very time.\n",
       "1708    He's probably exaggerating.     He's probably exaggerating.\n",
       "17674     Give me your credit card.     Lgive em your credgit card."
      ]
     },
     "execution_count": 72,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_altered.head()"
   ]
  }
 ],
 "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": "3.11.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}