spam_detection.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import pandas as pd\n",
    "import unidecode\n",
    "import random\n",
    "import string\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "scrolled": true
   },
   "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>label</th>\n",
       "      <th>message</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>ham</td>\n",
       "      <td>Go until jurong point, crazy.. Available only ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>ham</td>\n",
       "      <td>Ok lar... Joking wif u oni...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>spam</td>\n",
       "      <td>Free entry in 2 a wkly comp to win FA Cup fina...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>ham</td>\n",
       "      <td>U dun say so early hor... U c already then say...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>ham</td>\n",
       "      <td>Nah I don't think he goes to usf, he lives aro...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  label                                            message\n",
       "0   ham  Go until jurong point, crazy.. Available only ...\n",
       "1   ham                      Ok lar... Joking wif u oni...\n",
       "2  spam  Free entry in 2 a wkly comp to win FA Cup fina...\n",
       "3   ham  U dun say so early hor... U c already then say...\n",
       "4   ham  Nah I don't think he goes to usf, he lives aro..."
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# load data\n",
    "data = pd.read_csv('archive/spam.csv', encoding='latin-1')\n",
    "data = data.iloc[:, :2]\n",
    "data.columns = [\"label\", \"message\"]\n",
    "data.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# converting to ascii\n",
    "data.iloc[:, 1] = data.iloc[:, 1].apply(lambda x: unidecode.unidecode(x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# ham = 0 and spam = 1\n",
    "data.iloc[:, 0] = data.iloc[:, 0].apply(lambda x: 0 if x == \"ham\" else 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "spams = data[data.iloc[:, 0] == 1]\n",
    "hams = data[data.iloc[:, 0] == 0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# creating train and test set\n",
    "n_test_examples = 50\n",
    "train_spams = spams.iloc[:-n_test_examples]\n",
    "train_hams = hams.iloc[:-n_test_examples]\n",
    "\n",
    "test_spams = spams.iloc[-n_test_examples:]\n",
    "test_hams = hams.iloc[-n_test_examples:]\n",
    "test_examples = pd.concat([test_spams, test_hams])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# charset for training\n",
    "charset = string.printable[:-6]\n",
    "charset = {char:index for index, char in enumerate(charset)}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def one_hot_char(char):\n",
    "    v = torch.zeros(1, len(charset))\n",
    "    v[0][charset[char]] = 1\n",
    "    return v\n",
    "\n",
    "def one_hot_word(word):\n",
    "    return torch.vstack([one_hot_char(char) for char in word if charset.get(char, None) is not None]).reshape((-1,1,len(charset)))\n",
    "\n",
    "def get_training_example():\n",
    "    random_class_index = random.randint(0, 1)\n",
    "    data = train_spams if random_class_index else train_hams\n",
    "    random_msg_index = random.randint(0, data.shape[0]-1)\n",
    "    return random_class_index, data.iloc[random_msg_index][1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "class RNN(nn.Module):\n",
    "    \n",
    "    def __init__(self, input_size_char_rnn, hidden_size_char_rnn, hidden_size_word_rnn, output_size_word_rnn):\n",
    "        super().__init__()\n",
    "        self.hidden_size_char_rnn = hidden_size_char_rnn\n",
    "        self.hidden_size_word_rnn = hidden_size_word_rnn\n",
    "        \n",
    "        self.char_rnn_i2h = nn.Linear(input_size_char_rnn + hidden_size_char_rnn, hidden_size_char_rnn)\n",
    "    \n",
    "        input_size_word_rnn = hidden_size_char_rnn\n",
    "        self.word_rnn_i2h = nn.Linear(input_size_word_rnn + hidden_size_word_rnn, hidden_size_word_rnn)\n",
    "        self.word_rnn_i2o = nn.Linear(input_size_word_rnn + hidden_size_word_rnn, output_size_word_rnn)\n",
    "        \n",
    "        self.softmax = nn.LogSoftmax(dim=1)\n",
    "    \n",
    "    def forward(self, input_tensor_word_rnn, hidden_tensor_word_rnn):\n",
    "        hidden_char_rnn = torch.zeros(1, self.hidden_size_char_rnn)\n",
    "        \n",
    "        for char in input_tensor_word_rnn:\n",
    "            hidden_char_rnn = self.forward_for_char_rnn(char, hidden_char_rnn)\n",
    "            \n",
    "        concat_input = torch.cat((hidden_char_rnn, hidden_tensor_word_rnn), dim=1)\n",
    "            \n",
    "        hidden_word_rnn = self.word_rnn_i2h(concat_input)\n",
    "        \n",
    "        output = self.word_rnn_i2o(concat_input)\n",
    "        output = self.softmax(output)\n",
    "        \n",
    "        return output, hidden_word_rnn\n",
    "    \n",
    "    \n",
    "    def forward_for_char_rnn(self, input_tensor, hidden_tensor):\n",
    "        concat_input = torch.cat((input_tensor, hidden_tensor), dim=1)\n",
    "\n",
    "        hidden = self.char_rnn_i2h(concat_input)\n",
    "\n",
    "        return hidden"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_size_char_rnn = len(charset) \n",
    "hidden_size_char_rnn = 256\n",
    "hidden_size_word_rnn = 384\n",
    "output_size_word_rnn = 2\n",
    "learning_rate = 0.0001"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "rnn = RNN(input_size_char_rnn, hidden_size_char_rnn, hidden_size_word_rnn, output_size_word_rnn)\n",
    "loss_fn = nn.NLLLoss()\n",
    "optimizer = torch.optim.Adam(rnn.parameters(), lr=learning_rate)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train(label, msg):\n",
    "\n",
    "    hidden_word_rnn = torch.zeros(1, hidden_size_word_rnn)\n",
    "\n",
    "    for word in msg.split(' '):\n",
    "        if word:\n",
    "            output, hidden_word_rnn = rnn(one_hot_word(word), hidden_word_rnn)\n",
    "\n",
    "    loss = loss_fn(output, torch.tensor([label]))\n",
    "\n",
    "    optimizer.zero_grad()\n",
    "    loss.backward()\n",
    "    optimizer.step()\n",
    "\n",
    "    return output, loss.item()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "def test():\n",
    "    n_total = test_examples.shape[0]\n",
    "    n_correct = 0\n",
    "    with torch.no_grad():\n",
    "        for i in range(n_total):\n",
    "            hidden_word_rnn = torch.zeros(1, hidden_size_word_rnn)\n",
    "            for word in test_examples.iloc[i][1].split(' '):\n",
    "                if word:\n",
    "                    output, hidden_word_rnn = rnn(one_hot_word(word), hidden_word_rnn)\n",
    "            predicted_class = torch.argmax(output).item()\n",
    "            if predicted_class == test_examples.iloc[i][0]:\n",
    "                n_correct += 1\n",
    "    accuracy = n_correct/n_total\n",
    "    print('Accuracy:', accuracy)\n",
    "    return accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.5\n",
      "Accuracy: 0.9\n",
      "1000 2.0 0.3857 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.87\n",
      "2000 4.0 0.2928 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.98\n",
      "3000 6.0 0.2120 | Original: 1 | Predicted: 0\n",
      "Accuracy: 0.9\n",
      "4000 8.0 0.1823 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.98\n",
      "5000 10.0 0.1103 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.93\n",
      "6000 12.0 0.1743 | Original: 1 | Predicted: 0\n",
      "Accuracy: 0.99\n",
      "7000 14.0 0.1218 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.93\n",
      "8000 16.0 0.1875 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.97\n",
      "9000 18.0 0.0852 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.99\n",
      "10000 20.0 0.1536 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.96\n",
      "11000 22.0 0.1196 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.93\n",
      "12000 24.0 0.1297 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.94\n",
      "13000 26.0 0.1111 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.94\n",
      "14000 28.0 0.0852 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.9\n",
      "15000 30.0 0.0966 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.92\n",
      "16000 32.0 0.0962 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.95\n",
      "17000 34.0 0.1076 | Original: 0 | Predicted: 1\n",
      "Accuracy: 0.98\n",
      "18000 36.0 0.0823 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.88\n",
      "19000 38.0 0.0867 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.93\n",
      "20000 40.0 0.0514 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.94\n",
      "21000 42.0 0.1611 | Original: 0 | Predicted: 1\n",
      "Accuracy: 0.95\n",
      "22000 44.0 0.0952 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.95\n",
      "23000 46.0 0.0868 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.95\n",
      "24000 48.0 0.0443 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.97\n",
      "25000 50.0 0.0373 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.94\n",
      "26000 52.0 0.0973 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.94\n",
      "27000 54.0 0.0749 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.96\n",
      "28000 56.0 0.0342 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.95\n",
      "29000 58.0 0.5550 | Original: 0 | Predicted: 1\n",
      "Accuracy: 0.97\n",
      "30000 60.0 0.0695 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.96\n",
      "31000 62.0 0.0722 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.95\n",
      "32000 64.0 0.0358 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.95\n",
      "33000 66.0 0.0441 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.98\n",
      "34000 68.0 0.0184 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.95\n",
      "35000 70.0 0.0223 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.95\n",
      "36000 72.0 0.0443 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.91\n",
      "37000 74.0 0.0259 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.96\n",
      "38000 76.0 0.0512 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.98\n",
      "39000 78.0 0.0093 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.97\n",
      "40000 80.0 0.0529 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.95\n",
      "41000 82.0 0.0965 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.97\n",
      "42000 84.0 0.0236 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.95\n",
      "43000 86.0 0.0253 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.94\n",
      "44000 88.0 0.0207 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.94\n",
      "45000 90.0 0.0357 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.94\n",
      "46000 92.0 0.0290 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.97\n",
      "47000 94.0 0.0183 | Original: 0 | Predicted: 0\n",
      "Accuracy: 0.95\n",
      "48000 96.0 0.0205 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.96\n",
      "49000 98.0 0.0164 | Original: 1 | Predicted: 1\n",
      "Accuracy: 0.94\n",
      "50000 100.0 0.0495 | Original: 0 | Predicted: 0\n"
     ]
    }
   ],
   "source": [
    "rolling_loss = 0\n",
    "all_loses = []\n",
    "accuracies = [test()]\n",
    "plot_steps = 200\n",
    "print_steps = 1000\n",
    "n_iters = 50000\n",
    "\n",
    "for i in range(n_iters):\n",
    "    print(i, end='\\r')\n",
    "    label, msg = get_training_example()\n",
    "    \n",
    "    output, loss = train(label, msg)\n",
    "        \n",
    "    rolling_loss = (1 - 1/plot_steps) * rolling_loss + (1/plot_steps) * loss\n",
    "    \n",
    "    if (i+1) % plot_steps == 0:\n",
    "        all_loses.append(rolling_loss)\n",
    "        \n",
    "    \n",
    "    if (i+1) % print_steps == 0:\n",
    "        accuracies.append(test())\n",
    "        predicted_class = torch.argmax(output).item()\n",
    "        print(f\"{i+1} {(i+1) * 100 / n_iters}% {rolling_loss:.4f} | Original: {label} | Predicted: {predicted_class}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "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": [
    "plt.figure()\n",
    "plt.plot(all_loses)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "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": [
    "plt.figure()\n",
    "plt.plot(accuracies)\n",
    "plt.show()"
   ]
  }
 ],
 "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": "3.9.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}