abstracts.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Problem description\n",
    "\n",
    "Data for this project consists of two tables in a tab-separated columns format. Each row in those files corresponds to an abstract of a scientific article from ACM Digital Library, which was assigned to one or more topics from the ACM Computing Classification System.\n",
    "\n",
    "The training data (DM2020_training_docs_and_labels.csv) has three columns: the first one is an identifier of a document, the second one stores the text of the abstract, and the third one contains a list of comma-separated topic labels.\n",
    "\n",
    "The test data (DM2020_test_docs.csv) has a similar format, but the labels in the third column are missing.\n",
    "\n",
    "The task and the format of submissions: the task for you is to predict the labels of documents from the test data and submit them to the evaluation system. A correctly formatted submission should be a text file with exactly 100000 lines. Each line should correspond to a document from the test data set (the order matters!) and contain a list of one or more predicted labels, separated by commas.\n",
    "\n",
    "Evaluation: the quality of submissions will be evaluated using the average F1-score measure, i.e., for each test document, the F1-score between the predicted and true labels will be computed, and the values obtained for all test cases will be averaged."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Solution\n",
    "\n",
    "## Loading data\n",
    "\n",
    "Let's load the training set.\n",
    "\n",
    "Our texts have multiple labels, so we will use one hot encoding."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import csv\n",
    "\n",
    "import pandas as pd\n",
    "from sklearn.preprocessing import MultiLabelBinarizer\n",
    "\n",
    "\n",
    "def load_training_data(filepath):\n",
    "    with open(filepath, 'rb') as f:\n",
    "        decoded_lines = (line.decode(encoding='iso_8859_1') for line in f)\n",
    "        parsed_lines = csv.reader(decoded_lines, dialect=\"excel-tab\")\n",
    "\n",
    "        names = []\n",
    "        texts = []\n",
    "        label_lists = []\n",
    "        \n",
    "        for name, text, labels in parsed_lines:\n",
    "            names.append(name)\n",
    "            texts.append(text)\n",
    "            \n",
    "            labels = labels.split(',')\n",
    "            label_lists.append(labels)\n",
    "    \n",
    "    mlb = MultiLabelBinarizer()\n",
    "    df_labels = pd.DataFrame(mlb.fit_transform(label_lists), columns=mlb.classes_)\n",
    "    \n",
    "    df = pd.DataFrame({'name': names, 'text': texts})\n",
    "    return pd.concat((df, df_labels), axis=1)\n",
    "\n",
    "\n",
    "df = load_training_data('data/DM2020_training_docs_and_labels.txt')\n",
    "labels = df.columns[2:]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Label distribution\n",
    "\n",
    "Firstly, let's look into label distribution. We can plot frequency of each label:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2160x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2160x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2160x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2160x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABrgAAAJfCAYAAAAkWNaXAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAgAElEQVR4nOzdb6xYdZ3n8c8XLtLVZQvFLjEUc5sMcRT/tNgAZjZGIfzRrsKDGYPZrI0QWRN2M5OsztQ1hKwOSSeb6I4aScjgDJiZVRY0kKkZbdAx8YFA+bOOyBg6TpUSldKC7gJ1RX/74B7Yq2npveWW22/7eiU395zf+Z1zf6dPePDmnFNjjAAAAAAAAEAXxy33AgAAAAAAAGAxBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWplZ7gW8kFe+8pVjdnZ2uZcBAAAAAADAS+zee+99fIyxen/HjujANTs7m+3bty/3MgAAAAAAAHiJVdUPD3TMKwoBAAAAAABoReACAAAAAACgFYELAAAAAACAVo7ob3ABAAAAAAB09ctf/jK7du3Kvn37lnspR7QVK1ZkzZo1OeGEExZ8jsAFAAAAAABwGOzatSsnnXRSZmdnU1XLvZwj0hgje/bsya5du7J27doFn+cVhQAAAAAAAIfBvn37cuqpp4pbL6Cqcuqppy76KTeBCwAAAAAA4DARtw7uUP6NBC4AAAAAAAAW5W1ve1u2b9++bH/fN7gAAAAAAABeArObty7p9XZu2bik13upPPvss5mZeXGJ6qBPcFXVa6rqgXk/P6+qP6qqVVW1raoenn6fMs2vqvpUVe2oqu9U1dnzrrVpmv9wVW16USsHAAAAAADgBe3cuTOvfe1r84EPfCBnnXVWLrroojzzzDO/8QTW448/ntnZ2STJX/3VX+Wyyy7LhRdemNnZ2XzmM5/JJz7xiaxfvz7nnXde9u7d+/y1P//5z2fdunV5/etfn7vvvjtJ8tRTT+WKK67IOeeck/Xr1+f2229//rrvfve7c/755+eCCy540fd10MA1xvj+GGPdGGNdkjcneTrJl5NsTnLnGOPMJHdO+0nyjiRnTj9XJbk+SapqVZJrk5yb5Jwk1z4XxQAAAAAAADg8Hn744Vx99dV58MEHc/LJJ+e22257wfnf/e5386UvfSn33HNPPvrRj+blL3957r///rzlLW/JzTff/Py8p59+Og888EA++9nP5oorrkiSXHfddTn//PNz99135xvf+EY+/OEP56mnnkqS3Hfffbn11lvzzW9+80Xf02K/wXVBkn8aY/wwyaVJbprGb0py2bR9aZKbx5xvJzm5ql6V5OIk28YYe8cYTyTZluSSF30HAAAAAAAAHNDatWuzbt26JMmb3/zm7Ny58wXnv/3tb89JJ52U1atXZ+XKlXnXu96VJHnDG97wG+e+973vTZK89a1vzc9//vM8+eST+drXvpYtW7Zk3bp1edvb3pZ9+/blRz/6UZLkwgsvzKpVq5bknhb7gsPLk/yPafu0McaPp+2fJDlt2j49ySPzztk1jR1oHAAAAAAAgMPkxBNPfH77+OOPzzPPPJOZmZn8+te/TpLs27fvgPOPO+645/ePO+64PPvss88fq6rfOK+qMsbIbbfdlte85jW/ceyuu+7KK17xiqW5oSziCa6qelmSdyf5n799bIwxkoylWFBVXVVV26tq++7du5fikgAAAAAAAMwzOzube++9N0ly6623HtI1vvjFLyZJvvWtb2XlypVZuXJlLr744nz605/OXDpK7r///qVZ8G9ZzCsK35HkvjHGT6f9n06vHsz0+7Fp/NEkZ8w7b800dqDx3zDGuGGMsWGMsWH16tWLWB4AAAAAAAAL8aEPfSjXX3991q9fn8cff/yQrrFixYqsX78+H/zgB3PjjTcmSa655pr88pe/zBvf+MacddZZueaaa5Zy2c+r5wraQSdWfSHJV8cYfznt/7cke8YYW6pqc5JVY4w/rqqNSf5jkncmOTfJp8YY51TVqiT3Jjl7uuR9Sd48xth7oL+5YcOGsX379kO9NwAAAAAAgGXz0EMP5bWvfe1yL6OF/f1bVdW9Y4wN+5u/oG9wVdUrklyY5D/MG96S5JaqujLJD5O8Zxr/Subi1o4kTyd5f5KMMfZW1ceT3DPN+9gLxS0AAAAAAADYnwUFrjHGU0lO/a2xPUku2M/ckeTqA1znc0k+t/hlAgAAAAAAwJzFfIMLAAAAAAAAlp3ABQAAAAAAcJjMvfiOF3Io/0YCFwAAAAAAwGGwYsWK7NmzR+R6AWOM7NmzJytWrFjUeQv6BhcAAAAAAACLs2bNmuzatSu7d+9e7qUc0VasWJE1a9Ys6hyBCwAAAAAA4DA44YQTsnbt2uVexlHJKwoBAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACglQUFrqo6uapurap/rKqHquotVbWqqrZV1cPT71OmuVVVn6qqHVX1nao6e951Nk3zH66qTYfrpgAAAAAAADh6LfQJrj9P8ndjjN9N8qYkDyXZnOTOMcaZSe6c9pPkHUnOnH6uSnJ9klTVqiTXJjk3yTlJrn0uigEAAAAAAMBCHTRwVdXKJG9NcmOSjDH+7xjjySSXJrlpmnZTksum7UuT3DzmfDvJyVX1qiQXJ9k2xtg7xngiybYklyzp3QAAAAAAAHDUW8gTXGuT7E7yl1V1f1X9RVW9IslpY4wfT3N+kuS0afv0JI/MO3/XNHag8d9QVVdV1faq2r579+7F3Q0AAAAAAABHvYUErpkkZye5foyxPslT+f+vI0ySjDFGkrEUCxpj3DDG2DDG2LB69eqluCQAAAAAAABHkYUErl1Jdo0x7pr2b81c8Prp9OrBTL8fm44/muSMeeevmcYONA4AAAAAAAALdtDANcb4SZJHquo109AFSb6X5I4km6axTUlun7bvSPK+mnNekp9NrzL8apKLquqUqjolyUXTGAAAAAAAACzYzALn/ackf11VL0vygyTvz1wcu6WqrkzywyTvmeZ+Jck7k+xI8vQ0N2OMvVX18ST3TPM+NsbYuyR3AQAAAAAAwDGj5j6fdWTasGHD2L59+3IvAwAAAAAAgJdYVd07xtiwv2ML+QYXAAAAAAAAHDEELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoJWZ5V4AAAAAAMem2c1bD+m8nVs2LvFKAIBuPMEFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsLClxVtbOq/qGqHqiq7dPYqqraVlUPT79Pmcarqj5VVTuq6jtVdfa862ya5j9cVZsOzy0BAAAAAABwNJtZxNy3jzEen7e/OcmdY4wtVbV52v+TJO9Icub0c26S65OcW1WrklybZEOSkeTeqrpjjPHEEtwHAAAAAAC85GY3bz2k83Zu2bjEK4Fjy4t5ReGlSW6atm9Kctm88ZvHnG8nObmqXpXk4iTbxhh7p6i1LcklL+LvAwAAAAAAcAxaaOAaSb5WVfdW1VXT2GljjB9P2z9Jctq0fXqSR+adu2saO9A4AAAAAAAALNhCX1H4b8YYj1bVv06yrar+cf7BMcaoqrEUC5oC2lVJ8upXv3opLgkAAAAAAMBRZEFPcI0xHp1+P5bky0nOSfLT6dWDmX4/Nk1/NMkZ805fM40daPy3/9YNY4wNY4wNq1evXtzdAAAAAAAAcNQ7aOCqqldU1UnPbSe5KMl3k9yRZNM0bVOS26ftO5K8r+acl+Rn06sMv5rkoqo6papOma7z1SW9GwAAAAAAAI56C3lF4WlJvlxVz83/mzHG31XVPUluqaork/wwyXum+V9J8s4kO5I8neT9STLG2FtVH09yzzTvY2OMvUt2JwAAAAAAABwTDhq4xhg/SPKm/YzvSXLBfsZHkqsPcK3PJfnc4pcJAAAAAAAAcxb0DS4AAAAAAAA4UghcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArAhcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArAhcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArAhcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArAhcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArAhcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANDKggNXVR1fVfdX1d9O+2ur6q6q2lFVX6yql03jJ077O6bjs/Ou8ZFp/PtVdfFS3wwAAAAAAABHv8U8wfWHSR6at/9nST45xvidJE8kuXIavzLJE9P4J6d5qarXJbk8yVlJLkny2ao6/sUtHwAAAAAAgGPNggJXVa1JsjHJX0z7leT8JLdOU25Kctm0fem0n+n4BdP8S5N8YYzxizHGPyfZkeScpbgJAAAAAAAAjh0LfYLrvyf54yS/nvZPTfLkGOPZaX9XktOn7dOTPJIk0/GfTfOfH9/POc+rqquqantVbd+9e/cibgUAAAAAAIBjwUEDV1X92ySPjTHufQnWkzHGDWOMDWOMDatXr34p/iQAAAAAAACNzCxgzu8leXdVvTPJiiT/KsmfJzm5qmamp7TWJHl0mv9okjOS7KqqmSQrk+yZN/6c+ecAAAAAAADAghz0Ca4xxkfGGGvGGLNJLk/y9THGv0vyjSS/P03blOT2afuOaT/T8a+PMcY0fnlVnVhVa5OcmeTuJbsTAAAAAAAAjgkLeYLrQP4kyReq6k+T3J/kxmn8xiSfr6odSfZmLopljPFgVd2S5HtJnk1y9RjjVy/i7wMAAAAAAHAMWlTgGmP8fZK/n7Z/kOSc/czZl+QPDnD+dUmuW+wiAQAAAAAA4DkHfUUhAAAAAAAAHElezCsKAXiRZjdvPaTzdm7ZuMQrAQAAAADowxNcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArAhcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQisAFAAAAAABAKzPLvQCAFzK7eeshnbdzy8YlXgkAAAAAAEcKT3ABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArfgGFwAAAAAARw3fdIdjgye4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFZmlnsBAAAAAHA0mt289ZDO27ll4xKvBACOPp7gAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABo5aCBq6pWVNXdVfW/qurBqvqv0/jaqrqrqnZU1Rer6mXT+InT/o7p+Oy8a31kGv9+VV18uG4KAAAAAACAo9dCnuD6RZLzxxhvSrIuySVVdV6SP0vyyTHG7yR5IsmV0/wrkzwxjX9ympeqel2Sy5OcleSSJJ+tquOX8mYAAAAAAAA4+h00cI05/2faPWH6GUnOT3LrNH5Tksum7Uun/UzHL6iqmsa/MMb4xRjjn5PsSHLOktwFAAAAAAAAx4wFfYOrqo6vqgeSPJZkW5J/SvLkGOPZacquJKdP26cneSRJpuM/S3Lq/PH9nAMAAAAAAAALsqDANcb41RhjXZI1mXvq6ncP14Kq6qqq2l5V23fv3n24/gwAAAAAAABNLShwPWeM8WSSbyR5S5KTq2pmOrQmyaPT9qNJzkiS6fjKJHvmj+/nnPl/44YxxoYxxobVq1cvZnkAAAAAAAAcAw4auKpqdVWdPG3/iyQXJnkoc6Hr96dpm5LcPm3fMe1nOv71McaYxi+vqhOram2SM5PcvVQ3AgAAAAAAwLFh5uBT8qokN1XV8ZkLYreMMf62qr6X5AtV9adJ7k9y4zT/xiSfr6odSfYmuTxJxhgPVtUtSb6X5NkkV48xfrW0twMAAAAAAMDR7qCBa4zxnSTr9zP+g8x9j+u3x/cl+YMDXOu6JNctfpkAAAAAAAAwZ1Hf4AIAAAAAAIDlJnABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArAhcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArM8u9AAAAAODwmt289ZDO27ll4xKvBAAAloYnuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWZpZ7AXA0mt289ZDO27ll4xKvBAAAAAAAjj6e4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABaEbgAAAAAAABoReACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABamVnuBXBsmt289ZDO27ll4xKvBAAAAAAA6MYTXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArAhcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0MrMci8AAAAAgCPD7Oath3Tezi0bl3glAAAv7KBPcFXVGVX1jar6XlU9WFV/OI2vqqptVfXw9PuUabyq6lNVtaOqvlNVZ8+71qZp/sNVtenw3RYAAAAAAABHq4W8ovDZJP95jPG6JOclubqqXpdkc5I7xxhnJrlz2k+SdyQ5c/q5Ksn1yVwQS3JtknOTnJPk2ueiGAAAAAAAACzUQQPXGOPHY4z7pu3/neShJKcnuTTJTdO0m5JcNm1fmuTmMefbSU6uqlcluTjJtjHG3jHGE0m2JblkSe8GAAAAAACAo95CnuB6XlXNJlmf5K4kp40xfjwd+kmS06bt05M8Mu+0XdPYgcZ/+29cVVXbq2r77t27F7M8AAAAAAAAjgEzC51YVf8yyW1J/miM8fOqev7YGGNU1ViKBY0xbkhyQ5Js2LBhSa4JsFA+qAwAAAAAcORb0BNcVXVC5uLWX48xvjQN/3R69WCm349N448mOWPe6WumsQONAwAAAAAAwIIdNHDV3KNaNyZ5aIzxiXmH7kiyadrelOT2eePvqznnJfnZ9CrDrya5qKpOqapTklw0jQEAAAAAAMCCLeQVhb+X5N8n+YeqemAa+y9JtiS5paquTPLDJO+Zjn0lyTuT7EjydJL3J8kYY29VfTzJPdO8j40x9i7JXQAAAAAAAHDMOGjgGmN8K0kd4PAF+5k/klx9gGt9LsnnFrNAAAAAAAAAmG8hT3ABAMABzW7eekjn7dyycYlXAgAAABwrDvoNLgAAAAAAADiSeIKLJP7PawA4nPx3FgAAAGBpeYILAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaMU3uOAo4NsuAAAAAAAcSzzBBQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArAhcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0MrMci8AAAAAltvs5q2HdN7OLRuXeCUAAMBCeIILAAAAAACAVgQuAAAAAAAAWvGKQgAAAACOCV5HCgBHD09wAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQicAEAAAAAANCKwAUAAAAAAEArAhcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQisAFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0MrMci8AAAAAjjWzm7ce0nk7t2xc4pUAAEBPnuACAAAAAACgFYELAAAAAACAVgQuAAAAAAAAWhG4AAAAAAAAaEXgAgAAAAAAoBWBCwAAAAAAgFYELgAAAAAAAFoRuAAAAAAAAGhF4AIAAAAAAKAVgQsAAAAAAIBWBC4AAAAAAABamVnuBQDw0pndvPWQztu5ZeMSrwQAAAAA4NB5ggsAAAAAAIBWPMEFAAAAAAAcEbyBiIXyBBcAAAAAAACtCFwAAAAAAAC0InABAAAAAADQim9wHaG8ZxQAAAAAAGD/PMEFAAAAAABAKwIXAAAAAAAArQhcAAAAAAAAtCJwAQAAAAAA0IrABQAAAAAAQCsCFwAAAAAAAK0IXAAAAAAAALQys9wLAAAAAAAAFmZ289ZDOm/n/2vvzsNtuco6Af8+chMMCSEMYZAAkXkUmaFBQWU0MonQIqIMgjZi04DdgoIgRMzTIuAAMmgEHAABgSBzg2KDRhLCEKagQJiaOWbjhWMAAB6MSURBVCABQZLw9R9VJxxO7uXm3lTtc2rf932e/Zw6tc+pX9Xeu9auVavWquOPnXhNYHvpwQUAAAAAAMCiaOACAAAAAABgUTRwAQAAAAAAsCh7vQdXVZ2Q5CeTfL67rz/Ou1SSlyQ5JskZSe7T3V+uqkryB0l+Isl/JHlAd586/s8vJHncuNjjuvsF024KAAAAAAA7jXtGAXO4ID24np/kzlvmPSbJm7v7GknePP6eJHdJco3x8dAkf5Kc1yD2hCS3SHLzJE+oqkte2JUHAAAAAADgwLPXBq7u/sckZ26ZffckGz2wXpDkHpvmv7AHJyU5sqqukOROSd7U3Wd295eTvCnnbzQDAAAAAACAvdrfe3Bdrrs/M05/NsnlxukrJvnkpr/71DhvT/MBAAAAAABgn+xvA9d5uruT9ATrkiSpqodW1SlVdcoXvvCFqRYLAAAAAADAmtjfBq7PjUMPZvz5+XH+p5NcadPfHT3O29P88+nu53b3Tbv7pkcdddR+rh4AAAAAAADran8buE5M8gvj9C8kedWm+T9fg1sm+fdxKMM3JLljVV2yqi6Z5I7jPAAAAAAAANgnu/b2B1X1oiS3S3KZqvpUkickOT7J31TVg5N8PMl9xj9/bZKfSPJvSf4jyQOTpLvPrKonJzl5/LsndfeZE24HAAAAsEMc85jX7Nf/nXH8sROvCQAA62qvDVzdfd89PPXju/nbTvIre1jOCUlO2Ke1AwAAAAAAgC32d4hCAAAAAAAA2BYauAAAAAAAAFgUDVwAAAAAAAAsigYuAAAAAAAAFkUDFwAAAAAAAIuigQsAAAAAAIBF0cAFAAAAAADAomjgAgAAAAAAYFE0cAEAAAAAALAou7Z7BQAAgAPHMY95zX793xnHHzvxmgAAALBkenABAAAAAACwKBq4AAAAAAAAWBQNXAAAAAAAACyKBi4AAAAAAAAWRQMXAAAAAAAAi6KBCwAAAAAAgEXZtd0rAAAAwL475jGv2a//O+P4YydeEwAAgNXTgwsAAAAAAIBF0YPrAnJ1JAAAAAAAwM6gBxcAAAAAAACLooELAAAAAACARdHABQAAAAAAwKJo4AIAAAAAAGBRdm33CgAAMK1jHvOa/fq/M44/duI1AQAAAJiHHlwAAAAAAAAsih5cAAAA7Dh6owIAAN+LHlwAAAAAAAAsih5cAMCOsD9X6rtKH/he9AACAABYX3pwAQAAAAAAsCh6cAEAAExAjzEAAIDV0YMLAAAAAACARdGDiwOCq2kBAAAAAGB96MEFAAAAAADAomjgAgAAAAAAYFEMUQgAAAAAsM325xYbbq8BHMj04AIAAAAAAGBRNHABAAAAAACwKBq4AAAAAAAAWBQNXAAAAAAAACyKBi4AAAAAAAAWRQMXAAAAAAAAi6KBCwAAAAAAgEXZtd0rAAAAbJ9jHvOa/fq/M44/duI1AQAAgAtODy4AAAAAAAAWRQ8uAAAAAABgt4z6wE6lBxcAAAAAAACLogcXAACL4upBAAAAQAMXACyEk/oAAAAAMDBEIQAAAAAAAIuiBxcAAAAAwBZG0QDY2fTgAgAAAAAAYFE0cAEAAAAAALAoGrgAAAAAAABYFA1cAAAAAAAALMqu7V4BAIBVc7NoAAAAgGXTgwsAAAAAAIBF0YMLAACAvdL7FQAA2Ek0cAEAAAAAAHChrPqiOEMUAgAAAAAAsCh6cAEAAADsUIYHBQDYPQ1cwD7bnwqWyhUAAADAzqDxHFgHGrgAAAAAYA1otADgQOIeXAAAAAAAACyKHlwAsJ9cHQkAsDM4LgMA9pfjiOXSwAXAbBwgLJv3DwAAAICdarENXE66AQCwCo47AQB2BsdlAGzmHlwAAAAAAAAsysp7cFXVnZP8QZKDkvxpdx+/6nUAAAAAAC4cPaoAdr79KauXUk6vtIGrqg5K8swkd0jyqSQnV9WJ3f2BVa4HAOtJ5QoAgFVY5xNFAABLseoeXDdP8m/d/dEkqaoXJ7l7Eg1cAAAQjfUAAADrTJ1vOtXdqwur+ukkd+7uXxx/v3+SW3T3wzf9zUOTPHT89VpJTt+PqMsk+eKFXF158uTt7Cx58uQdOHnrvG3y5Mnbvrx13jZ58uRtX946b5s8efK2L2+dt02ePHnbl7eUbbtKdx+1uydWfg+uvenu5yZ57oVZRlWd0t03nWiV5MmTtwOz5MmTd+DkrfO2yZMnb/vy1nnb5MmTt31567xt8uTJ2768dd42efLkbV/eOmzbRaZc2AXw6SRX2vT70eM8AAAAAAAAuEBW3cB1cpJrVNUPVNUhSX4myYkrXgcAAAAAAAAWbKVDFHb3OVX18CRvSHJQkhO6+/0zRF2oIQ7lyZO3iCx58uQdOHnrvG3y5Mnbvrx13jZ58uRtX946b5s8efK2L2+dt02ePHnbl7f4bavunnqZAAAAAAAAMJtVD1EIAAAAAAAAF4oGLgAAAAAAABZFAxcAAAAAAACLooELAAAAAACARdm13Sswhaq6U5Kjk7y5u8/YNP9B3X3CDHmV5OZJrjjO+nSSd3R3T521l/W4dnd/aIbl3inJPfLd2/eq7n791Fl7WY/f6u4nzbTsg7v77C3zLtPdX5wh6yJJ0t3frqpDklw/yRndfebUWXvIf1h3P2sFOYcnuWaSj3b3V2ZY/iFJzt7Yz6rqR5PcOMkHuvt1M+T9YHe/d+rl7iXzykm+2t1fqapjktw0yYe6+30zZt40yZWSnJvkw3OUKVvyjkzy80mOyabvoO7+7xPnVJJ7J+kkL0vyY0nunuRDSZ7d3d+eMm9T7uWyqezs7s/NkbOXdTi8u7+26tw5VdWlVlVmjnl36+4TV5i3su2rqqsnuWGSD3b3B2ZY/pFzfAfsJXNXd58zTh+e5NoZvotme02r6qAkx+b8ZdnT5srclP2W7v6xmTPWsiyrqmtn+C7YfIx7Ynd/cMqcnULZOWmWsnOazKMy1NvPHbNmPV5ZRVm9tQ5ZVT+X4VzB+5I8b45zBOt+PmJctvr6PFmz1tk35ayqznfPJG/t7jPH8uX3k9woyQeSPLq7PzVl3pi57fuffW+y/LXZ95wvmy1zVWXZ05K8vLvfPuVyv0fetpdj43pMVpbVitd9clX1lCS3SXJqkrsmeUZ3/9H43KndfeOJ8+6Y5FlJ/jXDByAZDtKvnuRh3f3GKfP2si6f6O4rT7zMZ2QodF+YZONg4OgMO/S/dvcjpszby7rMsX0/muQvknxfhs/MQzcaRWf6vNwjyXOSfDvJLyf5jSRfS3KtJP+tu189cd6jts5K8tgkT0kmr8w9q7sfNk7fJslfJ/lIhn3hl7r7tVNljRnvSXK77v5yVf3PJPdM8tokt01ySnc/duK8c5N8NMmLk7xojpMZW/Iek+SXkvxnkqcm+bUkb09yyyR/NvVJ06q6bYYKwFeS3GTMumSSs5Pcv7s/OWXeptx/SnJSktMy7BdJku5+wcQ5z0py2SSHJPlqkosmOTHDSY7PTV2WVdUPJXl2kkvku78bvpLhu+HUKfP2si5zlJ03SPK8DAdAr0vy69395fG5d3T3zSfMunWSP83w+XhQkuOSXDXDe3mf7v7nqbLGvJ/aOivJM5M8LEm6+28nzntcdx83Tl83ySuTHDzm/tfu/peJ8/4+yb27+4tVdf8kj0/yj0lukeS5G8dME+adk+Qfkrwow0H6rCdsq+oBGcqyLyV5RIb37mMZjmX+V3e/aKbc1yb5Zs5flv32xDlbK46VYdtOH/N+cOK8tS3LqurXk9w3w/f65mPcn0ny4u4+fqqsMW9l5ea4TGXntHnKzmnzrpvkDzOcILpykndlOE57a5JHdPe/T5m3KXf2snpzHbKqHpfkhzPUi34yyae6+5FTZY0Z634+Qn192ryV1tk35a6qzveB7r7uOP2SMfOlSW6f5H7dfYeJ83bE/mff26+8td73nC+b5yLDFZZlX0jy8SRHJXlJhtf0XVNmbMraEeXYuC7TlWXdvehHhg/ZrnH6yAw78NPH3981Q94Hkxyzm/k/kOGKvqnz/nAPjz/K0HI9dd6H9zC/MjRwTZ331T08zkpyzgx5Jye53jj90xl26FvO+Hl5V5LLj5+Prya51jj/Khm+ZKbOOytDYfhbSZ4wPr68MT1x1qmbpv8+yY3H6avOtG3v2zR9SpJDx+ldSd4703t3/SS/k+TfkrwnyWN2t/9PlPf+JIcmufT4Ph41zj9s87ZPvH0bGT+Q5BXj9B2SvHGObdz6uZnzkeS08efBGU7eHDLz5+XdSW6xm/m3TPKeGfIetYfHo5OcOUPe25LcOcP37K+Nn9erbXyWJs56R5IbJLlVki8muc04/8ZJ3j7Dtp2d5O+SnJDkz8fHWePPE2bI21x2vibJXcbpmyf5pxnyNpedJye59Dh9sZn2hdMynNT7q3Hfe1WGBoRDp87alHeZfOd7duNzebk5tm9T7mzL3pJzYpK/zNCz4ioZTg5/cpy+ygx5a1uWJflwkoN3M/+QzHOMu7Jyc1ymsnPaPGXntHkn5Tv1oJsnecE4/ZAkL5tjG8flz15Wb96fM5wUPmycPjjj8ejEeet+PkJ9fdq8ldbZd5c75yPJ6Zum37nluXfPkLey/c++N3neWu97cb5s8syt7+Ocj419LMOFRo8ft/dD4+fzmhNnreVxxDrcg+u8oRV6uNLtrkmOqKqXZqiwTp6X71z1udmnMxzETu2BGYY3eOeWxylJvjVD3jer6ma7mX+zDFe/Te0rSa7R3UdseVw8yWdmyDuku9+fJN39sgxDMb5gvHqkZ8hLd3+2uz+W5BPdvXHF9cczzz3wrjcu97Akv9fD1Ylf7u7f7omvKt/iiB6v6u7uj2aebftqVV1/nP5ihiuLkmGfnCOvu/t93f2b3X31DBXwyyZ523gVx9TO7e5vZNgnvpHh5Ea6++szZCXJQd39hXH6ExkOJNPdb8p3uinP4S+q6iFVdYWqutTGY4acje+Fs5Oc3N3fGn8/J5uuvJnQYb2bq8e7+6QM++PUnpKhx93FtzwOzzz7w8W7+/Xd/ZXufmqShyd5fVXdMtOXnQd392k99Db4Qne/LUnGMubQibOS5L+Myz25ux/Y3Q9M8sVx+kEz5G32/T0OGdHd78g823d2VW3s019LslGm/GeSg+bI6+6/6+77ZbgS7K+S3CfJp6rqr2fIO7e7vzh+z36tuz+SJD3/kHqvG69+m1V33y3Jy5M8N8kNe7iS9uzu/vh4LDG1dS7Lvp3k+3cz/wqZ53thleVmouycmrJzWoduqgdtNMamu5+Xof4yl1WU1YdW1Y2q6iYZjq+/npx3DHruDHnrfj5CfX0+q6izb1hVne8fqupJVXXoOH3P5LzeSHP0DF3l/mffm9a673vOl81jVWVZJ0l3f7i7n9zd18twHPh9GTryTGktjyPW4R5cH6mq23b3W5Oku89N8uCqOi7JvWbIOyHJyVX14gxX0CbD/Wt+JsmfzZB3cobW6PMVEFX1xBnyHpDkT6rq4vnOB/5KGQ4OHjBD3gsznFjfXWVqjsrc2VV1+e7+bJJ09/ur6sczXIF6tRnyUlUX6eF+Pw/aNO+gzNAA292fSHLvqrp7kjdV1dOnztjk2uPQSZXkmKq6ZA/doS+SeRqXfznJX41drz+f5JSq+scMFeSnzJBXm38ZK+PvqKpHJ/mRGfJOHU9gHJbkzRkOJl+f4d5Rc3T3PqWq/izJW5LcLcOwOKmqi2WeEzcbvpXk95L8Zr5zoNwZrmSa0mdrvH9Ld995Y2ZVXT7zVAheV1WvyVCmbf5u+Pkkc9y/8NQkr+zud259oqp+cYa8VNUlehxCqLv/vqruleHE+9QHeJsPwLcOpTBHuXlyVd0hya+OQ1L9emaqxI2uWlUnZihjjq6qi3X3f4zPzXFA+cgkb6yql2e4EuwtVfWGDMM7//kMeeeVnWMl5G+S/E1VXSJDRXlqn6iq383QKPKhqvr9JH+bYWiaOS6U2XBSkleM33lnZ9ju7u4jpg7q7ldU1RuTPLmqHpx5vmM3rHNZ9j+SvLmq/jXf2bYrZxiO4+ETZyVZabmZKDunpuyc1keq6vEZjjt/KkNv0VTVwZn3JPsqyurPJNkYGunMqrpCd3+mqi6d8YKria37+Qj19Wmtus6+YVV1voePGaePvz+yqr6e5NVJ7j9xVrLa/c++N6EDYN9zvmweqyrLauuMHu5x9t6c/7j+wlrL44h1uAfXocl5FYGtz12xuz99/v+60JnXye5vUD3HDYcvleSbmypwKzGeBN58c/HPrjJ/LlV1+wxXtb5ny/xLJHl4d//OxHk3yzA0xTe3zD8mw9Axfzll3paMw5I8McNQQ5N/wVTVVbbM+n/dfXZVXSbJj/TE914YMw9KcscM3XY3rjp4Q89zk86f7e45Gln3lLcryb0zfFm+LMPwLT+boXfVM6e+MmU8ofCQJNfN0J38hO4+dyxTLztTz4BU1UeT3LxnuEntBcw/LEMPhc/PsOy7ZPffDZOPbV9V10rypd29jlV1uamvwK6qn81wQ9yTtsy/cpLHd/dDJsy6W5L/s/V7r6quluRe3f2/p8raTfb3J3lGkpt299QHrRsZt90y653d/bWqulySn+7uZ86QeYkM5cnmsvNVPcPNqavq13rorbISVXVEkl/JUHb+cZI7ZbhK7ONJjuvuWRq5qupjGfb303qFB9NVdcMkt+ruZ8+Ysc5l2UVy/hsqnzxeIDepVZab43KVndNnKjunyzsyw/1VNo47j+/us8bX+Dpb95MJc7elrB6zD0py0Tnq8et8PkJ9ffKMldfZx9yV1/nGz8iu7v7SzDkr2f/se/a9/ch0vmz63JWUZTVenD1nxpa8tTuOWHwDF6tTVdeeo0K3U/KA+dXQC+Eeq260B5jSeEXk7carTlm4VVcqgdVQVsP2UOcD1oGybDnW4R5ce1RVp64474nrnJfkjeuct+7v3yrz1nnb5F1oX0/y7qp6TlX94cZjxrzz2YbvhofKW16WPHl78dEM93p4bFU9auMxY975KMsmNefQJuez5q+lPHk7KW9by2rnI+QdwHkHYp3vieuYJU/eAZ53QJVlS37v1uEeXHvU3TdeceT57h+wtLzvsaNWkiOXnrcXi3//dlDeOm+bvAvnleNj22zDd8P5xlOWt4gsefK+l4+Nj0My7z0s9khZto8L2/NJ7Upy+JRZF8CiX0t58haUt61ltfMR8g7gvAOxzud8izx565d3oJVli33vDFHId6mqs5I8Osl/7ubp3+/uyyw5DwCAA09VfTPDTaLP2c3Tj+zuVV9YBQAAwIW0+CEKq+qsqvrqbh5nVdVXZ8q8U1U9eLz54eb5D1qDvJOTvK+7X7D1keSsNchb9/dvpXnrvG3yVmeObtCr/m6owX2q6t7j9I+P3dcfVlWTf9euc946b5u85eftZV0mH2artuE4dzfr8JZV5Kwg79Qkr+zu3976yEzHnFut0WspT97i8jblTlpWr7qcrqrLbPn958bvvYdW1eS94eTJm3BdnjjDMle9/92zqi41Th9VVS+sqtOq6iVVdfRSs+TJk7dP6/LEGZa5srKsqp5WVbeecpk7IU8Prn1UVU9JcpsMleS7JnlGd//R+NypU3cd3Ia8SyX55qpuoLcNeev+/q0sb523Td70eXtZl7t296tXlTeHqnpWkstmGP7mq0kumuTEJMcm+Vx3P0LezsuSJ2/idfml7n7OqvLmUFXv3ToryTWTnJ4k3f2DS82rqmsl+VJ3f3E3z12uuz83Vda4zLV9LeXJ2+l5e1mXRZfVm4/Tq+pxSX44yV8n+ckkn+ruR8qTt115e1mXdajzfaC7rztOvyTJSUlemuT2Se7X3XdYYpY8efL2aV0WXZZV1ReSfDzJUUlekuRF3f2uxed1t8c+PJKclmTXOH1kktcmefr4+7uWnrfuj3V//1aZt87bJk/Zsj+v5/jz4CRfSnLI+PuuJO+VtzOz5MnzON/reWKSv0xy7SRXSXJMkk+O01dZep7XUp68AyNvnR/ZdJye4UK1w8bpgze+E+XJ2668dX8kOX3T9Du3PPfupWbJkyfvwHlsfC9kuNDo8Unen+RDSZ6Q5JpLzVv8EIXbYFd3n5Mk3f2VDD0fjqiql2aeG9euOm+P5uiGuQ156/7+rTJvnbdN3gxqhwyJOJON1/LsJCd397fG389J8m15OzZLnrx9UoMdMSTiHLr7bklenuS5SW7Y3WckObu7P97dH1963p7UDMNLrvtrKU/eTs7bndqmIRFncGhV3aiqbpLkoO7+enLe9+C58uRtZ17toCERZ/IPVfWkqjp0nL5nklTVjyb59wVnyZMnb5PaQUMizqCTpLs/3N1P7u7rJblPku/LcOH7IvMWXxHfBh+pqttu/NLd53b3gzMMr3CdNcj7Xt65Bnnr/v6tMm+dt03exGoYEvE3k9wgyZur6lc3Pf3wqfO2wWer6vAk6e47b8ysqssn+Za8HZslT96+emaGA/L7J/mLJL+c4X6iP5Lk6TPkrVx3vyLJXZLcrqpelZkvqFp13h7MctJt3V9LefJ2al5VvXfL47Qkt974fa7cFflMkqcleWqSM6vqCklSVZfOeNGHPHnbmPfGjYkahkS8f4bzOncY12PpHp7hAqrTk9w7ycur6qwkD8mwrUvNkidP3nf7ne4+c5z+4yTvynAM87okfz5D3iqdr97T3e/t7sd299WXmuceXPtobC1Od39jN89dsbs/veS8dbfu798q89Z52+TNkndakht19zlVdWSGsd9P7+5HVtW7uvtGU+btFFV1WIahQD4vbzlZ8uR9j+We1t03qKqDk3w2yRW6+1tVtSvJqb3C+8isQlXdMMmtuvvZ65i3Suv+WsqTt5PyqurEDPdlPC7JNzKcXPm/Ge4/m1X1GlulqjooyUV7dfe2lidvd8s9r15XVacm+eHu/vp43HRqd99gyrztVFWXyDAqypfWKUuePHlJVZ3e3dcap9/Z3TfZ9Ny7u/uH5sqeW1Ud3t1fW7c8DVycT1XdKcnRSd7cw/ARG/Mf1N0nLD0PWL2q+mB3X2fT7wdlGKbmiCTXHbspA+xoW07cvH5Lr7FFV3YORFX1lu7+se1eD2B6NQxf9MgkT+3uE6vqo9191e1eL1hnVfWhJPfNMFrUCd19w03POU4CFqGqnpPkc0l+N8PFMm/r7lfUMCTiE7v7tt9zAaycIQonNF6hsui8WvEwYqvO28u6LP792yl567xt8vbbThpudaXW5P3bEXnrvG3yFpO36iERd4ylv3+1g4YsW/prKU/eEvJ2yPCnK7Uu7528ReetekjEHUOdSJ68tcpb9ZCIO8KS3zs9uPguteJhxFadB2yPMtwqsMZWPQQj++5AHLIMGKzz8KewBKseghFgKqsegpH9o4GL77LqYcQMWwYAwCoYsgwAAGC9aODaR2OXxN29aJWku/uIhef9XZLf6+63bpl/XJLf6O5Jh7Xchrx1f/9WlrfO2yZv+ry9rMup3X3jVeXNYd3fP2WLPHkXaF2UZTs8b8w8LMmTk1wtyU26++ipM8actX4t5cnbyXl7WZdFl9Xr/t7JW3beXtZl0fteok4kT96BkreXdVl0Wbau750GLr7LqocRM2wZAACrZsgyAACA5dPABQAAAAAAwKLs2u4VYDlW3Q1z6d0+ge/YSV3KAfaXsmw9OeaE9aKshu1h3wPWgbJsefTgAgAAAAAAYFEust0rAAAAAAAAAPvCEIV8l1V3w9TtEwCAuTnmBAAAWD+GKAQAAAAAAGBRDFEIAAAAAADAomjgAgAAAAAAYFE0cAEAAAAAALAoGrgAAAAAAABYFA1cAAAAK1BVx1TVB6vqeVX1/qp6Y1UdWlUPqaqTq+o9VfXyqrrY+PfPr6o/qaqTquqjVXW7qjphXMbzNy33jlX1z1V1alW9tKoO37aNBAAAWBENXAAAAKtzjSTP7O7rJflKknsl+dvuvll33zDJB5M8eNPfXzLJrZI8MsmJSZ6e5HpJblBVP1RVl0nyuCS37+4bJzklyaNWtjUAAADbZNd2rwAAAMAB5GPd/e5x+p1Jjkly/ao6LsmRSQ5P8oZNf//q7u6qOi3J57r7tCSpqveP/3t0kusmeXtVJckhSf55BdsBAACwrTRwAQAArM5/bpo+N8mhSZ6f5B7d/Z6qekCS2+3m77+95X+/naE+d26SN3X3fWdaXwAAgB3JEIUAAADb6+JJPlNVBye53z7+70lJbl1VV0+Sqjqsqq459QoCAADsNBq4AAAAttfjk/xLkrcn+dC+/GN3fyHJA5K8qKrem2F4wmtPvYIAAAA7TXX3dq8DAAAAAAAAXGB6cAEAAAAAALAoGrgAAAAAAABYFA1cAAAAAAAALIoGLgAAAAAAABZFAxcAAAAAAACLooELAAAAAACARdHABQAAAAAAwKJo4AIAAAAAAGBR/j/+krWWDhmhiwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 2160x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 2160x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "plt.rcParams[\"figure.figsize\"] = [30,10]\n",
    "\n",
    "sums = [sum(df[label]) for label in labels]\n",
    "df_sum = pd.DataFrame({'name': labels, 'number': sums})\n",
    "interval = 60\n",
    "\n",
    "for a, b in zip(range(0, len(labels), interval), range(interval, len(labels) + interval, interval)):\n",
    "    df_sum[a:b].plot.bar(x='name', ylim=(0, 7200))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Most frequent labels:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "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>name</th>\n",
       "      <th>number</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>69</th>\n",
       "      <td>C.2.1</td>\n",
       "      <td>7030</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>151</th>\n",
       "      <td>F.2.2</td>\n",
       "      <td>6946</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>77</th>\n",
       "      <td>C.4</td>\n",
       "      <td>6125</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>281</th>\n",
       "      <td>I.5.2</td>\n",
       "      <td>5125</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>292</th>\n",
       "      <td>I.6.5</td>\n",
       "      <td>5035</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>187</th>\n",
       "      <td>G.3</td>\n",
       "      <td>4767</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>308</th>\n",
       "      <td>J.2</td>\n",
       "      <td>4658</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>309</th>\n",
       "      <td>J.3</td>\n",
       "      <td>4437</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>226</th>\n",
       "      <td>H.5.2</td>\n",
       "      <td>4288</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>211</th>\n",
       "      <td>H.3.3</td>\n",
       "      <td>4256</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>251</th>\n",
       "      <td>I.2.8</td>\n",
       "      <td>3913</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>249</th>\n",
       "      <td>I.2.6</td>\n",
       "      <td>3758</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>291</th>\n",
       "      <td>I.6.4</td>\n",
       "      <td>3704</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>184</th>\n",
       "      <td>G.2.2</td>\n",
       "      <td>3583</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>76</th>\n",
       "      <td>C.3</td>\n",
       "      <td>3468</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>70</th>\n",
       "      <td>C.2.2</td>\n",
       "      <td>3435</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>150</th>\n",
       "      <td>F.2.1</td>\n",
       "      <td>3430</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>212</th>\n",
       "      <td>H.3.4</td>\n",
       "      <td>3415</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>72</th>\n",
       "      <td>C.2.4</td>\n",
       "      <td>3403</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>283</th>\n",
       "      <td>I.5.4</td>\n",
       "      <td>3332</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      name  number\n",
       "69   C.2.1    7030\n",
       "151  F.2.2    6946\n",
       "77     C.4    6125\n",
       "281  I.5.2    5125\n",
       "292  I.6.5    5035\n",
       "187    G.3    4767\n",
       "308    J.2    4658\n",
       "309    J.3    4437\n",
       "226  H.5.2    4288\n",
       "211  H.3.3    4256\n",
       "251  I.2.8    3913\n",
       "249  I.2.6    3758\n",
       "291  I.6.4    3704\n",
       "184  G.2.2    3583\n",
       "76     C.3    3468\n",
       "70   C.2.2    3435\n",
       "150  F.2.1    3430\n",
       "212  H.3.4    3415\n",
       "72   C.2.4    3403\n",
       "283  I.5.4    3332"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_sum.sort_values(by='number', ascending=False).head(20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Least frequent labels:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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>name</th>\n",
       "      <th>number</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>159</th>\n",
       "      <td>F.3.m</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>B.1.m</td>\n",
       "      <td>3</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>301</th>\n",
       "      <td>I.7.3</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>160</th>\n",
       "      <td>F.4</td>\n",
       "      <td>4</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>40</th>\n",
       "      <td>B.5.m</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>35</th>\n",
       "      <td>B.5</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>154</th>\n",
       "      <td>F.3</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>B.2</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>142</th>\n",
       "      <td>F.1</td>\n",
       "      <td>5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>239</th>\n",
       "      <td>I.1.m</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>81</th>\n",
       "      <td>C.5.2</td>\n",
       "      <td>7</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>330</th>\n",
       "      <td>K.5</td>\n",
       "      <td>8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>B.2.m</td>\n",
       "      <td>8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>B.2.3</td>\n",
       "      <td>9</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>78</th>\n",
       "      <td>C.5</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>B.4</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>85</th>\n",
       "      <td>C.5.m</td>\n",
       "      <td>10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>53</th>\n",
       "      <td>B.8</td>\n",
       "      <td>11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>297</th>\n",
       "      <td>I.7</td>\n",
       "      <td>11</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>B.1.3</td>\n",
       "      <td>11</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      name  number\n",
       "159  F.3.m       3\n",
       "12   B.1.m       3\n",
       "301  I.7.3       4\n",
       "160    F.4       4\n",
       "40   B.5.m       5\n",
       "35     B.5       5\n",
       "154    F.3       5\n",
       "13     B.2       5\n",
       "142    F.1       5\n",
       "239  I.1.m       7\n",
       "81   C.5.2       7\n",
       "330    K.5       8\n",
       "19   B.2.m       8\n",
       "17   B.2.3       9\n",
       "78     C.5      10\n",
       "27     B.4      10\n",
       "85   C.5.m      10\n",
       "53     B.8      11\n",
       "297    I.7      11\n",
       "9    B.1.3      11"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_sum.sort_values(by='number', ascending=True).head(20)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We see that there is only one case of class \"J\" (Computer Applications). We correct it to \"J.0\" (Computer Applications - GENERAL) and reload the dataset."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Text cleanup\n",
    "\n",
    "Our texts are a little bit \"dirty\" - they need some cleaning before we can feed them to classificators.\n",
    "\n",
    "We observe that some texts have unwanted characters (`\"<U+00BF>\"`) that are probably a result of problems with encoding at some stage of data acquisition. We will get rid of them with regex and remove all markup using `BeautifulSoup`.\n",
    "\n",
    "The texts are in english and have some english abbreviations, e.g \"what's\" - we will take care of them as well.\n",
    "\n",
    "Last but not least - we have to strip the texts with all whitespace and replace it with spaces and replace uppercase letters with lowercase letters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Before cleaning:\n",
      "None Responding to marketplace needs, today's embedded processors must feature a flexible core that allows easy modification with fast time to market. In this environment, embedded processors are increasingly reliant on flexible support tools. This paper presents one such tool, called Quick Piping, a new, high-level formalism for modeling processor pipelines. Quick Piping consists of three primary components that together provide an easy-to-build, reus<U+00BF>able processor description: Pipeline graphs-a new high-level formalism for model<U+00BF>ing processor pipelines,pipe--a companion domain-specific language for specify<U+00BF>ing a pipeline graph, pipe miner--a compiler specification generator for pipe descriptions. pipe miner processes a pipe description and produces a compiler specification that is used to build a compiler that reads the corresponding machine's instruc<U+00BF>tion set and automatically generates resource vectors.Despite their ubiquity and importance in achieving high perfor<U+00BF>mance in modern processors, pipelines--and improving the mech<U+00BF>anisms for specifying their operation--have received little attention. Until now, handwritten resource vectors have served to specify information about a processor's pipeline and encode rele<U+00BF>vant information about each instruction's resource usage. Describ<U+00BF>ing the complete set of resource vectors for a machine can be quite tedious and error prone, since it commonly must be developed by hand on an instruction-by-instruction basis.With its use of pipeline graphs, the pipe language, and the pipe miner compiler specification generator, Quick Piping gives the embedded processor architect and compiler writer an intuitive high-level abstraction of pipelines, a language for specifying a pipeline, and a tool for automatically producing pipeline resource vectors. The resulting specifications are quick to develop, easy to understand, simple to modify and maintain, and can be automati<U+00BF>cally processed to produce the low-level information required by processor control units and instruction schedulers.\n",
      "\n",
      "After cleaning:\n",
      "none responding to marketplace needs today embedded processors must feature a flexible core that allows easy modification with fast time to market in this environment embedded processors are increasingly reliant on flexible support tools this paper presents one such tool called quick piping a new high level formalism for modeling processor pipelines quick piping consists of three primary components that together provide an easy to build reus able processor description pipeline graphs a new high level formalism for model ing processor pipelines pipe a companion domain specific language for specify ing a pipeline graph pipe miner a compiler specification generator for pipe descriptions pipe miner processes a pipe description and produces a compiler specification that is used to build a compiler that reads the corresponding machine instruc tion set and automatically generates resource vectors despite their ubiquity and importance in achieving high perfor mance in modern processors pipelines and improving the mech anisms for specifying their operation have received little attention until now handwritten resource vectors have served to specify information about a processor pipeline and encode rele vant information about each instruction resource usage describ ing the complete set of resource vectors for a machine can be quite tedious and error prone since it commonly must be developed by hand on an instruction by instruction basis with its use of pipeline graphs the pipe language and the pipe miner compiler specification generator quick piping gives the embedded processor architect and compiler writer an intuitive high level abstraction of pipelines a language for specifying a pipeline and a tool for automatically producing pipeline resource vectors the resulting specifications are quick to develop easy to understand simple to modify and maintain and can be automati cally processed to produce the low level information required by processor control units and instruction schedulers\n"
     ]
    }
   ],
   "source": [
    "import re\n",
    "from bs4 import BeautifulSoup\n",
    "\n",
    "\n",
    "def clean_text(text: str):\n",
    "    text = re.sub(r\"<U\\+00F\\S>\", \"\", text)\n",
    "    text = re.sub(r\"<U\\+00\\S\\S>\", \" \", text)\n",
    "    text = BeautifulSoup(text, features=\"html.parser\").get_text()\n",
    "    \n",
    "    text = text.lower()\n",
    "    text = re.sub(r\"what's\", \"what is \", text)\n",
    "    text = re.sub(r\"\\'s\", \" \", text)\n",
    "    text = re.sub(r\"\\'ve\", \" have \", text)\n",
    "    text = re.sub(r\"can't\", \"can not \", text)\n",
    "    text = re.sub(r\"n't\", \" not \", text)\n",
    "    text = re.sub(r\"\\'d\", \" would \", text)\n",
    "    text = re.sub(r\"\\'ll\", \" will \", text)\n",
    "    text = re.sub('\\W', ' ', text)\n",
    "    text = re.sub('\\s+', ' ', text)\n",
    "    text = text.strip(' ')\n",
    "    return text\n",
    "\n",
    "print('Before cleaning:\\n{}\\n'.format(df['text'][0]))\n",
    "df['text'] = df['text'].apply(clean_text)\n",
    "print('After cleaning:\\n{}'.format(df['text'][0]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Classificators\n",
    "\n",
    "We will check 3 classification models:\n",
    "* **Naive Bayes**\n",
    "* **Linear SVC**\n",
    "* **Logistic Regression**\n",
    "\n",
    "\n",
    "The models do not support multi-target classification, so we have to fit and predict each label at a time. We use `OneVsRestClassifier` for that.\n",
    "\n",
    "We need to convert our texts to a format that can be accepted by classifiers. One of the possible formats is **td-idf** (term frequency–inverse document frequency), which reflects how important a word is to a document in a collection or corpus."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.feature_extraction.text import (\n",
    "    ENGLISH_STOP_WORDS,\n",
    "    TfidfVectorizer,\n",
    ")\n",
    "\n",
    "import nltk\n",
    "from nltk.corpus import stopwords\n",
    "\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.multiclass import OneVsRestClassifier\n",
    "from sklearn.naive_bayes import MultinomialNB\n",
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.svm import LinearSVC\n",
    "\n",
    "\n",
    "STOP_WORDS = set(stopwords.words('english')).update(ENGLISH_STOP_WORDS)\n",
    "\n",
    "\n",
    "def get_naive_bayes_pipeline(n_jobs=-1):\n",
    "    return Pipeline([\n",
    "        ('tfidf', TfidfVectorizer(analyzer='word', stop_words=STOP_WORDS, min_df=2, max_df=0.9, strip_accents='unicode')),\n",
    "        ('clf', OneVsRestClassifier(MultinomialNB(), n_jobs=n_jobs)),\n",
    "    ])\n",
    "\n",
    "def get_logistic_pipeline(C=1.0, solver='sag', max_iter=100, n_jobs=-1):\n",
    "    return Pipeline([\n",
    "        ('tfidf', TfidfVectorizer(analyzer='word', stop_words=STOP_WORDS, min_df=4, max_df=0.9, strip_accents='unicode')),\n",
    "        ('clf', OneVsRestClassifier(LogisticRegression(C=C, solver=solver, max_iter=max_iter), n_jobs=n_jobs)),\n",
    "    ])\n",
    "\n",
    "def get_svm_pipeline(C=1.0, n_jobs=-1):\n",
    "    return Pipeline([\n",
    "        ('tfidf', TfidfVectorizer(analyzer='word', stop_words=STOP_WORDS, min_df=4, max_df=0.9, strip_accents='unicode')),\n",
    "        ('clf', OneVsRestClassifier(LinearSVC(C=C), n_jobs=n_jobs)),\n",
    "    ])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def fit_and_predict(pipeline, df_train, df_test, labels):\n",
    "    predictions = pd.DataFrame(df_test['name'])\n",
    "    \n",
    "    for i, category in enumerate(labels):\n",
    "        pipeline.fit(df_train['text'], df_train[category])\n",
    "        pred = pipeline.predict(df_test['text'])\n",
    "        predictions = predictions.assign(**{category: pred})\n",
    "    \n",
    "    return predictions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Before predicting labels for our task and submitting the result to a challange platform, we have to check which classifiers and tricks work best, so we split our training data into training data (75%) and test data (25%) and check our models."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "\n",
    "df_train, df_test = train_test_split(df, random_state=42, test_size=0.25, shuffle=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Naive Bayes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "pred_nb = fit_and_predict(get_naive_bayes_pipeline(), df_train, df_test, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "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>name</th>\n",
       "      <th>A.0</th>\n",
       "      <th>A.1</th>\n",
       "      <th>A.2</th>\n",
       "      <th>A.m</th>\n",
       "      <th>B.0</th>\n",
       "      <th>B.1</th>\n",
       "      <th>B.1.0</th>\n",
       "      <th>B.1.1</th>\n",
       "      <th>B.1.2</th>\n",
       "      <th>...</th>\n",
       "      <th>K.7.3</th>\n",
       "      <th>K.7.4</th>\n",
       "      <th>K.7.m</th>\n",
       "      <th>K.8</th>\n",
       "      <th>K.8.0</th>\n",
       "      <th>K.8.1</th>\n",
       "      <th>K.8.2</th>\n",
       "      <th>K.8.3</th>\n",
       "      <th>K.8.m</th>\n",
       "      <th>K.m</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>75721</th>\n",
       "      <td>1342625.txt</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>80184</th>\n",
       "      <td>1381812.txt</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19864</th>\n",
       "      <td>336578.txt</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>76699</th>\n",
       "      <td>1247849.txt</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>92991</th>\n",
       "      <td>570742.txt</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 358 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              name  A.0  A.1  A.2  A.m  B.0  B.1  B.1.0  B.1.1  B.1.2  ...  \\\n",
       "75721  1342625.txt    0    0    0    0    0    0      0      0      0  ...   \n",
       "80184  1381812.txt    0    0    0    0    0    0      0      0      0  ...   \n",
       "19864   336578.txt    0    0    0    0    0    0      0      0      0  ...   \n",
       "76699  1247849.txt    0    0    0    0    0    0      0      0      0  ...   \n",
       "92991   570742.txt    0    0    0    0    0    0      0      0      0  ...   \n",
       "\n",
       "       K.7.3  K.7.4  K.7.m  K.8  K.8.0  K.8.1  K.8.2  K.8.3  K.8.m  K.m  \n",
       "75721      0      0      0    0      0      0      0      0      0    0  \n",
       "80184      0      0      0    0      0      0      0      0      0    0  \n",
       "19864      0      0      0    0      0      0      0      0      0    0  \n",
       "76699      0      0      0    0      0      0      0      0      0    0  \n",
       "92991      0      0      0    0      0      0      0      0      0    0  \n",
       "\n",
       "[5 rows x 358 columns]"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_nb.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Our challange is to achieve the best possible **F1 score**, so we will care about it the most, but we will also calculate **precision** and **accuracy**, to see what is happening and to be able to tweak our models."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import (\n",
    "    f1_score,\n",
    "    precision_score,\n",
    "    recall_score,\n",
    ")\n",
    "\n",
    "\n",
    "def calculate_f1(predictions, df_test, labels):\n",
    "    return f1_score(df_test.get(labels), predictions.get(labels), average='samples', zero_division=0)\n",
    "\n",
    "\n",
    "def calculate_precision(predictions, df_test, labels):\n",
    "    return precision_score(df_test.get(labels), predictions.get(labels), average='samples', zero_division=0)\n",
    "\n",
    "\n",
    "def calculate_recall(predictions, df_test, labels):\n",
    "    return recall_score(df_test.get(labels), predictions.get(labels), average='samples', zero_division=0)\n",
    "\n",
    "\n",
    "def print_scores(predictions, df_test, labels):\n",
    "    f1 = calculate_f1(predictions, df_test, labels)\n",
    "    print(f'F1 score: {f1}')\n",
    "\n",
    "    precision = calculate_precision(predictions, df_test, labels)\n",
    "    print(f'Precision: {precision}')\n",
    "\n",
    "    recall = calculate_recall(predictions, df_test, labels)\n",
    "    print(f'Recall: {recall}')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.011645238095238094\n",
      "Precision: 0.02001333333333333\n",
      "Recall: 0.008970380952380952\n"
     ]
    }
   ],
   "source": [
    "print_scores(pred_nb, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As we can see, Naive Bayes doesn't do well."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Linear SVC"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.2984154747474748\n",
      "Precision: 0.4365833333333334\n",
      "Recall: 0.2567800981240981\n"
     ]
    }
   ],
   "source": [
    "pred_svc_c_1 = fit_and_predict(get_svm_pipeline(C=1.0), df_train, df_test, labels)\n",
    "print_scores(pred_svc_c_1, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Linear SVC also has regularization parameter `C`. Let's check if increasing `C` improves performance.\n",
    "\n",
    "The default is `1.0`. Let's check `5.0`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.3107845863025863\n",
      "Precision: 0.40365934920634916\n",
      "Recall: 0.29466466955266957\n"
     ]
    }
   ],
   "source": [
    "pred_svc_c_5 = fit_and_predict(get_svm_pipeline(C=5.0), df_train, df_test, labels)\n",
    "print_scores(pred_svc_c_5, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Logistic regression"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.2245212079032079\n",
      "Precision: 0.3446484285714286\n",
      "Recall: 0.1878690808080808\n"
     ]
    }
   ],
   "source": [
    "pred_logistic_c_1 = fit_and_predict(get_logistic_pipeline(C=1.0), df_train, df_test, labels)\n",
    "print_scores(pred_logistic_c_1, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We will check how (and if) the regularization parameter (`C`) influences the result of Logistic Regression.\n",
    "\n",
    "The default is `1.0`. Let's check `0.1` and `10.0`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.050435142857142866\n",
      "Precision: 0.08423333333333334\n",
      "Recall: 0.039630904761904756\n"
     ]
    }
   ],
   "source": [
    "pred_logistic_c_01 = fit_and_predict(get_logistic_pipeline(C=0.1), df_train, df_test, labels)\n",
    "print_scores(pred_logistic_c_01, df_test, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.29959420712620716\n",
      "Precision: 0.4243862380952382\n",
      "Recall: 0.2657165743145743\n"
     ]
    }
   ],
   "source": [
    "pred_logistic_c_10 = fit_and_predict(get_logistic_pipeline(C=10.0), df_train, df_test, labels)\n",
    "print_scores(pred_logistic_c_10, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see that setting `C=10.0` improves all stats greatly. Let's check it for `C=20`!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.30456998712398714\n",
      "Precision: 0.42108809523809526\n",
      "Recall: 0.2750774314574314\n"
     ]
    }
   ],
   "source": [
    "pred_logistic_c_20 = fit_and_predict(get_logistic_pipeline(C=20.0), df_train, df_test, labels)\n",
    "print_scores(pred_logistic_c_20, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see that recall increased but precision decreased, so we suppose that it won't improve on further parameter increase."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "There is one more *clever* way of inreasing recall and decrease precision - decreasing the decision threshold:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "def fit_and_predict_logistic(df_train, df_test, labels, C=1.0, threshold=0.5, n_jobs=-1):\n",
    "    pipeline = get_logistic_pipeline(C=C, n_jobs=n_jobs)\n",
    "    predictions = pd.DataFrame(df_test['name'])\n",
    "\n",
    "    for i, category in enumerate(labels):\n",
    "        pipeline.fit(df_train['text'], df_train[category])\n",
    "        pred = (pipeline.predict_proba(df_test['text'])[:, 1] >= threshold).astype(int)\n",
    "        predictions = predictions.assign(**{category: pred})\n",
    "\n",
    "    return predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.3985782013280837\n",
      "Precision: 0.4314285452325452\n",
      "Recall: 0.4550399422799423\n"
     ]
    }
   ],
   "source": [
    "pred_logistic_c_1_t_015 = fit_and_predict_logistic(df_train, df_test, labels, C=1.0, threshold=0.15)\n",
    "print_scores(pred_logistic_c_1_t_015, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "After playing with multiple parameter configuration:\n",
    "\n",
    "* `C = 1.0, threshold = 0.5`\n",
    "\n",
    "```\n",
    "F1 score: 0.22399187456987457\n",
    "Precision: 0.34360509523809524\n",
    "Recall: 0.18752174747474745\n",
    "```\n",
    "\n",
    "* `C = 1.0, threshold = 0.4`\n",
    "\n",
    "```\n",
    "F1 score: 0.2720341351981352\n",
    "Precision: 0.3949718888888888\n",
    "Recall: 0.2381718759018759\n",
    "```\n",
    "\n",
    "* `C = 1.0, threshold = 0.3`\n",
    "\n",
    "```\n",
    "F1 score: 0.32396974003774004\n",
    "Precision: 0.4366673492063492\n",
    "Recall: 0.30217359018759016\n",
    "```\n",
    "\n",
    "* `C = 1.0, threshold = 0.2`\n",
    "\n",
    "```\n",
    "F1 score: 0.37612265922965926\n",
    "Precision: 0.4485183953823954\n",
    "Recall: 0.3906865281385281\n",
    "```\n",
    "\n",
    "* **`C = 1.0, threshold = 0.15`**\n",
    "\n",
    "```\n",
    "F1 score: 0.39857453899042133\n",
    "Precision: 0.43142635475635477\n",
    "Recall: 0.4550319422799423\n",
    "```\n",
    "\n",
    "* **`C = 1.0, threshold = 0.1`**\n",
    "\n",
    "```\n",
    "F1 score: 0.4063851057473359\n",
    "Precision: 0.38242067363029125\n",
    "Recall: 0.5433441803751803\n",
    "```\n",
    "\n",
    "* `C = 1.0, threshold = 0.05`\n",
    "\n",
    "```\n",
    "F1 score: 0.35440506616835415\n",
    "Precision: 0.26071897747888584\n",
    "Recall: 0.6852806132756133\n",
    "```\n",
    "\n",
    "* `C = 10.0, threshold = 0.5`\n",
    "\n",
    "```\n",
    "F1 score: 0.299329197025197\n",
    "Precision: 0.4241084285714285\n",
    "Recall: 0.265616240981241\n",
    "```\n",
    "\n",
    "* `C = 10.0, threshold = 0.4`\n",
    "\n",
    "```\n",
    "F1 score: 0.33141764857364864\n",
    "Precision: 0.4403473809523809\n",
    "Recall: 0.30966100288600285\n",
    "```\n",
    "\n",
    "* `C = 10.0, threshold = 0.3`\n",
    "\n",
    "```\n",
    "F1 score: 0.36254445110445116\n",
    "Precision: 0.4426806984126984\n",
    "Recall: 0.36432986002886003\n",
    "```\n",
    "\n",
    "* `C = 10.0, threshold = 0.2`\n",
    "\n",
    "```\n",
    "F1 score: 0.3869891514890339\n",
    "Precision: 0.41897179309579313\n",
    "Recall: 0.434980101010101\n",
    "```\n",
    "\n",
    "* `C = 10.0, threshold = 0.1`\n",
    "\n",
    "```\n",
    "F1 score: 0.3861434005596265\n",
    "Precision: 0.3424029859029859\n",
    "Recall: 0.5458180072150073\n",
    "```\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Combining results of multiple classificators\n",
    "\n",
    "If we have more than one classifier that performs well, then we could try to merge their results.\n",
    "\n",
    "If we have classifiers that have **high precision** and **low recall**, it means that they give labels accurately, but they don't give many labels. If we simply add labels for every text, we will *possibly* increase recall depending on how much the predictions overlap.\n",
    "\n",
    "If we have classifiers that have **low precision** and **high recall**, the labels that are needed are generaly given, but there are too many labels given.. If we take intersection of labels for every text, we will strip those labels that are not returned by all classifiers and *maybe* increase the precision.\n",
    "\n",
    "Our case is **high precision** and **low recall**."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "def combine_predictions(preds, labels):\n",
    "    res = preds[0].copy()\n",
    "    \n",
    "    for pred in preds[1:]:\n",
    "        for label in labels:\n",
    "            res[label] = res[label] | pred[label]\n",
    "    \n",
    "    return res"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.4030546928365752\n",
      "Precision: 0.43403316428016425\n",
      "Recall: 0.46128337085137083\n"
     ]
    }
   ],
   "source": [
    "pred_combined = combine_predictions([pred_logistic_c_1_t_015, pred_svc_c_1], labels)\n",
    "print_scores(pred_combined, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As we can see, it works!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Filling empty lines\n",
    "\n",
    "Even after merging results of multiple classifiers, some texts have no labels. One or more labels are required by the challenge rules, so we need to fill the empty lines. Filling them with most frequent label (or many labels) seems to be reasonable idea - we want to maximise the chance of hitting the right label.\n",
    "\n",
    "Let's check how many labels gives the best result:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "def fill_empty(predictions, default_labels: list):\n",
    "    res = predictions.copy()\n",
    "    \n",
    "    for i, row in res.iterrows():\n",
    "        if not row.get(labels).eq(1).any():\n",
    "            for label in default_labels:\n",
    "                res.at[i, label] = 1\n",
    "    return res"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Without filling:\n",
      "F1 score: 0.4030546928365752\n",
      "Precision: 0.43403316428016425\n",
      "Recall: 0.46128337085137083\n",
      "\n",
      "Filled with C.2.1:\n",
      "F1 score: 0.4041833941352765\n",
      "Precision: 0.43615316428016426\n",
      "Recall: 0.4621050851370851\n",
      "\n",
      "Filled with [C.2.1, C.4]:\n",
      "F1 score: 0.4055887563286387\n",
      "Precision: 0.4371131642801643\n",
      "Recall: 0.4636587994227994\n",
      "\n",
      "Filled with [C.2.1, F.2.2, C.4]:\n",
      "F1 score: 0.4059864657303481\n",
      "Precision: 0.4369531642801643\n",
      "Recall: 0.4645361327561327\n",
      "\n",
      "Filled with [C.2.1, F.2.2, C.4, I.5.2, I.6.5]:\n",
      "F1 score: 0.4064508544527368\n",
      "Precision: 0.43680916428016425\n",
      "Recall: 0.4661591803751804\n",
      "\n",
      "Filled with [C.2.1, F.2.2, C.4, I.5.2, I.6.5]:\n",
      "F1 score: 0.4064508544527368\n",
      "Precision: 0.43680916428016425\n",
      "Recall: 0.4661591803751804\n",
      "\n",
      "Filled with [C.2.1, F.2.2, C.4, I.5.2, I.6.5, G.3]:\n",
      "F1 score: 0.40664104126592365\n",
      "Precision: 0.4367731642801643\n",
      "Recall: 0.46710356132756126\n",
      "\n",
      "Filled with [C.2.1, F.2.2, C.4, I.5.2, I.6.5, G.3, J.2]:\n",
      "F1 score: 0.4068625269240563\n",
      "Precision: 0.4368103071373072\n",
      "Recall: 0.46811270418470413\n",
      "\n",
      "Filled with [C.2.1, F.2.2, C.4, I.5.2, I.6.5, G.3, J.2, J.3]:\n",
      "F1 score: 0.40676986033574264\n",
      "Precision: 0.4366331642801643\n",
      "Recall: 0.4686957518037518\n"
     ]
    }
   ],
   "source": [
    "print('Without filling:')\n",
    "print_scores(pred_combined, df_test, labels)\n",
    "\n",
    "pred_filled_1 = fill_empty(pred_combined, ['C.2.1'])\n",
    "print('\\nFilled with C.2.1:')\n",
    "print_scores(pred_filled_1, df_test, labels)\n",
    "\n",
    "pred_filled_2 = fill_empty(pred_combined, ['C.2.1', 'C.4'])\n",
    "print('\\nFilled with [C.2.1, C.4]:')\n",
    "print_scores(pred_filled_2, df_test, labels)\n",
    "\n",
    "pred_filled_3 = fill_empty(pred_combined, ['C.2.1', 'F.2.2', 'C.4'])\n",
    "print('\\nFilled with [C.2.1, F.2.2, C.4]:')\n",
    "print_scores(pred_filled_3, df_test, labels)\n",
    "\n",
    "pred_filled_5 = fill_empty(pred_combined, ['C.2.1', 'F.2.2', 'C.4', 'I.5.2', 'I.6.5'])\n",
    "print('\\nFilled with [C.2.1, F.2.2, C.4, I.5.2, I.6.5]:')\n",
    "print_scores(pred_filled_5, df_test, labels)\n",
    "\n",
    "pred_filled_5 = fill_empty(pred_combined, ['C.2.1', 'F.2.2', 'C.4', 'I.5.2', 'I.6.5'])\n",
    "print('\\nFilled with [C.2.1, F.2.2, C.4, I.5.2, I.6.5]:')\n",
    "print_scores(pred_filled_5, df_test, labels)\n",
    "\n",
    "pred_filled_6 = fill_empty(pred_combined, ['C.2.1', 'F.2.2', 'C.4', 'I.5.2', 'I.6.5', 'G.3'])\n",
    "print('\\nFilled with [C.2.1, F.2.2, C.4, I.5.2, I.6.5, G.3]:')\n",
    "print_scores(pred_filled_6, df_test, labels)\n",
    "\n",
    "pred_filled_7 = fill_empty(pred_combined, ['C.2.1', 'F.2.2', 'C.4', 'I.5.2', 'I.6.5', 'G.3', 'J.2'])\n",
    "print('\\nFilled with [C.2.1, F.2.2, C.4, I.5.2, I.6.5, G.3, J.2]:')\n",
    "print_scores(pred_filled_7, df_test, labels)\n",
    "\n",
    "pred_filled_8 = fill_empty(pred_combined, ['C.2.1', 'F.2.2', 'C.4', 'I.5.2', 'I.6.5', 'G.3', 'J.2', 'J.3'])\n",
    "print('\\nFilled with [C.2.1, F.2.2, C.4, I.5.2, I.6.5, G.3, J.2, J.3]:')\n",
    "print_scores(pred_filled_8, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As we can see, the more labels, the higher recall, but lower precision. I've decided to settle on 7 labels, as they give the best result."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "default_labels = ['C.2.1', 'F.2.2', 'C.4', 'I.5.2', 'I.6.5', 'G.3', 'J.2']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Stemming\n",
    "\n",
    "Sometimes it is better to reduce input size by **steming** - trimming words' endings and leaving the stem - it groups some words into one stem and reduces number of columns in any counting vector representation. It is known to increase recall and reduce precision, so we will try it."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "from nltk.stem import SnowballStemmer\n",
    "\n",
    "\n",
    "def stem(text):\n",
    "    stemmer = SnowballStemmer('english')\n",
    "    return ' '.join(stemmer.stem(word) for word in text.split())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.39857787232375463\n",
      "Precision: 0.43143235475635466\n",
      "Recall: 0.4550319422799423\n"
     ]
    }
   ],
   "source": [
    "df['text'] = df['text'].apply(clean_text)\n",
    "df['text'] = df['text'].apply(stem)\n",
    "df_train, df_test = train_test_split(df, random_state=42, test_size=0.25, shuffle=True)\n",
    "pred_stemmed_logistic = fit_and_predict_logistic(df_train, df_test, labels, C=1.0, threshold=0.15)\n",
    "print_scores(pred_stemmed_logistic, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "F1 without stemming: 0.39857454\n",
    "\n",
    "F1 with stemming: 0.39857787\n",
    "\n",
    "We can see that the result is almost identical, but the number of used words in tf-idf matrix decreased, so we will use stemming."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Association rules\n",
    "\n",
    "We can leverage association rules obtained using Apriori algorithm in R:\n",
    "```R\n",
    "#install.packages(c(\"arules\"))\n",
    "library(arules)\n",
    "\n",
    "data = data.table::fread(\"data/DM2020_training_docs_and_labels.txt\", header=FALSE, sep=\"\\t\")\n",
    "data = setNames(data, c(\"name\", \"text\", \"labels\"))\n",
    "data[, {labels := strsplit(labels, \",\")}]\n",
    "\n",
    "transactions = as(data[, labels], \"transactions\")\n",
    "\n",
    "minSupport = 0.0001\n",
    "frequentItemSets = apriori(transactions, parameter=list(supp=minSupport, minlen=1, maxlen=8, target=\"frequent itemsets\"), appearance=NULL, control=list(verbose=FALSE))\n",
    "rules = ruleInduction(frequentItemSets, transactions, confidence=0.7, control=list(method=\"apriori\", verbose=FALSE))\n",
    "\n",
    "rules = sort(rules, decreasing=TRUE, by=\"confidence\")\n",
    "inspect(head(rules, n=100))\n",
    "```\n",
    "\n",
    "We take only those with confidence greater than `0.9`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "rules = [\n",
    "    ({'C.2.2', 'G.2.3'}, 'G.2.2'),\n",
    "    ({'B.8.0', 'I.5.4'}, 'C.3'),\n",
    "    ({'F.4.2', 'I.2.7', 'I.6.4'}, 'I.6.5'),\n",
    "    ({'B.7.1', 'B.8.0', 'I.5.4'}, 'C.3'),\n",
    "    ({'B.8.0', 'C.4', 'I.5.4'}, 'C.3'),\n",
    "    ({'H.5.1', 'H.5.4', 'J.1'}, 'K.3.1'),\n",
    "    ({'I.6.4', 'I.6.6', 'J.1'}, 'I.6.3'),\n",
    "    ({'C.2.1', 'I.6.3', 'I.6.6'}, 'C.4'),\n",
    "    ({'F.2.2', 'I.5.3', 'I.6.5'}, 'J.3'),\n",
    "    ({'B.7.1', 'C.4', 'I.5.4'}, 'C.3'),\n",
    "    ({'I.5.1', 'I.6.4', 'I.6.5', 'J.3'}, 'I.5.2'),\n",
    "    ({'I.4.6', 'I.4.7', 'I.4.8', 'I.5.2', 'J.2'}, 'I.4.9'),\n",
    "    ({'D.4.7', 'D.4.8', 'F.2.2'}, 'D.4.1'),\n",
    "    ({'I.5.1', 'I.6.4', 'J.3'}, 'I.5.2'),\n",
    "    ({'I.4.7', 'I.4.8', 'I.5.2', 'J.2'}, 'I.4.9'),\n",
    "    ({'H.3.5', 'H.5.1', 'J.1'}, 'K.3.1'),\n",
    "    ({'C.2.0', 'C.2.3', 'K.6.4'}, 'K.6.5'),\n",
    "    ({'C.5.3', 'H.3.5', 'J.1'}, 'K.3.1'),\n",
    "    ({'C.2.1', 'H.1.1', 'I.6.5'}, 'E.4'),\n",
    "    ({'G.1.0', 'I.6.6', 'J.2'}, 'I.6.5'),\n",
    "    ({'F.2.2', 'I.6.4', 'J.3'}, 'I.5.2'),\n",
    "    ({'I.4.6', 'I.4.7', 'I.4.9', 'J.2'}, 'I.5.2'),\n",
    "    ({'H.3.1', 'I.2.6', 'I.6.5'}, 'I.2.7'),\n",
    "    ({'I.4.6', 'I.6.4', 'J.2'}, 'I.5.2'),\n",
    "    ({'G.0', 'I.6.8'}, 'J.2'),\n",
    "    ({'G.1.0', 'G.1.10', 'I.6.4', 'J.7'}, 'I.6.5'),\n",
    "    ({'B.8.0', 'C.4', 'I.5.4'}, 'B.7.1'),\n",
    "    ({'H.3.5', 'I.2.0', 'K.3.1'}, 'J.1'),\n",
    "    ({'H.3.5', 'I.2.0', 'J.1'}, 'K.3.1'),\n",
    "    ({'B.7.1', 'C.1.3', 'I.5.4'}, 'C.3'),\n",
    "    ({'B.8.0', 'C.3', 'C.4', 'I.5.4'}, 'B.7.1'),\n",
    "    ({'I.4.6', 'I.4.7', 'I.4.8', 'J.2'}, 'I.4.9'),\n",
    "    ({'I.4.6', 'I.4.7', 'J.2'}, 'I.5.2'),\n",
    "    ({'B.8.0', 'I.5.4'}, 'B.7.1'),\n",
    "    ({'B.8.0', 'C.3', 'I.5.4'}, 'B.7.1'),\n",
    "    ({'H.1.2', 'H.5.1', 'J.1'}, 'K.3.1'),\n",
    "    ({'I.4.6', 'I.4.8', 'I.5.2', 'J.2'}, 'I.4.9'),\n",
    "    ({'I.4.6', 'I.4.7', 'I.4.8', 'I.4.9', 'J.2'}, 'I.5.2'),\n",
    "    ({'H.3.5', 'I.6.4', 'K.3.1'}, 'J.1'),\n",
    "    ({'H.3.5', 'I.6.4', 'J.1'}, 'K.3.1'),\n",
    "]\n",
    "\n",
    "\n",
    "def apply_rules(predictions, rules):\n",
    "    res = predictions.copy()\n",
    "    \n",
    "    for i, row in predictions.iterrows():\n",
    "            pred_labels = set()\n",
    "            \n",
    "            for label in labels:\n",
    "                if row[label] == 1:\n",
    "                    pred_labels.add(label)\n",
    "            \n",
    "            for rule_set, rule_label in rules:\n",
    "                if rule_set.issubset(pred_labels):\n",
    "                    res.at[i, rule_label] = 1\n",
    "    return res"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Before applying the rules:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.39857787232375463\n",
      "Precision: 0.43143235475635466\n",
      "Recall: 0.4550319422799423\n"
     ]
    }
   ],
   "source": [
    "print_scores(pred_stemmed_logistic, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "After applying the rules:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.3985619939798763\n",
      "Precision: 0.4314019579309579\n",
      "Recall: 0.4550552756132756\n"
     ]
    }
   ],
   "source": [
    "pred_stemmed_logistic_rules = apply_rules(pred_stemmed_logistic, rules)\n",
    "print_scores(pred_stemmed_logistic_rules, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As we can see, it doesn't affect the result much. It slightly increases recall at the expense of accuracy. It looks like our models are already good at \"spotting\" those associations."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## One last trick\n",
    "\n",
    "Let's look at distribution of the number of labels of each document."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = load_training_data('data/DM2020_training_docs_and_labels.txt')\n",
    "df_count = df.sum(axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import seaborn as sns\n",
    "\n",
    "f, axes = plt.subplots(1, 2)\n",
    "sns.kdeplot(df_count, ax=axes[0])\n",
    "sns.boxplot(df_count, ax=axes[1])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "A we can see, most texts in our dataset have from 1 to 5 labels, peaking at 3. We could leverage that observation to create better models.\n",
    "\n",
    "Our model of logistic regression gives us probability values and classifies texts using a threshold. If the threshold is **low**, we get too many texts with **more than 5 labels**. If the threshold is **high**, we get a lot of texts with **0 labels**.\n",
    "\n",
    "What if we assigned labels based on the threshold, and if we get less than `min_labels` labels, we take `min_labels` labels with highest probabilities. Similarily, if we get more than `max_labels` labels, we could take `max_labels` labels with highest probabilities. We suspect that this technique will result in higher scores than just assigning `default_labels` if we get 0 labels.\n",
    "\n",
    "Let's check it out."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "df['text'] = df['text'].apply(clean_text)\n",
    "df['text'] = df['text'].apply(stem)\n",
    "df_train, df_test = train_test_split(df, random_state=42, test_size=0.25, shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_proba_logistic(df_train, df_test, labels, C=1.0, n_jobs=-1):\n",
    "    pipeline = get_logistic_pipeline(C=C, n_jobs=n_jobs)\n",
    "    predictions = pd.DataFrame(df_test['name'])\n",
    "\n",
    "    for i, category in enumerate(labels):\n",
    "        print(f'{category},', end='')\n",
    "        pipeline.fit(df_train['text'], df_train[category])\n",
    "        pred = pipeline.predict_proba(df_test['text'])[:, 1]\n",
    "        predictions = predictions.assign(**{category: pred})\n",
    "    return predictions\n",
    "\n",
    "\n",
    "def get_preds_from_proba(predictions, threshold=0.1, min_labels=2, max_labels=6):\n",
    "    pred = predictions.copy()\n",
    "    \n",
    "    how_many = (pred.iloc[:, 1:] >= threshold).sum(axis=1)\n",
    "    ranked = pred.iloc[:, 1:].rank(axis=1, ascending=False)\n",
    "    \n",
    "    for i, row in pred.iterrows():\n",
    "        if min_labels <= how_many.loc[i] <= max_labels:\n",
    "            pred.at[i, 1:] = (row[1:] >= threshold)\n",
    "        else:\n",
    "            if how_many[i] > max_labels:\n",
    "                pred.at[i, 1:] = (ranked.loc[i, :] < (max_labels + 1)) \n",
    "            else: # how_many[i] < min_labels:\n",
    "                pred.at[i, 1:] = (ranked.loc[i, :] < (min_labels + 1))\n",
    "    \n",
    "    pred = pred.drop('name', axis=1)\n",
    "    pred = pred.astype(int)\n",
    "    return pred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "A.0,A.1,A.2,A.m,B.0,B.1,B.1.0,B.1.1,B.1.2,B.1.3,B.1.4,B.1.5,B.1.m,B.2,B.2.0,B.2.1,B.2.2,B.2.3,B.2.4,B.2.m,B.3,B.3.0,B.3.1,B.3.2,B.3.3,B.3.4,B.3.m,B.4,B.4.0,B.4.1,B.4.2,B.4.3,B.4.4,B.4.5,B.4.m,B.5,B.5.0,B.5.1,B.5.2,B.5.3,B.5.m,B.6,B.6.0,B.6.1,B.6.2,B.6.3,B.6.m,B.7,B.7.0,B.7.1,B.7.2,B.7.3,B.7.m,B.8,B.8.0,B.8.1,B.8.2,B.8.m,B.m,C.0,C.1,C.1.0,C.1.1,C.1.2,C.1.3,C.1.4,C.1.m,C.2,C.2.0,C.2.1,C.2.2,C.2.3,C.2.4,C.2.5,C.2.6,C.2.m,C.3,C.4,C.5,C.5.0,C.5.1,C.5.2,C.5.3,C.5.4,C.5.5,C.5.m,C.m,D.0,D.1,D.1.0,D.1.1,D.1.2,D.1.3,D.1.4,D.1.5,D.1.6,D.1.7,D.1.m,D.2,D.2.0,D.2.1,D.2.10,D.2.11,D.2.12,D.2.13,D.2.2,D.2.3,D.2.4,D.2.5,D.2.6,D.2.7,D.2.8,D.2.9,D.2.m,D.3,D.3.0,D.3.1,D.3.2,D.3.3,D.3.4,D.3.m,D.4,D.4.0,D.4.1,D.4.2,D.4.3,D.4.4,D.4.5,D.4.6,D.4.7,D.4.8,D.4.9,D.4.m,D.m,E.0,E.1,E.2,E.3,E.4,E.5,E.m,F.0,F.1,F.1.0,F.1.1,F.1.2,F.1.3,F.1.m,F.2,F.2.0,F.2.1,F.2.2,F.2.3,F.2.m,F.3,F.3.0,F.3.1,F.3.2,F.3.3,F.3.m,F.4,F.4.0,F.4.1,F.4.2,F.4.3,F.4.m,F.m,G.0,G.1,G.1.0,G.1.1,G.1.10,G.1.2,G.1.3,G.1.4,G.1.5,G.1.6,G.1.7,G.1.8,G.1.9,G.1.m,G.2,G.2.0,G.2.1,G.2.2,G.2.3,G.2.m,G.3,G.4,G.m,H.0,H.1,H.1.0,H.1.1,H.1.2,H.1.m,H.2,H.2.0,H.2.1,H.2.2,H.2.3,H.2.4,H.2.5,H.2.6,H.2.7,H.2.8,H.2.m,H.3,H.3.0,H.3.1,H.3.2,H.3.3,H.3.4,H.3.5,H.3.6,H.3.7,H.3.m,H.4,H.4.0,H.4.1,H.4.2,H.4.3,H.4.m,H.5,H.5.0,H.5.1,H.5.2,H.5.3,H.5.4,H.5.5,H.5.m,H.m,I.0,I.1,I.1.0,I.1.1,I.1.2,I.1.3,I.1.4,I.1.m,I.2,I.2.0,I.2.1,I.2.10,I.2.11,I.2.2,I.2.3,I.2.4,I.2.5,I.2.6,I.2.7,I.2.8,I.2.9,I.2.m,I.3,I.3.0,I.3.1,I.3.2,I.3.3,I.3.4,I.3.5,I.3.6,I.3.7,I.3.8,I.3.m,I.4,I.4.0,I.4.1,I.4.10,I.4.2,I.4.3,I.4.4,I.4.5,I.4.6,I.4.7,I.4.8,I.4.9,I.4.m,I.5,I.5.0,I.5.1,I.5.2,I.5.3,I.5.4,I.5.5,I.5.m,I.6,I.6.0,I.6.1,I.6.2,I.6.3,I.6.4,I.6.5,I.6.6,I.6.7,I.6.8,I.6.m,I.7,I.7.0,I.7.1,I.7.2,I.7.3,I.7.4,I.7.5,I.7.m,I.m,J.0,J.1,J.2,J.3,J.4,J.5,J.6,J.7,J.m,K.0,K.1,K.2,K.3,K.3.0,K.3.1,K.3.2,K.3.m,K.4,K.4.0,K.4.1,K.4.2,K.4.3,K.4.4,K.4.m,K.5,K.5.0,K.5.1,K.5.2,K.5.m,K.6,K.6.0,K.6.1,K.6.2,K.6.3,K.6.4,K.6.5,K.6.m,K.7,K.7.0,K.7.1,K.7.2,K.7.3,K.7.4,K.7.m,K.8,K.8.0,K.8.1,K.8.2,K.8.3,K.8.m,K.m,"
     ]
    }
   ],
   "source": [
    "proba_logistic = get_proba_logistic(df_train, df_test, labels, C=1.0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Results for min_labels=1, max_labels=1:\n",
      "F1 score: 0.3337041690858161\n",
      "Precision: 0.60276\n",
      "Recall: 0.24867904256854256\n",
      "\n",
      "Results for min_labels=1, max_labels=2:\n",
      "F1 score: 0.3952175832272303\n",
      "Precision: 0.53706\n",
      "Recall: 0.349014613997114\n",
      "\n",
      "Results for min_labels=1, max_labels=3:\n",
      "F1 score: 0.40747159565924274\n",
      "Precision: 0.5079199999999999\n",
      "Recall: 0.394013106060606\n",
      "\n",
      "Results for min_labels=1, max_labels=4:\n",
      "F1 score: 0.40853521603886306\n",
      "Precision: 0.49578666666666654\n",
      "Recall: 0.41271036002886\n",
      "\n",
      "Results for min_labels=2, max_labels=2:\n",
      "F1 score: 0.4076533870573871\n",
      "Precision: 0.48994\n",
      "Recall: 0.3853682806637807\n",
      "\n",
      "Results for min_labels=2, max_labels=3:\n",
      "F1 score: 0.41990739948939954\n",
      "Precision: 0.4608\n",
      "Recall: 0.4303667727272727\n",
      "\n",
      "Results for min_labels=2, max_labels=4:\n",
      "F1 score: 0.4209710198690198\n",
      "Precision: 0.44866666666666666\n",
      "Recall: 0.4490640266955267\n",
      "\n",
      "Results for min_labels=3, max_labels=3:\n",
      "F1 score: 0.42194833062843595\n",
      "Precision: 0.4165733333333333\n",
      "Recall: 0.47798586796536796\n",
      "\n",
      "Results for min_labels=3, max_labels=4:\n",
      "F1 score: 0.4230119510080563\n",
      "Precision: 0.40444\n",
      "Recall: 0.49668312193362185\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# new scores\n",
    "for min_labels in range(1, 4):\n",
    "    for max_labels in range(min_labels, 5):\n",
    "        pred_with_bounds = get_preds_from_proba(proba_logistic, threshold=0.2, min_labels=min_labels, max_labels=max_labels)\n",
    "        print(f'Results for min_labels={min_labels}, max_labels={max_labels}:')\n",
    "        print_scores(pred_with_bounds, df_test, labels)\n",
    "        print('')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Results for min_labels=1, max_labels=1:\n",
      "F1 score: 0.3362643838383838\n",
      "Precision: 0.60796\n",
      "Recall: 0.2505395743145743\n",
      "\n",
      "Results for min_labels=1, max_labels=2:\n",
      "F1 score: 0.40227117349317354\n",
      "Precision: 0.52306\n",
      "Recall: 0.3638941298701299\n",
      "\n",
      "Results for min_labels=1, max_labels=3:\n",
      "F1 score: 0.4173662346542347\n",
      "Precision: 0.48306\n",
      "Recall: 0.4242084155844156\n",
      "\n",
      "Results for min_labels=1, max_labels=4:\n",
      "F1 score: 0.4185505692085692\n",
      "Precision: 0.46371\n",
      "Recall: 0.4544254011544011\n",
      "\n",
      "Results for min_labels=2, max_labels=2:\n",
      "F1 score: 0.40918571317571323\n",
      "Precision: 0.49212\n",
      "Recall: 0.386564606060606\n",
      "\n",
      "Results for min_labels=2, max_labels=3:\n",
      "F1 score: 0.42428077433677436\n",
      "Precision: 0.45212\n",
      "Recall: 0.44687889177489176\n",
      "\n",
      "Results for min_labels=2, max_labels=4:\n",
      "F1 score: 0.4254651088911089\n",
      "Precision: 0.43277\n",
      "Recall: 0.47709587734487735\n",
      "\n",
      "Results for min_labels=3, max_labels=3:\n",
      "F1 score: 0.422604148962149\n",
      "Precision: 0.41745333333333334\n",
      "Recall: 0.4784620822510822\n",
      "\n",
      "Results for min_labels=3, max_labels=4:\n",
      "F1 score: 0.4237884835164835\n",
      "Precision: 0.39810333333333336\n",
      "Recall: 0.5086790678210679\n",
      "\n"
     ]
    }
   ],
   "source": [
    "for min_labels in range(1, 4):\n",
    "    for max_labels in range(min_labels, 5):\n",
    "        pred_with_bounds = get_preds_from_proba(proba_logistic, threshold=0.15, min_labels=min_labels, max_labels=max_labels)\n",
    "        print(f'Results for min_labels={min_labels}, max_labels={max_labels}:')\n",
    "        print_scores(pred_with_bounds, df_test, labels)\n",
    "        print('')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As we can see, we got quite good result (0.4255) for bounds [2, 4]."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.4254651088911089\n",
      "Precision: 0.43277\n",
      "Recall: 0.47709587734487735\n"
     ]
    }
   ],
   "source": [
    "pred_logistic_with_bounds = get_preds_from_proba(proba_logistic, threshold=0.15, min_labels=2, max_labels=4)\n",
    "print_scores(pred_logistic_with_bounds, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can try to merge those results with results from SVC. We know that SVC gave us high precision, so we hope that the increase from recall will exceed the loss of precision."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score: 0.42648932067932066\n",
      "Precision: 0.4316039047619048\n",
      "Recall: 0.48085454401154404\n"
     ]
    }
   ],
   "source": [
    "pred_combined = combine_predictions([pred_logistic_with_bounds, pred_svc_c_1], labels)\n",
    "print_scores(pred_combined, df_test, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As we predicted, precision is slightly lower, recall is slightly higher, but the F1 score is higher."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Challange data\n",
    "\n",
    "Let's combine our techniques to create a submission."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "\n",
    "\n",
    "def load_test_data(filepath):\n",
    "    with open(filepath, 'rb') as f:\n",
    "        decoded_lines = (line.decode(encoding='iso_8859_1') for line in f)\n",
    "\n",
    "        parsed_lines = csv.reader(decoded_lines, dialect=\"excel-tab\")\n",
    "\n",
    "        names = []\n",
    "        texts = []\n",
    "    \n",
    "        for name, text, _ in parsed_lines:\n",
    "            names.append(name)\n",
    "            texts.append(text)\n",
    "\n",
    "    return pd.DataFrame({'name': names, 'text': texts})\n",
    "\n",
    "\n",
    "def write_prediction(predictions, filepath):\n",
    "    with open(filepath, 'w') as f:\n",
    "        for i, (_, row) in enumerate(predictions.iterrows()):\n",
    "            pred_labels = []\n",
    "            for label in labels:\n",
    "                if row[label] == 1:\n",
    "                    pred_labels.append(label)\n",
    "            \n",
    "            f.write(','.join(pred_labels) + '\\n')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = load_training_data('data/DM2020_training_docs_and_labels.txt')\n",
    "df_challenge = load_test_data('data/DM2020_test_docs.txt')\n",
    "\n",
    "df['text'] = df['text'].apply(clean_text)\n",
    "df_challenge['text'] = df_challenge['text'].apply(clean_text)\n",
    "\n",
    "df['text'] = df['text'].apply(stem)\n",
    "df_challenge['text'] = df_challenge['text'].apply(stem)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "challenge_svc = fit_and_predict(get_svm_pipeline(C=1.0), df, df_challenge, labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [],
   "source": [
    "challenge_proba_logistic = get_proba_logistic(df, df_challenge, labels, C=1.0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "challenge_logistic = get_preds_from_proba(challenge_proba_logistic, threshold=0.13, min_labels=2, max_labels=4)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [],
   "source": [
    "challenge_combined = combine_predictions([challenge_logistic, challenge_svc], labels)\n",
    "write_prediction(challenge_combined, 'results/challenge_t013_b2-4.txt')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Our score on partial dataset is `0.4296`.\n",
    "After final evaluation on the entire dataset, our score is `0.429108`, which gives us the third place in the challange!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "*Author: Jakub Godula*"
   ]
  }
 ],
 "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": "3.7.5"
  },
  "pycharm": {
   "stem_cell": {
    "cell_type": "raw",
    "source": [],
    "metadata": {
     "collapsed": false
    }
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}