LinearSVM_Top10.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "0b162649",
   "metadata": {},
   "outputs": [
    {
     "ename": "ModuleNotFoundError",
     "evalue": "No module named 'dask'",
     "output_type": "error",
     "traceback": [
      "\u001b[1;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[1;31mModuleNotFoundError\u001b[0m                       Traceback (most recent call last)",
      "\u001b[1;32m~\\AppData\\Local\\Temp\\16/ipykernel_10340/2377632533.py\u001b[0m in \u001b[0;36m<module>\u001b[1;34m\u001b[0m\n\u001b[0;32m      2\u001b[0m \u001b[1;32mimport\u001b[0m \u001b[0mpandas\u001b[0m \u001b[1;32mas\u001b[0m \u001b[0mpd\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      3\u001b[0m \u001b[1;32mimport\u001b[0m \u001b[0mnumpy\u001b[0m \u001b[1;32mas\u001b[0m \u001b[0mnp\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[1;32m----> 4\u001b[1;33m \u001b[1;32mimport\u001b[0m \u001b[0mdask\u001b[0m\u001b[1;33m.\u001b[0m\u001b[0mbag\u001b[0m \u001b[1;32mas\u001b[0m \u001b[0mdb\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0m\u001b[0;32m      5\u001b[0m \u001b[1;32mimport\u001b[0m \u001b[0mjson\u001b[0m\u001b[1;33m\u001b[0m\u001b[1;33m\u001b[0m\u001b[0m\n\u001b[0;32m      6\u001b[0m \u001b[1;33m\u001b[0m\u001b[0m\n",
      "\u001b[1;31mModuleNotFoundError\u001b[0m: No module named 'dask'"
     ]
    }
   ],
   "source": [
    "# Importing dependencies\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import dask.bag as db\n",
    "import json\n",
    " \n",
    "# Opening JSON file\n",
    "f = open('golden_json_with_full_abstract_and_titles_and_isOpenAccess.json')\n",
    "\n",
    "docs = json.load(f)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "e21c06a9",
   "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>paper</th>\n",
       "      <th>mag</th>\n",
       "      <th>venue_mag</th>\n",
       "      <th>author</th>\n",
       "      <th>reference</th>\n",
       "      <th>title</th>\n",
       "      <th>abstract</th>\n",
       "      <th>petalID</th>\n",
       "      <th>doi</th>\n",
       "      <th>venue</th>\n",
       "      <th>level1</th>\n",
       "      <th>level2</th>\n",
       "      <th>level3</th>\n",
       "      <th>url</th>\n",
       "      <th>isBiomimicry</th>\n",
       "      <th>fullDocLink</th>\n",
       "      <th>isOpenAccess</th>\n",
       "      <th>abstract_full</th>\n",
       "      <th>title_full</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>2103410568</td>\n",
       "      <td>['bubble nest', 'nest', 'mixing', 'bubble', 'p...</td>\n",
       "      <td>['Biology Letters']</td>\n",
       "      <td>[2346835213, 2098042950]</td>\n",
       "      <td>[2130285640, 2066345165, 2054319467, 204771406...</td>\n",
       "      <td>['building', 'home', 'foam', 'tungara', 'frog'...</td>\n",
       "      <td>['frogs', 'build', 'foam', 'nests', 'floating'...</td>\n",
       "      <td>0</td>\n",
       "      <td>10.1098/RSBL.2009.0934</td>\n",
       "      <td>[\"Weird Nature: An Astonishing Exploration of ...</td>\n",
       "      <td>['physically_assemble/disassemble', 'protect_f...</td>\n",
       "      <td>['physically_assemble_structure', 'protect_fro...</td>\n",
       "      <td>['protect_from_animals', 'protect_from_loss_of...</td>\n",
       "      <td>https://royalsocietypublishing.org/doi/10.1098...</td>\n",
       "      <td>Y</td>\n",
       "      <td>https://royalsocietypublishing.org/doi/10.1098...</td>\n",
       "      <td>True</td>\n",
       "      <td>Frogs that build foam nests floating on water ...</td>\n",
       "      <td>Building a home from foam—túngara frog foam ne...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2138292607</td>\n",
       "      <td>['sunset', 'earth s magnetic field', 'compass'...</td>\n",
       "      <td>['Proceedings of the National Academy of Scien...</td>\n",
       "      <td>[2132083079, 2425702268, 2552946098]</td>\n",
       "      <td>[1493129647, 2037761037, 1984592609, 213699427...</td>\n",
       "      <td>['nocturnal', 'mammal', 'greater', 'mouse', 'e...</td>\n",
       "      <td>['evidence', 'suggests', 'bats', 'detect', 'ge...</td>\n",
       "      <td>1</td>\n",
       "      <td>10.1073/PNAS.0912477107</td>\n",
       "      <td>['Proceedings of the National Academy of Scien...</td>\n",
       "      <td>['sense_send_or_process_information']</td>\n",
       "      <td>['sense_signals/environmental_cues']</td>\n",
       "      <td>['sense_spatial_awareness/balance/orientation']</td>\n",
       "      <td>https://www.pnas.org/content/107/15/6941</td>\n",
       "      <td>Y</td>\n",
       "      <td>https://www.pnas.org/content/107/15/6941.full.pdf</td>\n",
       "      <td>True</td>\n",
       "      <td>Recent evidence suggests that bats can detect ...</td>\n",
       "      <td>A nocturnal mammal, the greater mouse-eared ba...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2005539166</td>\n",
       "      <td>['sepia mestus', 'optomotor response', 'cuttle...</td>\n",
       "      <td>['The Journal of Experimental Biology']</td>\n",
       "      <td>[2163942483, 3088803717]</td>\n",
       "      <td>[2035108601, 2155571491, 2159857711, 207521876...</td>\n",
       "      <td>['polarization', 'sensitivity', 'two', 'specie...</td>\n",
       "      <td>['existence', 'polarization', 'sensitivity', '...</td>\n",
       "      <td>2</td>\n",
       "      <td>10.1242/JEB.042937</td>\n",
       "      <td>['The Journal of Experimental Biology', 'Curre...</td>\n",
       "      <td>['sense_send_or_process_information']</td>\n",
       "      <td>['sense_signals/environmental_cues']</td>\n",
       "      <td>['sense_light_in_the_non-visible_spectrum', 's...</td>\n",
       "      <td>https://jeb.biologists.org/content/213/19/3364</td>\n",
       "      <td>Y</td>\n",
       "      <td>https://journals.biologists.com/jeb/article-pd...</td>\n",
       "      <td>True</td>\n",
       "      <td>SUMMARY The existence of polarization sensitiv...</td>\n",
       "      <td>Polarization sensitivity in two species of cut...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>2151557512</td>\n",
       "      <td>['energy source', 'cellulase', 'archaea', 'cel...</td>\n",
       "      <td>['Nature Communications']</td>\n",
       "      <td>[3146712807, 3146712807, 2189911855, 265906908...</td>\n",
       "      <td>[2141885858, 2120772351, 2108929776, 212024499...</td>\n",
       "      <td>['identification', 'characterization', 'multid...</td>\n",
       "      <td>['archaea', 'microorganisms', 'wide', 'range',...</td>\n",
       "      <td>3</td>\n",
       "      <td>10.1038/NCOMMS1373</td>\n",
       "      <td>['Nature Communications']</td>\n",
       "      <td>['chemically_modify_or_change_energy_state', '...</td>\n",
       "      <td>['chemically_break_down', 'protect_from_non-li...</td>\n",
       "      <td>['chemically_break_down_organic_compounds', 'p...</td>\n",
       "      <td>https://www.nature.com/articles/ncomms1373</td>\n",
       "      <td>Y</td>\n",
       "      <td>https://www.nature.com/articles/ncomms1373.pdf</td>\n",
       "      <td>True</td>\n",
       "      <td>Despite extensive studies on microbial and enz...</td>\n",
       "      <td>Identification and characterization of a multi...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>2160542693</td>\n",
       "      <td>['galactan', 'coralline algae', 'cell wall', '...</td>\n",
       "      <td>['Journal of Phycology']</td>\n",
       "      <td>[1790079306, 2557664461, 189902495, 2140516860]</td>\n",
       "      <td>[2001396309, 2057007315, 1968834637, 212515227...</td>\n",
       "      <td>['differences', 'polysaccharide', 'structure',...</td>\n",
       "      <td>['articulated', 'coralline', 'calliarthron', '...</td>\n",
       "      <td>4</td>\n",
       "      <td>10.1111/J.1529-8817.2010.00828.X</td>\n",
       "      <td>['The Journal of Experimental Biology', 'Natur...</td>\n",
       "      <td>['maintain_structural_integrity']</td>\n",
       "      <td>['manage_structural_forces']</td>\n",
       "      <td>['manage_shear', 'manage_stress/strain']</td>\n",
       "      <td>https://onlinelibrary.wiley.com/doi/abs/10.111...</td>\n",
       "      <td>Y</td>\n",
       "      <td></td>\n",
       "      <td>False</td>\n",
       "      <td>The articulated coralline Calliarthron cheilos...</td>\n",
       "      <td>DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        paper                                                mag  \\\n",
       "0  2103410568  ['bubble nest', 'nest', 'mixing', 'bubble', 'p...   \n",
       "1  2138292607  ['sunset', 'earth s magnetic field', 'compass'...   \n",
       "2  2005539166  ['sepia mestus', 'optomotor response', 'cuttle...   \n",
       "3  2151557512  ['energy source', 'cellulase', 'archaea', 'cel...   \n",
       "4  2160542693  ['galactan', 'coralline algae', 'cell wall', '...   \n",
       "\n",
       "                                           venue_mag  \\\n",
       "0                                ['Biology Letters']   \n",
       "1  ['Proceedings of the National Academy of Scien...   \n",
       "2            ['The Journal of Experimental Biology']   \n",
       "3                          ['Nature Communications']   \n",
       "4                           ['Journal of Phycology']   \n",
       "\n",
       "                                              author  \\\n",
       "0                           [2346835213, 2098042950]   \n",
       "1               [2132083079, 2425702268, 2552946098]   \n",
       "2                           [2163942483, 3088803717]   \n",
       "3  [3146712807, 3146712807, 2189911855, 265906908...   \n",
       "4    [1790079306, 2557664461, 189902495, 2140516860]   \n",
       "\n",
       "                                           reference  \\\n",
       "0  [2130285640, 2066345165, 2054319467, 204771406...   \n",
       "1  [1493129647, 2037761037, 1984592609, 213699427...   \n",
       "2  [2035108601, 2155571491, 2159857711, 207521876...   \n",
       "3  [2141885858, 2120772351, 2108929776, 212024499...   \n",
       "4  [2001396309, 2057007315, 1968834637, 212515227...   \n",
       "\n",
       "                                               title  \\\n",
       "0  ['building', 'home', 'foam', 'tungara', 'frog'...   \n",
       "1  ['nocturnal', 'mammal', 'greater', 'mouse', 'e...   \n",
       "2  ['polarization', 'sensitivity', 'two', 'specie...   \n",
       "3  ['identification', 'characterization', 'multid...   \n",
       "4  ['differences', 'polysaccharide', 'structure',...   \n",
       "\n",
       "                                            abstract  petalID  \\\n",
       "0  ['frogs', 'build', 'foam', 'nests', 'floating'...        0   \n",
       "1  ['evidence', 'suggests', 'bats', 'detect', 'ge...        1   \n",
       "2  ['existence', 'polarization', 'sensitivity', '...        2   \n",
       "3  ['archaea', 'microorganisms', 'wide', 'range',...        3   \n",
       "4  ['articulated', 'coralline', 'calliarthron', '...        4   \n",
       "\n",
       "                                doi  \\\n",
       "0            10.1098/RSBL.2009.0934   \n",
       "1           10.1073/PNAS.0912477107   \n",
       "2                10.1242/JEB.042937   \n",
       "3                10.1038/NCOMMS1373   \n",
       "4  10.1111/J.1529-8817.2010.00828.X   \n",
       "\n",
       "                                               venue  \\\n",
       "0  [\"Weird Nature: An Astonishing Exploration of ...   \n",
       "1  ['Proceedings of the National Academy of Scien...   \n",
       "2  ['The Journal of Experimental Biology', 'Curre...   \n",
       "3                          ['Nature Communications']   \n",
       "4  ['The Journal of Experimental Biology', 'Natur...   \n",
       "\n",
       "                                              level1  \\\n",
       "0  ['physically_assemble/disassemble', 'protect_f...   \n",
       "1              ['sense_send_or_process_information']   \n",
       "2              ['sense_send_or_process_information']   \n",
       "3  ['chemically_modify_or_change_energy_state', '...   \n",
       "4                  ['maintain_structural_integrity']   \n",
       "\n",
       "                                              level2  \\\n",
       "0  ['physically_assemble_structure', 'protect_fro...   \n",
       "1               ['sense_signals/environmental_cues']   \n",
       "2               ['sense_signals/environmental_cues']   \n",
       "3  ['chemically_break_down', 'protect_from_non-li...   \n",
       "4                       ['manage_structural_forces']   \n",
       "\n",
       "                                              level3  \\\n",
       "0  ['protect_from_animals', 'protect_from_loss_of...   \n",
       "1    ['sense_spatial_awareness/balance/orientation']   \n",
       "2  ['sense_light_in_the_non-visible_spectrum', 's...   \n",
       "3  ['chemically_break_down_organic_compounds', 'p...   \n",
       "4           ['manage_shear', 'manage_stress/strain']   \n",
       "\n",
       "                                                 url isBiomimicry  \\\n",
       "0  https://royalsocietypublishing.org/doi/10.1098...            Y   \n",
       "1           https://www.pnas.org/content/107/15/6941            Y   \n",
       "2     https://jeb.biologists.org/content/213/19/3364            Y   \n",
       "3         https://www.nature.com/articles/ncomms1373            Y   \n",
       "4  https://onlinelibrary.wiley.com/doi/abs/10.111...            Y   \n",
       "\n",
       "                                         fullDocLink isOpenAccess  \\\n",
       "0  https://royalsocietypublishing.org/doi/10.1098...         True   \n",
       "1  https://www.pnas.org/content/107/15/6941.full.pdf         True   \n",
       "2  https://journals.biologists.com/jeb/article-pd...         True   \n",
       "3     https://www.nature.com/articles/ncomms1373.pdf         True   \n",
       "4                                                           False   \n",
       "\n",
       "                                       abstract_full  \\\n",
       "0  Frogs that build foam nests floating on water ...   \n",
       "1  Recent evidence suggests that bats can detect ...   \n",
       "2  SUMMARY The existence of polarization sensitiv...   \n",
       "3  Despite extensive studies on microbial and enz...   \n",
       "4  The articulated coralline Calliarthron cheilos...   \n",
       "\n",
       "                                          title_full  \n",
       "0  Building a home from foam—túngara frog foam ne...  \n",
       "1  A nocturnal mammal, the greater mouse-eared ba...  \n",
       "2  Polarization sensitivity in two species of cut...  \n",
       "3  Identification and characterization of a multi...  \n",
       "4  DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = pd.DataFrame(docs)\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f437730d",
   "metadata": {},
   "source": [
    "## Pre-Processing"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "3dc992c0",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "paper            1456\n",
       "mag              1456\n",
       "venue_mag        1456\n",
       "author           1456\n",
       "reference        1456\n",
       "title            1456\n",
       "abstract         1456\n",
       "petalID          1456\n",
       "doi              1456\n",
       "venue            1456\n",
       "level1           1456\n",
       "level2           1456\n",
       "level3           1456\n",
       "url              1456\n",
       "isBiomimicry     1456\n",
       "fullDocLink      1456\n",
       "isOpenAccess     1456\n",
       "abstract_full    1438\n",
       "title_full       1456\n",
       "dtype: int64"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = df[df['isBiomimicry'] == 'Y']\n",
    "df.count()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "46168681",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "paper            1131\n",
       "mag              1131\n",
       "venue_mag        1131\n",
       "author           1131\n",
       "reference        1131\n",
       "title            1131\n",
       "abstract         1131\n",
       "petalID          1131\n",
       "doi              1131\n",
       "venue            1131\n",
       "level1           1131\n",
       "level2           1131\n",
       "level3           1131\n",
       "url              1131\n",
       "isBiomimicry     1131\n",
       "fullDocLink      1131\n",
       "isOpenAccess     1131\n",
       "abstract_full    1113\n",
       "title_full       1131\n",
       "dtype: int64"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df = df[df['level1'] != '[]' ]\n",
    "df.count()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "648b250d",
   "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>abstract_full</th>\n",
       "      <th>title_full</th>\n",
       "      <th>level1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Frogs that build foam nests floating on water ...</td>\n",
       "      <td>Building a home from foam—túngara frog foam ne...</td>\n",
       "      <td>['physically_assemble/disassemble', 'protect_f...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Recent evidence suggests that bats can detect ...</td>\n",
       "      <td>A nocturnal mammal, the greater mouse-eared ba...</td>\n",
       "      <td>['sense_send_or_process_information']</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>SUMMARY The existence of polarization sensitiv...</td>\n",
       "      <td>Polarization sensitivity in two species of cut...</td>\n",
       "      <td>['sense_send_or_process_information']</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Despite extensive studies on microbial and enz...</td>\n",
       "      <td>Identification and characterization of a multi...</td>\n",
       "      <td>['chemically_modify_or_change_energy_state', '...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>The articulated coralline Calliarthron cheilos...</td>\n",
       "      <td>DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...</td>\n",
       "      <td>['maintain_structural_integrity']</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                       abstract_full  \\\n",
       "0  Frogs that build foam nests floating on water ...   \n",
       "1  Recent evidence suggests that bats can detect ...   \n",
       "2  SUMMARY The existence of polarization sensitiv...   \n",
       "3  Despite extensive studies on microbial and enz...   \n",
       "4  The articulated coralline Calliarthron cheilos...   \n",
       "\n",
       "                                          title_full  \\\n",
       "0  Building a home from foam—túngara frog foam ne...   \n",
       "1  A nocturnal mammal, the greater mouse-eared ba...   \n",
       "2  Polarization sensitivity in two species of cut...   \n",
       "3  Identification and characterization of a multi...   \n",
       "4  DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...   \n",
       "\n",
       "                                              level1  \n",
       "0  ['physically_assemble/disassemble', 'protect_f...  \n",
       "1              ['sense_send_or_process_information']  \n",
       "2              ['sense_send_or_process_information']  \n",
       "3  ['chemically_modify_or_change_energy_state', '...  \n",
       "4                  ['maintain_structural_integrity']  "
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_cleaned = df[['abstract_full', 'title_full', 'level1']]\n",
    "df_cleaned.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "a0c00665",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "<ipython-input-6-3b2ff8dfb6cb>:1: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  df_cleaned['abstract'] = df_cleaned['title_full'] + df_cleaned['abstract_full']\n",
      "<ipython-input-6-3b2ff8dfb6cb>:2: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy\n",
      "  df_cleaned['title'] = df_cleaned['title_full']\n"
     ]
    },
    {
     "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>abstract</th>\n",
       "      <th>title</th>\n",
       "      <th>level1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Building a home from foam—túngara frog foam ne...</td>\n",
       "      <td>Building a home from foam—túngara frog foam ne...</td>\n",
       "      <td>['physically_assemble/disassemble', 'protect_f...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>A nocturnal mammal, the greater mouse-eared ba...</td>\n",
       "      <td>A nocturnal mammal, the greater mouse-eared ba...</td>\n",
       "      <td>['sense_send_or_process_information']</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Polarization sensitivity in two species of cut...</td>\n",
       "      <td>Polarization sensitivity in two species of cut...</td>\n",
       "      <td>['sense_send_or_process_information']</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Identification and characterization of a multi...</td>\n",
       "      <td>Identification and characterization of a multi...</td>\n",
       "      <td>['chemically_modify_or_change_energy_state', '...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...</td>\n",
       "      <td>DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...</td>\n",
       "      <td>['maintain_structural_integrity']</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                            abstract  \\\n",
       "0  Building a home from foam—túngara frog foam ne...   \n",
       "1  A nocturnal mammal, the greater mouse-eared ba...   \n",
       "2  Polarization sensitivity in two species of cut...   \n",
       "3  Identification and characterization of a multi...   \n",
       "4  DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...   \n",
       "\n",
       "                                               title  \\\n",
       "0  Building a home from foam—túngara frog foam ne...   \n",
       "1  A nocturnal mammal, the greater mouse-eared ba...   \n",
       "2  Polarization sensitivity in two species of cut...   \n",
       "3  Identification and characterization of a multi...   \n",
       "4  DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...   \n",
       "\n",
       "                                              level1  \n",
       "0  ['physically_assemble/disassemble', 'protect_f...  \n",
       "1              ['sense_send_or_process_information']  \n",
       "2              ['sense_send_or_process_information']  \n",
       "3  ['chemically_modify_or_change_energy_state', '...  \n",
       "4                  ['maintain_structural_integrity']  "
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_cleaned['abstract'] = df_cleaned['title_full'] + df_cleaned['abstract_full']\n",
    "df_cleaned['title'] = df_cleaned['title_full']\n",
    "df_cleaned = df_cleaned[['abstract', 'title', 'level1']]\n",
    "df_cleaned.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "ce346db3",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Changing 'level1' column from string to list\n",
    "for i in range(len(df_cleaned['level1'])):\n",
    "    x = df_cleaned['level1'][i]\n",
    "    x = x.strip('][').split(', ')\n",
    "    for j in range(len(x)):\n",
    "        x[j] = x[j].replace(\"'\", \"\")\n",
    "    df_cleaned['level1'][i] = x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "6ce723b4",
   "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>abstract</th>\n",
       "      <th>title</th>\n",
       "      <th>level1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Building a home from foam—túngara frog foam ne...</td>\n",
       "      <td>Building a home from foam—túngara frog foam ne...</td>\n",
       "      <td>[physically_assemble/disassemble, protect_from...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>A nocturnal mammal, the greater mouse-eared ba...</td>\n",
       "      <td>A nocturnal mammal, the greater mouse-eared ba...</td>\n",
       "      <td>[sense_send_or_process_information]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Polarization sensitivity in two species of cut...</td>\n",
       "      <td>Polarization sensitivity in two species of cut...</td>\n",
       "      <td>[sense_send_or_process_information]</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Identification and characterization of a multi...</td>\n",
       "      <td>Identification and characterization of a multi...</td>\n",
       "      <td>[chemically_modify_or_change_energy_state, pro...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...</td>\n",
       "      <td>DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...</td>\n",
       "      <td>[maintain_structural_integrity]</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                            abstract  \\\n",
       "0  Building a home from foam—túngara frog foam ne...   \n",
       "1  A nocturnal mammal, the greater mouse-eared ba...   \n",
       "2  Polarization sensitivity in two species of cut...   \n",
       "3  Identification and characterization of a multi...   \n",
       "4  DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...   \n",
       "\n",
       "                                               title  \\\n",
       "0  Building a home from foam—túngara frog foam ne...   \n",
       "1  A nocturnal mammal, the greater mouse-eared ba...   \n",
       "2  Polarization sensitivity in two species of cut...   \n",
       "3  Identification and characterization of a multi...   \n",
       "4  DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...   \n",
       "\n",
       "                                              level1  \n",
       "0  [physically_assemble/disassemble, protect_from...  \n",
       "1                [sense_send_or_process_information]  \n",
       "2                [sense_send_or_process_information]  \n",
       "3  [chemically_modify_or_change_energy_state, pro...  \n",
       "4                    [maintain_structural_integrity]  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_cleaned.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "08729fa0",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package stopwords to\n",
      "[nltk_data]     C:\\Users\\aeltabak\\AppData\\Roaming\\nltk_data...\n",
      "[nltk_data]   Package stopwords is already up-to-date!\n"
     ]
    }
   ],
   "source": [
    "# Importing dependencies\n",
    "import nltk\n",
    "nltk.download('stopwords')\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.feature_extraction.text import TfidfVectorizer\n",
    "from sklearn.feature_extraction.text import CountVectorizer\n",
    "from sklearn.decomposition import PCA\n",
    "from sklearn.multiclass import OneVsRestClassifier\n",
    "from nltk.corpus import stopwords\n",
    "stop_words = set(stopwords.words('english'))\n",
    "from sklearn.svm import LinearSVC\n",
    "from sklearn.pipeline import Pipeline\n",
    "from sklearn.metrics import accuracy_score, roc_curve, auc, hamming_loss\n",
    "from sklearn.preprocessing import MultiLabelBinarizer\n",
    "from sklearn.calibration import CalibratedClassifierCV"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "f4c03f56",
   "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>abstract</th>\n",
       "      <th>title</th>\n",
       "      <th>attach</th>\n",
       "      <th>change_size_or_color</th>\n",
       "      <th>chemically_modify_or_change_energy_state</th>\n",
       "      <th>maintain_ecological_community</th>\n",
       "      <th>maintain_structural_integrity</th>\n",
       "      <th>move</th>\n",
       "      <th>physically_assemble/disassemble</th>\n",
       "      <th>process_resources</th>\n",
       "      <th>protect_from_harm</th>\n",
       "      <th>sense_send_or_process_information</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Building a home from foam—túngara frog foam ne...</td>\n",
       "      <td>Building a home from foam—túngara frog foam ne...</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>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>A nocturnal mammal, the greater mouse-eared ba...</td>\n",
       "      <td>A nocturnal mammal, the greater mouse-eared ba...</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>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>Polarization sensitivity in two species of cut...</td>\n",
       "      <td>Polarization sensitivity in two species of cut...</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>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>Identification and characterization of a multi...</td>\n",
       "      <td>Identification and characterization of a multi...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...</td>\n",
       "      <td>DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</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",
       "</div>"
      ],
      "text/plain": [
       "                                            abstract  \\\n",
       "0  Building a home from foam—túngara frog foam ne...   \n",
       "1  A nocturnal mammal, the greater mouse-eared ba...   \n",
       "2  Polarization sensitivity in two species of cut...   \n",
       "3  Identification and characterization of a multi...   \n",
       "4  DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...   \n",
       "\n",
       "                                               title  attach  \\\n",
       "0  Building a home from foam—túngara frog foam ne...       0   \n",
       "1  A nocturnal mammal, the greater mouse-eared ba...       0   \n",
       "2  Polarization sensitivity in two species of cut...       0   \n",
       "3  Identification and characterization of a multi...       0   \n",
       "4  DIFFERENCES IN POLYSACCHARIDE STRUCTURE BETWEE...       0   \n",
       "\n",
       "   change_size_or_color  chemically_modify_or_change_energy_state  \\\n",
       "0                     0                                         0   \n",
       "1                     0                                         0   \n",
       "2                     0                                         0   \n",
       "3                     0                                         1   \n",
       "4                     0                                         0   \n",
       "\n",
       "   maintain_ecological_community  maintain_structural_integrity  move  \\\n",
       "0                              0                              0     0   \n",
       "1                              0                              0     0   \n",
       "2                              0                              0     0   \n",
       "3                              0                              0     0   \n",
       "4                              0                              1     0   \n",
       "\n",
       "   physically_assemble/disassemble  process_resources  protect_from_harm  \\\n",
       "0                                1                  0                  1   \n",
       "1                                0                  0                  0   \n",
       "2                                0                  0                  0   \n",
       "3                                0                  0                  1   \n",
       "4                                0                  0                  0   \n",
       "\n",
       "   sense_send_or_process_information  \n",
       "0                                  0  \n",
       "1                                  1  \n",
       "2                                  1  \n",
       "3                                  0  \n",
       "4                                  0  "
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# convert general category into label columns\n",
    "mlb = MultiLabelBinarizer()\n",
    "labels = mlb.fit_transform(df_cleaned.level1)\n",
    "\n",
    "# concatenate with the abstracts\n",
    "df = pd.concat([df_cleaned[['abstract', 'title']], pd.DataFrame(labels)], axis=1)\n",
    "df.columns = ['abstract', 'title'] + list(mlb.classes_)\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "6b5b071d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Removed following categories from training : []\n"
     ]
    }
   ],
   "source": [
    "# sample and keep columns that have at least 1 positive example\n",
    "sample_df = df.sample(frac = 1, random_state = 4)\n",
    "keep = sample_df.iloc[:,2:].apply(sum) > 1\n",
    "sample_df = pd.concat([sample_df.iloc[:,:2],sample_df.iloc[:,2:].iloc[:,keep.values]], axis = 1)\n",
    "\n",
    "categories = sample_df.columns[2:]\n",
    "\n",
    "# removed categories\n",
    "print('Removed following categories from training : {}'.format(str(keep[~keep].index.to_list())))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "14352e9f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAHKCAYAAAD4lAtVAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAAsTAAALEwEAmpwYAABHxElEQVR4nO3deZxcVZn/8c+XRTZZggRE2QKDaEDWsAj8FEEEWWQRFBRF0EFHFNBxAZcBHRnUERFxQEHAKCBGEUSUXRZBEBO2AIEBASXCQFSWyJ7w/f1xTiXVleqFVNW51bef9+tVr+57q6rPk3TXU6fOPec5sk0IIYR6WaTqAEIIIXRfJPcQQqihSO4hhFBDkdxDCKGGIrmHEEINLVZ1AAArrbSS11prrarDCCGEUWXatGl/sz2+3X19kdzXWmstpk6dWnUYIYQwqkj682D3xbBMCCHUUCT3EEKooUjuIYRQQ5HcQwihhoZN7pKWlHSTpNsk3Snpy/n8MZL+KunWfNul6TlHSbpP0j2SdurlPyCEEMKCRjJb5nlge9v/lLQ4cJ2ki/N9J9j+ZvODJU0E9gPWB14DXCHpdbbndjPwEEIIgxu25+7kn/lw8XwbqpTkHsC5tp+3/QBwH7BFx5GGEEIYsRGNuUtaVNKtwGPA5bb/kO/6uKTbJZ0haVw+91rgoaanz8znWn/mIZKmSpo6a9ashf8XhBBCWMCIkrvtubY3BlYDtpC0AXAKsA6wMfAIcHx+uNr9iDY/81Tbk2xPGj++7QKrEEIIC+llrVC1/YSkq4Gdm8faJZ0GXJQPZwKrNz1tNeDhDuMMIYQRW+vIX3f0/Ae/tmuXIqnOSGbLjJe0Qv5+KeBtwN2SVm162F7AHfn7C4H9JC0haQKwLnBTV6MOIYQwpJH03FcFJktalPRmMMX2RZJ+LGlj0pDLg8BHAGzfKWkKcBcwBzg0ZsqEEEJZwyZ327cDm7Q5//4hnnMscGxnoYUQQlhYsUI1hBBqKJJ7CCHUUCT3EEKooUjuIYRQQ5HcQwihhiK5hxBCDUVyDyGEGorkHkIINRTJPYQQaiiSewgh1FAk9xBCqKFI7iGEUEOR3EMIoYYiuYcQQg1Fcg8hhBqK5B5CCDUUyT2EEGooknsIIdRQJPcQQqihSO4hhFBDkdxDCKGGhk3ukpaUdJOk2yTdKenL+fyKki6XdG/+Oq7pOUdJuk/SPZJ26uU/IIQQwoJG0nN/Htje9kbAxsDOkrYCjgSutL0ucGU+RtJEYD9gfWBn4GRJi/Yg9hBCCIMYNrk7+Wc+XDzfDOwBTM7nJwN75u/3AM61/bztB4D7gC26GXQIIYShjWjMXdKikm4FHgMut/0HYBXbjwDkryvnh78WeKjp6TPzudafeYikqZKmzpo1q4N/QgghhFYjSu6259reGFgN2ELSBkM8XO1+RJufeartSbYnjR8/fkTBhhBCGJmXNVvG9hPA1aSx9EclrQqQvz6WHzYTWL3paasBD3caaAghhJEbyWyZ8ZJWyN8vBbwNuBu4EDgwP+xA4Jf5+wuB/SQtIWkCsC5wU5fjDiGEMITFRvCYVYHJecbLIsAU2xdJugGYIulDwF+AfQFs3ylpCnAXMAc41Pbc3oQfQgihnWGTu+3bgU3anP87sMMgzzkWOLbj6EIIISyUWKEaQgg1FMk9hBBqKJJ7CCHUUCT3EEKooUjuIYRQQ5HcQwihhiK5hxBCDUVyDyGEGorkHkIINRTJPYQQaiiSewgh1FAk9xBCqKFI7iGEUEOR3EMIoYYiuYcQQg1Fcg8hhBqK5B5CCDUUyT2EEGooknsIIdRQJPcQQqihYZO7pNUlXSVphqQ7JR2ezx8j6a+Sbs23XZqec5Sk+yTdI2mnXv4DQgghLGixETxmDvDvtm+WtCwwTdLl+b4TbH+z+cGSJgL7AesDrwGukPQ623O7GXgIIYTBDdtzt/2I7Zvz97OBGcBrh3jKHsC5tp+3/QBwH7BFN4INIYQwMi9rzF3SWsAmwB/yqY9Lul3SGZLG5XOvBR5qetpM2rwZSDpE0lRJU2fNmvXyIw8hhDCoESd3Sa8EzgOOsP0UcAqwDrAx8AhwfOOhbZ7uBU7Yp9qeZHvS+PHjX27cIYQQhjCi5C5pcVJiP9v2LwBsP2p7ru2XgNOYP/QyE1i96emrAQ93L+QQQgjDGclsGQGnAzNsf6vp/KpND9sLuCN/fyGwn6QlJE0A1gVu6l7IIYQQhjOS2TLbAO8Hpku6NZ/7PLC/pI1JQy4PAh8BsH2npCnAXaSZNofGTJkQQihr2ORu+zraj6P/ZojnHAsc20FcIYQQOhArVEMIoYYiuYcQQg1Fcg8hhBqK5B5CCDUUyT2EEGooknsIIdRQJPcQQqihSO4hhFBDkdxDCKGGRlJ+IIQQRmStI3/d8c948Gu7diGSED33EEKooUjuIYRQQ5HcQwihhiK5hxBCDUVyDyGEGorkHkIINRTJPYQQaiiSewgh1FAk9xBCqKFI7iGEUEOR3EMIoYaGTe6SVpd0laQZku6UdHg+v6KkyyXdm7+Oa3rOUZLuk3SPpJ16+Q8IIYSwoJH03OcA/277DcBWwKGSJgJHAlfaXhe4Mh+T79sPWB/YGThZ0qK9CD6EEEJ7wyZ324/Yvjl/PxuYAbwW2AOYnB82Gdgzf78HcK7t520/ANwHbNHluEMIIQzhZY25S1oL2AT4A7CK7UcgvQEAK+eHvRZ4qOlpM/O51p91iKSpkqbOmjVrIUIPIYQwmBEnd0mvBM4DjrD91FAPbXPOC5ywT7U9yfak8ePHjzSMEEIIIzCi5C5pcVJiP9v2L/LpRyWtmu9fFXgsn58JrN709NWAh7sTbgghhJEYyWwZAacDM2x/q+muC4ED8/cHAr9sOr+fpCUkTQDWBW7qXsghhBCGM5Jt9rYB3g9Ml3RrPvd54GvAFEkfAv4C7Atg+05JU4C7SDNtDrU9t9uBhxBCGNywyd32dbQfRwfYYZDnHAsc20FcIYQQOhArVEMIoYYiuYcQQg1Fcg8hhBqK5B5CCDU0ktkyIYQQXqa1jvx1xz/jwa/tutDPjZ57CCHUUCT3EEKooUjuIYRQQ5HcQwihhiK5hxBCDUVyDyGEGorkHkIINRTJPYQQaigWMYVRrdOFIp0sEgmhn0XPPYQQaiiSewgh1FAMy4TQoapriITQTvTcQwihhiK5hxBCDUVyDyGEGorkHkIINTRscpd0hqTHJN3RdO4YSX+VdGu+7dJ031GS7pN0j6SdehV4CCGEwY2k5/5DYOc250+wvXG+/QZA0kRgP2D9/JyTJS3arWBDCCGMzLDJ3fa1wD9G+PP2AM61/bztB4D7gC06iC+EEMJC6GTM/eOSbs/DNuPyudcCDzU9ZmY+twBJh0iaKmnqrFmzOggjhBBCq4VN7qcA6wAbA48Ax+fzavNYt/sBtk+1Pcn2pPHjxy9kGCGEENpZqORu+1Hbc22/BJzG/KGXmcDqTQ9dDXi4sxBDCCG8XAuV3CWt2nS4F9CYSXMhsJ+kJSRNANYFbuosxBBCCC/XsLVlJP0E2A5YSdJM4GhgO0kbk4ZcHgQ+AmD7TklTgLuAOcChtuf2JPIQQgiDGja5296/zenTh3j8scCxnQQVQgihM7FCNYQQaiiSewgh1FAk9xBCqKFI7iGEUEOR3EMIoYZim71RqNNt3WJLtxDqL3ruIYRQQ5HcQwihhiK5hxBCDUVyDyGEGorkHkIINRSzZcJC6XTGDsSsnRB6KXruIYRQQ5HcQwihhiK5hxBCDUVyDyGEGorkHkIINRTJPYQQaiiSewgh1FAk9xBCqKFI7iGEUEPDJndJZ0h6TNIdTedWlHS5pHvz13FN9x0l6T5J90jaqVeBhxBCGNxIeu4/BHZuOXckcKXtdYEr8zGSJgL7Aevn55wsadGuRRtCCGFEhk3utq8F/tFyeg9gcv5+MrBn0/lzbT9v+wHgPmCL7oQaQghhpBZ2zH0V248A5K8r5/OvBR5qetzMfG4Bkg6RNFXS1FmzZi1kGCGEENrp9gVVtTnndg+0fartSbYnjR8/vsthhBDC2LawJX8flbSq7UckrQo8ls/PBFZvetxqwMOdBNgQm0KHEMLILWzP/ULgwPz9gcAvm87vJ2kJSROAdYGbOgsxhBDCyzVsz13ST4DtgJUkzQSOBr4GTJH0IeAvwL4Atu+UNAW4C5gDHGp7bo9iDyGEMIhhk7vt/Qe5a4dBHn8scGwnQYUQQuhMrFANIYQaij1UQ6iJmHQQmkVyfxliU+gQwmgRwzIhhFBDkdxDCKGGIrmHEEINRXIPIYQaiuQeQgg1FMk9hBBqKJJ7CCHUUCT3EEKooUjuIYRQQ5HcQwihhiK5hxBCDUVyDyGEGorkHkIINRTJPYQQaiiSewgh1FAk9xBCqKFI7iGEUEOR3EMIoYY62mZP0oPAbGAuMMf2JEkrAj8F1gIeBN5t+/HOwgwhhPBydKPn/lbbG9uelI+PBK60vS5wZT4OIYRQUC+GZfYAJufvJwN79qCNEEIIQ+g0uRu4TNI0SYfkc6vYfgQgf1253RMlHSJpqqSps2bN6jCMEEIIzToacwe2sf2wpJWByyXdPdIn2j4VOBVg0qRJ7jCOEEIITTrqudt+OH99DDgf2AJ4VNKqAPnrY50GGUII4eVZ6OQuaRlJyza+B94O3AFcCByYH3Yg8MtOgwwhhPDydDIsswpwvqTGzznH9iWS/ghMkfQh4C/Avp2HGUII4eVY6ORu+35gozbn/w7s0ElQIYQQOhMrVEMIoYYiuYcQQg1Fcg8hhBqK5B5CCDUUyT2EEGooknsIIdRQJPcQQqihSO4hhFBDkdxDCKGGIrmHEEINRXIPIYQaiuQeQgg1FMk9hBBqKJJ7CCHUUCT3EEKooUjuIYRQQ5HcQwihhiK5hxBCDUVyDyGEGorkHkIINRTJPYQQaqhnyV3SzpLukXSfpCN71U4IIYQF9SS5S1oU+B/gHcBEYH9JE3vRVgghhAX1que+BXCf7fttvwCcC+zRo7ZCCCG0kO3u/1BpH2Bn2x/Ox+8HtrT98abHHAIckg/XA+7psNmVgL91+DO6oR/i6IcYoD/iiBjm64c4+iEG6I84uhHDmrbHt7tjsQ5/8GDU5tyAdxHbpwKndq1BaartSd36eaM5jn6IoV/iiBj6K45+iKFf4uh1DL0alpkJrN50vBrwcI/aCiGE0KJXyf2PwLqSJkh6BbAfcGGP2gohhNCiJ8MytudI+jhwKbAocIbtO3vRVpOuDfF0qB/i6IcYoD/iiBjm64c4+iEG6I84ehpDTy6ohhBCqFasUA0hhBqK5B5CCDUUyT2EEGooknvoKkkbVB1DP5G0TNUx9AtJi0haruo4xopI7h2QtKikK6qOA0DJAZL+Ix+vIWmLCkL5nqSbJH1M0goVtI+k8yTtKqmyv29JW0u6C5iRjzeSdHJFsWwr6aD8/XhJEwq2fY6k5fKb3F3APZI+U6r9llh2k3SLpH9IekrSbElPFY5hCUnvlfR5Sf/RuPWirVGd3CXtLeleSU9W8cuyPRd4RtLypdocwsnAm4D98/FsUvG2omxvC7yPtIhtan5x71g4jFOA9wL3SvqapNcXbh/gBGAn4O8Atm8D3lw6CElHA58DjsqnFgfOKhjCRNtPAXsCvwHWAN5fsP1m3wYOBF5leznby9ou/Unil6Q6W3OAp5tuXder8gOlfAPY3faMCmN4Dpgu6XKafkm2Dyscx5a2N5V0S27/8byArDjb90r6IjAV+A6wiSQBn7f9iwLtXwFckd909wcul/QQcBpwlu0Xex1DjuOh9M+eZ26JdlvsBWwC3JxjeljSsgXbX1zS4qTk/l3bL0qqav71Q8Adrnb+92q2dy7R0GhP7o9WnNgBfp1vVXsxl1o2pI/fwEulg5C0IXAQsCtwOenN92ZJrwFuAHqe3HMcrwIOIPUSbwHOBrYl9dy2KxDCQ5K2BpzfZA8jD9EU9oJtNxJqBdcAvg88CNwGXCtpTaDoUEiTzwK/kXQN8HzjpO1vFYzh95LeaHt6rxsalYuYJO2dv30L8GrgAgb+sookkKZ4XgG8Lh/eU6pn2BLD+4D3AJsCk4F9gC/ZnlI4jmtJPeSf23625b732/5xgRh+Abwe+DHwQ9uPNN1XpGCUpJWAE4G3kQrpXQYcbvvvvW67JY5PA+sCOwLHAQcD59g+qWQcLTEtZntOBe1eBvwTmE5Tx8f2lwvGcBfwL8ADpJylFII37HpbozS5nznE3bZ9cMFYtiMl0wdJv6jVgQNtX1sqhqZYXg/skOO4sopPNZKOsP3tlnOH2z6xYAy72P5Ny7klbD8/2HN6EMN427NKtTeUfM3j7aS/i0ttX16w7VWA/wJeY/sdedOeN9k+vVQMTbH0QyXINdudt/3nrrc1GpN7P5E0DXiv7Xvy8euAn9jerHAcP7b9/uHOFYjjZtubtpy7xfYmFcewwLkex3AvqXf2U+A820+UarufSLoYOBP4gu2NJC0G3GL7jRXE8jXgt7YvK912SxwbAf8vH/4uX2zvutE+W2Zy83Q7SeMknVE4jMUbiR3A9v+SZiSUtn7zQR5/L/YGI2l/Sb8CJki6sOl2FXnGSIEYXi1pM2ApSZtI2jTftgOWLhFDg+11gS+Sfi83S7pI0gGl2m/MHGtzKz39b6U8NPgSpKKCVHNhGeBQ4BJJz1Y4FfJw0vWflfPtLEmf6EVbo/2C6obNPaI8Q6RYDzGbKul00vgupGmA00o1Luko4POkhPYU8zdKeYGyle9+DzxC2l3m+Kbzs4HbC8WwE/BB0v4BzRfJZpP+j4qyfRNwk6T/yvFMptA0RNslZ8QM5el8cbtxQXcr4MnSQSitedjZ9vWl227xIdLMtqcBJH2dNNGg69dARvWwjKTbgO1sP56PVwSuKfmRT9ISpB7BtqTEei1wcsnx3RzHcbaPGv6R9SfpXbbPqziG5UjTEPcD1gHOB6bYLvbG3xTLpqS/TwPX2b6lcNsnARsAdwDjgX1sl3rDb47lBttvKt1uSwzTgc1tP5ePlwT+2IucNdqT+wdIizN+nk/tCxxbYkZGP5I0jjQzYsnGuVIXdiVdZ3tbSbMZuKViYzZAzxeLSDrA9lmS/r0lBqDslDdJD5BmcU2xfUOpdtvE8R+k10VjBtmewM9sf7VgDIuR9kkWFc0my3F8mfQp8hdVzXWX9CnSdNzz86k9STO6vt31tkZzcgeQtD7wVubPELmrULvTaZNAGnoxtWkokj4MHE4akrgV2Aq4wfb2JeOokqSP2P6+0qrMBRSe8qY8v3zZ1LT/WartljhmAJs09RSXAm62/YZC7R8KnN0YPs0dkP1tFy/FkDsey5BWhz5HwY5HSxyNT1ICru3VJ6lRn9wBJK3MwN7qXwq02XZKU1MMXZ/aNJTGxz3gRtsb52mRX7b9npJx5FgWBVah6ZpOid9JU9uH2T6hRHtDxLEB6TrMiqQX8SzSFNk7CsdxMSmZPpGPVyCt0t2tUPu32t645VzR2VP9QNJytp/KQ8cLsP2Pbrc5qi+oSnon6eLda4DHgDVJqwDXH+p53dCcvPNc3s3z4U22H+t1+208Z/s5SY053XdLWq90EPnK/9HAo8xfKGKgyCcZ23Pz30WlyZ10MftTtq+CeeshTgW2LtG4pJNI/+/PA3cqlccwaTHTdSViyBZpfIrJcS0KVFIWI7df1dDlOcBupMkWCwxbAmt3u8FRndyB/yQNP1xhexNJb2V+4awiJL0b+G/gatIv6iRJn7H98yGf2H0zc6/sAlItlceBhwvHAGloaL3SKzFb/F7Sd0lzzJvr/dxcMIZlGok9t321yi79n5q/TmP++C6kv9OSLgOmSPoeKYl9FLikcAzA4EOXQM+HLhuflGyXq8g5modllFec5Vkzm9h+SdJNtouVus1t79jorSvVdLnC9kalYmgT01uA5YGLS1+8yvPad3QFy8tbYmjlktcfJJ1PKtbVuLh/ADDJ9p6lYugHkgR8hIFlGH7gVFG1dCyVD11KutL2DsOd64bR3nN/QtIrSdMPz5b0GOliSUmLtAzD/J0KFoepaTWq7Wsa5yhfXvV+4GpJv6ai4ky231qqrSEcDHwZOI/5U2Q/WDoISbuRPuGuSXq9l5y9tAhwu+0NgO/1ur0RqGzoMk95XBpYKQ8NNdajLEcaVu660Z7c9wCeBT5JWjy0POkFVdIlki4FfpKP3wNcXDgGqHiFapO/5NsrqGhsNU83a/UkMM32rYXCWIdUZ2gR0utsB9LH/6KzqEg1zPcGppee/pc/Sd8maY1SF9SHUeXQ5UeAI0iJfBrzk/tT9GjfhdE+LPN1258b7lyBOPZm4NSm84d5SjfbnrdCFXimcZq8QnUsLmySdA4wCfhVPrUr8EdSpcif2f5GgRjuAT5NWrjTXIGw9Cyqq4AdbBcv/5zb/y1pKOQmBl7/eGcV8TQ0DV1eYvuFgu1+woUqco725N6uQNTtJeeYK21Z9kjLPOJVbD9YKobcbl+sUM3JpN0CopLj3ZcC72rMLc9Ddz8nrRidZntigRiuc9qVqlKSNicNy1RSwzwn0QU0hg5Lq3KablMMGwATGThj50fdbmdUDstI+jfgY8A6kpqXMS9LqnFS0s8YOL1tbj63efuH98xFkpax/bRSgapNgRNL9xRJvdWGJYF3Uf46yBqkTy4NLwJr2n5WUqmyEEdL+gFwJRXuNQAcS6phviQVDJNVlcTbqXqabo7haNJmMRNJ2w6+gzQ1NZJ7dg5pXPs44Mim87N7sRhgGIs1f6yz/YKq2d7uFGAjpXKinwVOJ/3BtO059YoXrJ1yvdLONyWdA9wo6Zf5eHfgJ5q/SXMJB5GGgRZnYCIpndxXtP32wm3Oo4HlKF5B+v94uvSq0KwfpunuA2xEKnt8UF4j84NeNDQqk7vtJ4EnJc1p7ZmqfA3zWZLeafvC3P4ewN8Ktt8wJy9334PUYz9d0oGlg2hZgbcI6aLuq0vGYPs/Jf2G+ddBPmq7Me/7fYXC2MgV1Cxv4wpJb3dFNczdUp1S0p5AsanKLR6igoqULZ7NF5rnKBWXe4weLGCCUZrcm7TOEFmM8jNEPkqahvndfDyTanZ3n50vrh4AvDmPLVZRV76xAk+k4ZgHSGVOS1sKeMr2mZLGS5pg+4GC7d8oaaIL1ToawqHAZyW9QBqqqqSeSoPtCyQdOfwju6dp9lTl03RJJcJXIG1FOY00ZHZTLxoalRdUB5khAmlstZIZIvminWzPbjl/oO3JBdp/NfBeUvnQ30lag1QOuetjef0uj2tOIn0Ef53S5tw/s71NwRhmkKZD9nyvzH6m+fsdQ/okNwl4iwuW3tUgheQaXLCgXDNJawHLuUflj0dlcm+QdBzwDdLm1I0rzy5UK2JE2s3oqSiOIrWs8yeGXYG1GDgjoWS53VuBTUjVDzfJ50rPoiq2V+YwcYg0FDUhD1etDqzqtJFIifab9zueQ9pr+DRXU39pSJJOst2TXZFa2tmQBV8fXb8WM9qHZe4nrfwrXiviZdDwDyliyeEf0hW/IpVTHbDDfGEv5OsPjWJVJWu6AOWT+BBOJv0etidNifwnadFMkdlctg8q0U6X9PyTndI2oBsCd9LjC+2jPbkfxvxaEW9t1IqoOKZW/fLRqFQcq/XB0MMUSd8HVpD0r6RSAD2ZkTAKbGl7U0m3wLytKIvN5pK0Gmknpm3IO0EBh9ueWSqGPrNViXUWMMo3yCbXigDm1Yog7fjST/ql517KxZIqm3oHYPubpEVL55H+Hv7D9neqjKlCL+ahssanmPGU/UR1JnAhadn9a0mf7M4c8hn1doOkIsl9tPfcKy1zmwsj7eO0u/tgqt6Qt6HUm8yNwPn5/+ZFKpid0VSC4vI258aa75BK/q4s6VjSPOsvFmx/vO3mZP5DSUcUbP/lKPEamUxK8P9Hjy+0j+oLqs0qrBVxre03l2pviDg+TtrO7PFB7t/ABXYBknQ/aV/I4oWqmmKovCxFP8nDlTvAvK0oZxRs+wrgh8wvrLc/cJB7UOK2U5I+aPuHPW7jPuBTtFyT6sU1mtok96pI+hKpMmXrxhBFV8pK+iqwH6mG+BnApVUk11zX5R1VFKpqKkuxNvCnpruWBa63fUDpmKomaR1gpu3nlXaD2hD4kfO2ewXaXwP4LtCYqXU9acy9+AVnSZOAL7Bg+eOSs6h+W6rOUiT3Dintct/Ktnuy6myYWAS8nbT0fRIwBTjd9p+GfGJ3Y/ghKbleTOGFIpKWB8bRH2Up+kKeFjqJNPXuEtKY93q2d6kwrEooVer8DAV6zUPEcDKwAun30NOaQ6N9zL1yLrht1nDy9L//A/6PNKd4HPBzSZfb/myhMB7It+L13BtlKYD9W6r/vVLSK90fNcVLe8n2nLyY6ETbJzVmzpQg6RvAV0mfbi8h1VU5wvZZpWJoMqtRJqRCS5GSevOkg55MhYyee4ckLQ78G9AYd78a+L7Lb293GHAgqa7ND4ALbL+YL2zea3udkvFUKV9/OIaW6n9jccxd0h9IG3Z8Adjd9gOS7nDaHalE+7c6bWm3F+lazCeBq1zBNpSSdiCN+VdSqTN3OL5m+zMl2ouee+dOIdVwOTkfvz+f+3DhOFYC9m79iJmLFO1WKog245qNOEom1iOovvpfvziIVP/o2JzYJwAle82N+ka7AD+x/Y80eliJSit12p4rqdhq9ei5d0jSba29kHbnCsSxYpvTsyv4BNEP45qVb9Ldj5T27ly9V7VMBmnza6Qe+7OkapArABfZ3rJUDE2xTHfFlTolHQ+sS9rzoXkCRoy596G5ktZpXLSUtDZpw47Sbibt2fk4aRbACsAjSpuG/6sXrLPeK/0wrtkP1f/6gqSrgXeSXuu3kkpUX2O73T6zXWf7SElfJ1XonCvpGdLex1Xoh0qdKwJ/Z2CJlCg/0Kc+A1yV53eLNBxRRT2NS4DzbV8KkFeJ7kyaMXMyUKqn1A87EFW+SXcfWd72U5I+DJxp+2gN3L2spyQtTSo7vAZwCGml6nrARaViaLItcGCe4VZJpc6StXZiWKYLJC1B+oMVcLft55vu29H25YM+uXsxTLU9qd25xkWtXseQ2zyLNK45oDCS7YNLtB8GkjSdNDNjMvAF238suaBL0k9Jdcs/YHsDpT2Gbyj199gSS+WVOkvW2omeexfkZD5Yb+jrNC2D76F/SPoccG4+fg/weL5CX3JB0UZVjWtK+rbtIyT9ivabdL+zgrCq9hXgUuC6nNjXBu4t2P46tt8jaX8Ap31sK7miavvPSttQ/r986ne2byscxpmkbSD3zccH5HM7druhSO69V+oP+b2kzX8vyMfX5XOLAu8uFANUO6754/z1mxW03Zds/4x08a5xfD9p0/JSXsi99UbhsnVoGq4rSdLhwL8yf3z7LEmn2j6pYBjFau3EsEyPtatz0oM2FgUm98Pyeo2CHYgknWe7ZIKrjKQlSdscrk9TTf9Sw2SSdiQVKpsIXEYajvig7atLtN8Sy+3Am2w/nY+XIQ0RlSw/UKzWTvTcayDPQhgv6RUuWDRtEDtX3P5IFC8NUaEfA3cDO5GGaN4HFCkclhfQjQP2Jm2kI9L4chUbyJPbb57JNpfyJbkPJtXaOYH0aeb3+VzXRXLvvQcLtnO9pAsZOH+26PS/PhnXHM5Y+rj6L7b3lbSH7cmSziGNwfdcXkD3caeS2L8u0eYwzgD+IOn8fLwncHqJhptKTm9Z6trPaN+so3KSlpb0JUmn5eN1m1eE2t578Gd31cOk6WWLkKogNm5F5XHNs4GV8+0sST3flzIMqrGI7QlJG5DKYq9VsP3LJX1a0uqSVmzcCrYPzPsU8QfSNOV/kNaDHGT724VC2CWXKjmqUHsx5t6pfprqleNZpjGmWFH7lY9rDkfSLc4bZ9ddnt9+HvBG0ljvK0k7U32vUPv9VDW1yCbxg7T936R5/ssAz5CvRUHvNrOJnnvn1rH9DXIPyfazVLC1nqQ3SbqLPJ4qaaNcXrR4KFQ/rjmcMbMjk+0f2H7c9rW217a9cqnEntuf0OY2L7HnC66lXCbpXVVMxbT9GdvLA7+2vZztZZu/9qLNGHPvXL9M9fo26aLZhQC2b5NUxQ5RZ1LduOZ02o+nD5ixY/uyEvH0gzxMdiYwGzgN2BQ4so/+D0qtA4G0A9IywBxJz1HBFpC2i5VeiOTeuWNIS/9Xl3Q2eapXFYHYfqilU1K8xo3tb+V6JtuSXjwH2S5VP7xY9ctR5GDbJ0raiXQN5CBSsu+X5N7zXrSkbWxfT5pj/lyv2xsmlr1Jb2grk/7tPXuDieTeIduXSZpG9VO9HpK0NWBJrwAOo9CUt2aStgLutH1zPl5W0pa2/9DrtksuIx9FGslzF1JtmduqWiE6iBIX/b4DbEaadlis5O4gvkGqq9/z12Yk9w5JuhI43vavm86davuQwqF8FDgReC0wk9QzO7RwDJBq2Te/gJ5uc66n8hvMScAbSIXDFgWeLvnxu49Mk3QZMAE4StKylC1H0Q9elHQmsJqk77TeafuwgrE8WiKxQyT3bpgAfE7S5ra/nM9NGuoJvZA/LbyvdLttyE1TsPJc59J/Z98lbRb+M9Lv4gPAvxSOoV98CNgYuN/2M5JeRTVVSwfzYIE2dgPeRiqzW6r09WCm5hl2FxB7qPa9J4AdgO/kglWVlACQNJ5UN2MtBu6AVLoa4/1KW/6dko8/RqqvXpTt+yQtansucKak35eOoUqSXm/7blJiB1i7itEYSfsCl9ieLemLpE9wX20M25VYB5I7PudKmjHUgjpJR9k+rsfhLEeaChl7qPa75jnTkj4I/DswzvZqheP4PfA7Us9k3oVU2+cVjmNl0hjn9qQ/2itJGyI/VjCGa0k9tR+QNgt/hFTPpPi+nVWRdJrtf1XalaqVbW/f5nwv4rjd9oaStgWOIxV1+7wr2IlpOCXqQJUUyb1Dkj5i+/tNx5sBh5buMatgzfZOlOgdKdXtfpQ03v5J0qrMk23f18t2w4IanR9JxwHTbZ/Tr4vIehmXpM/a/oakk2hfjrrr4/6R3BeSpOWcdrhpu5Ta9j8Kx/NV4Pe2f1Oy3Zer172jfqqQWaU85W5QvRjjHSSOi4C/kj5JbUbaS/WmfvwU1cu/TUm72/6VpAPb3W97ctfbjOS+cCRdZHu3vLy6sYy4ofjyakmzSQs0Xsi34gs0RqJEr03SpaTpZlVXyKxMnh0CaT711sBv8/FbgatLjHXnOJYmVQqdbvteSasCb+yjRVTz9MMnCkkn2e5KLaa4oLqQbO+Wv06oOhYA28WLhC2kEr2JB+mDCplVct6rM/ecJ9p+JB+vCvxPwVBWJS25f17SdsCGwI8Ktj9P02Kmwc79rM3TStumWz8oast0SNI2uTgWkg6Q9C1Ja1QQh3L7X8rHq0vaonQcI1BiykZfVMjsE2s1Env2KPC6gu2fB8yV9C+kMhQTSNvMVaHdjkvzztn+r4Kx9Fz03Dt3CrCRUg3zz5L+gH8MvKVwHCeTFqdsD/wn8E9SD23zwnEMp+e9o6b1BgGuzsNUPyF9atoPaDeDpldesj0nXwP4tu2TJJUqRwGkonqkoanxkj7VdNdypAVutRTJvXNzbFvSHsCJtk8f7KJJj21pe9PGC8f247kMQRGDzQJoaMwGKNE7ytP/2s1IKDL9r5/Y/nhOrI3NU061ff5Qz+myF5U2x/4AsHs+t3jB9iHNmnolKd81f4J7CtincCzD6don20junZst6SjS4qU359kapf94Ib2IFmV+dcrxlF1mPrVgW8P5dNP3S5I2hJ5TUSyVyL31S4CL88yYIrNj2jiIVBrjWNsPSJoAnFUyANvXANdI+mE/1R9S2kDklbafajp9Ytd+fsyW6YykVwPvBf5o+3d5vH0720UvGkl6H/Ae0grAyaQeyRdt98NFospJusZ26aGyyuS/y53z7XWkXYguAa60/c/CsSwFrGH7npLttonjcmBf20/k43HAubZ3KhjDOaQ3u7mkBYfLA9+y/d9dbyuSe2+p4O4vkl5PKoUg0ot4RtN942w/XiCG8aTNMCaSes1A2SGRlrUHi5DmV3/H9nqlYugnuYe4JfAO0t/Hs8BlTpvM9Lrt3UmrUl9he4KkjYGvuNA+oi2xLDDVsfT0x8Ziw9wZ24z0WpnmHuxUFsMyvbfk8A/pjlxL5O5B7r6SMpUZzwZ+CuxK6qEcCMwq0G6zacxfezAHeIBUQGtMsv0ScEO+/YeklUgbu5RwDLAFcHWO5dY8NFOFlyStYfsvMG8lc+ne7eJKe6nuCXzX9ouSehJDJPfe65ePRqWqRr0qX1Q+vGms85pCbTe8wS2bMkhaonAMfUHS60gzulZx2uN3Q+Cdtr9aKIQ5tp9sKVpW1WviC8B1TX+Pbybta1rS90nrMG4Drs1vME8N+YyFFPPcx45SL6gX89dHJO0qaROgaBE10qYMrW4oHEO/OA04ivl7/N5Omg5Zyh2S3gssKmndPKuqkgqdti8hfXr9KTAF2Mz2pYVj+I7t19rexcmfSauGuy6Se+/10643JXxV0vKk6pifJlVm/GSJhiW9OhduW0rSJpI2zbftgKVLxNCHlrZ9U8u5kjOHPgGsT6pdfg7wJHBEwfbnUfr4sDOwqe1fAUuXXugn6XBJy+VFh6dLupm0NqXrYlimC/JHq3VtX5FnBixme3a++/0VhtasyJuM7Yvyt0/Sox7JEHYi7V+7GnA88//NTwGfLxxLv/ib0qbtjSmy+5BKIBdh+xnScMgXSrU5hOaFfl8hbRp+HmUX+jXvaTueXu5paztuHdxIG2T8EfhTPl6XNFOlili2JW1ITf7DmdB034qFYpgMrNB0PA44o/D/w7uq/rvolxuwNnAFaYOIvwLXkUoSlGr/8jZ/D5dW9H9xc/56S9O52wrHcHv+eiKwV2s83bzFsEznDiUV+3kKwPa9pEp8RUk6mjSt6qh8anGaFou4XAniDZ3nEed2HwdKV9rbTNIKjQNJ45RKIo85tu+3/TbSm/3rbW9r+8GCIazU5u+h+Osjq3qhH8zf03YX4FL1cE/bGJbp3PO2X2jMBlDaL7SK2QB7kZJoY/uyh/MfTmmLNM+pz3POS/+dvcP2vGEYp1IMuwBfLBxHZVpqqDSfB4pWyOyH6YcN3wHOB1aWdCxpod+XCsdQbE/bSO6du0bS50kX8XYk7Rn6qwrieMG2G3NmlStVVuB44PeSfp6P9wWOLRzDopKWsP08zFshOdamQvZLFcx+mH4IgO2zJU1j/kK/Pd200K9UGKQFfruRxv2XoUdrYWKFaofy6r8PkTa8FXAp8AMX/o+V9GnSeP+OpL0qDwbOsd2uzGmvY5lIumjVWCl7V+H2Pwu8k3ShyqT/iwtdYEVmWFBeNLVVPrzRacPqKuL4se33D3euxzGcQr6oa/sNuQTCZba7flE3knuN5E8O895kbF9esO1+23awsdRepBdP0fnM/ULS2qSLd1uR3uhuAD5p+/6CMbyT1GOHtAvURUM9vodxDNhGL4+/T7c9sXQMzWUPJN3mHmw7GMMyHZI0nQXHEJ8kVUn8qu2/l4olJ/NiCb3FOaSPmo2l/w3Kx0W3HbR9MXBxyTb71Dmkuv575eP9SLXdtyzRuKSvkaYanp1PHa60+9FRQzyt2zEcRZoKu5Skp5g/RfYF4NRScWTFLupGz71Dkr5BqvDW2F2msfrvKWBb27u3fWL345jN4G8y/16yp1a1lv+LV5BmDj3tPttPtgRJf7C9Zcu5G21vNdhzutz+7cDGTvVtGr3lW9yDQlkjiOW4km8qg8RQrHpr9Nw7t43t5n0Pp0u63vY2kg4oGMe3SNvLnUPqmewHvBq4BzgD2K5EEJLaFSd7Eviz7SIrI92yn6ykPUnFq8aMpuGxqyQdCZxLesN7D/DrwuGsADSG5ZYv3PY8to+qeoio5EXd6Ll3SNJtwCG2/5CPtwBOs71RyXKiQ/XQejWmN0gcN5J6JbeT/njfSCqS9Crgo65o1/uSvdV+IOkB5lfGbGXbRYbJJO0HfJ20tZ9IifUo2+eWaL8lluNIb/KNIaL9gamFh4i2Au50XsGepytPbOSPboqee+c+DJwh6ZWkP96ngA/nqYjHFYzjJUnvBhpTEJu3Dyv5Dv4g8CHbd8K8mTOfIe3r+gt6scy6hdK2cg2LAJPon+qcRdiuqqzuPHkm2Uuki7mbk14fn7P9fxWFtCsDh4gmA7cwf+FfCacwsPT2023OdUUk9w7Z/iPwxlwsS82r8UiV50p5H2lWxMmkRHYjcECe4/3xgnG8vpHYAWzfJWkT2/dL7TqRPdF8nWMO6Q1nj1KN95M8xr0rsBZNr/cSi5hsvyTp47anABf2ur0RWoFqh4jUPE06/x/1JA9Hcu+QUp3wd5FfPE0rAL9SMo58wXSwi7fXFQzlnjyXt/Gx+z3A/+b/pxcHf1p35GR2u+0Tet3WKPEr4DlgOuWX2gNcntdg/JTUSwXKT43N/gu4RWkD9XlDRIVjuF/SYaTeOqRFjz2Z7BBj7h2SdAnpguE00qwZAGwfXziO8aQiZmsxsId2cOE4liL9wW5LegFdR/o08Ryp/GzP9++UdJXt0hUp+5Kk26uYmdLUfmPsf4BSY/5NcSxCGqr8HfOHiP5QeohI0sqkMgiNMr9XAEfYfqzrbUVy74ykO2xv0Adx/J70h9v6JnNeBbG8AliP9KK+x3bPe+wt7R9L+sjd2lu8uWQc/UDS10mrhKu6kN38Zm/S3+j3bD9bQSzX2n7z8I+sh0juHZJ0KnCS7ekVx3Gr7Y2rjCHHsR1p/u6DpN7R6sCBtq8tGMNVbU7bBTfp7heS9iJVB12ENCwm0v9FkTn/kqaQJhk0z1BZwfa7S7TfEsuXSJuDVzZEJGk14CRSJVmTPtkebntm19uK5N4ZSXcB/0LahPl55r94in4UVipp+3vbvynZbps4pgHvtX1PPn4d8BPbmxWMYe3WRVvtzo0Fku4nbcY8vXS9o9z+AtNwS07NbWm38iEiSZeT1qL8OJ86AHif7R273lYk987kEqYLcNobsWQcs0kV5p6ngh5aUxwLjPGWHvdtrSGSz00r+QbTLyRdSiqBXMXFVCT9kDQMc2M+3pL0Se5jFcRS+RBRu0/YvfrUHbNlOtRI4vlCSU9Kd44wjn4p8TpV0unM75m8j3QdoOckvZ60X+fyLXPdl6PC303FHgGulnQx6Y0fKFrPfUvgA5L+ko/XAGY0ajIV/oQ7mTRE9J18vH8+V3KI6G955fpPmmLoSf2pSO4dysuZjwdeAzwGrAnMICWZ0rGMI5X9nZfISo51Z/9G2p3qMNKnh2tJs2VKWI9UvGwFBk4LnU2aSTQWPZBvr8i30nauoM3BrNcyHHRVXmFe0sHAd4ETSJ8efp/PdV0My3Qo/3FsD1xhexNJbwX2t110QwJJHwYOJ20OfStpVeANY/Qi4pts31B1HP1E0jK2nx7+kfVV9RBRXoMx2XaRmlOxh2rnXnQq67uIpEVsX0XaRqu0w0nzd/+c53hvAswq1bik6ZJuH+xWKo5sL0nLSVpc0pWSGh+FxxxJb8oX/Wfk440klfok1W+2JO0S9qCkB0m17d/S+NvtdeO25wLj81Thnothmc49kevKXAucLekx0pL30p6z/ZwklLaYu1vSegXb361gW8N5u+3P5mmAM0lb/V1F04bhY8i3gZ3Iy/9t3yZpzMz1btEPQ0QPAtdLupCB0zG7fg0kknvn9iCtvvwk6eLh8qS9EUubKWkF4ALSku/HSSWAi2ieHSRpFdKnCICberH6bhiL56+7kKZh/qNgXZu+Y/uhln//3MEeW2elZ7AN4uF8W4Qe73Mbyb1DLeOYkyuMo7HTzjF5Ec/ywCWl48iVKf8buJp0QfUkSZ+x/fMhn9hdv5J0N2nBysdyaYbnCrbfTx6StDXgPBxwGHmIJpRn+8sAkpZLh6n0by/EBdUO5Sl3XwdWJiWzSuaX51gWBVZhYG2Zvwz+jJ7EcBuwY6O3nhPrFaUXreSZQ0/ZnqtUfnnZRh0RSTu64P6yVVLanPpE4G2kv83LSCsii23/GOaTNIm0cXuj1/4kcLDtrk8XjuTeIUn3Abu7R7upvIw4PgEcDTzK/Op/VayUnW77jU3HiwC3NZ+rWrtFTiGUkC/cHmr7d/l4W+DkXrxOY1imc49Wndizw0nzeKvukV2SV0U2Fmm8h/7bqHrMDMAr7fH7VdIQ1SXARqQqhGPx4nI/mN1I7AC2r8ury7sueu4LqWkF5FtIe5VewMAVgL8oHM9VpOGQKmbqtMayN/NL/l5r+/yKQxpgLPXcG0vb88yhPUkX/q+qorZLAEknAEuTOj+NPW0fB86D7lYujZ77wmusgDTwDPD2pvtM2lKu5yR9Kn97P2mZ+a+pZpl5I54JwG8ab26SlpK0lu0HS8YR5omZQ/1l4/z16JbzW5PyRtcWHUZyX0i2D4J5+zAe7ry9Xr6QV3KjjsaFmb/kW1XLzBt+RvpDbZibz23e/uGVeLDqAAqKmUN9xMNsIiPpQNtdmXUXwzIdknSL7U2GOzdWDFL1rniJ1zz9by0Gzhz6UckY+kXLzKGlgeVc3SbVYQjdHDKM8gOdWyS/eACQtCIVfCKSdHlexNQ4HpcvbJY2KxdTa8SxB/C3kgFI+jHwTdK4/+b5NqlkDP1C0r7AnJzYv0hapfuaisMKg+vamFkMy3TueFK9ip+TxszeDRxbQRzjG0NDALYfz2WIS/soqQzD/5D+P2YCHygcwyRgYhWbU/ShL9n+WZ5ytxPpTe8UUp2V0H+69jcbyb1Dtn8kaSrpQoiAvW3fVUEocyWt0Vi0pLSJSPHkZvtPwFa53o56uQJvCHeQZjA9UkHb/aZRamBX4BTbv5R0TIXxhKFFz72f5GReRUJv9gXgOknX5OM3A0XLDsO8ujL/BbzG9jskTQTeZPv0gmGsBNwl6SYGzhx65+BPqa2/Svo+aYXq1yUtQQzHViIv6NvH9pQhHnZ919qLT671kZeab0V697/B9t+a7lvf9p0FYriYtLz6C7Y3krQYcEvJFaqS3tLuvO1r2p2vs3wBdWfSHqr3SloVeKPtyyoObUySdK3tIlU5I7mPEaUW7kj6o+3Nm2cMtZtBE8pSyzaQpWsOhUTSl0jTUn/KwJK//+h2WzEsM3aUWrnytKRXkcf7JW1FKo7Uc5Kus71tXs7d3GuprJhb1bTgNpBrAHdTwTaQAZi/pd6hTecMrN3thiK5jx2lPqJ9irQxxDqSrgfGA/uUaNj2tvlrv2wW3g/+kzRUN2AbyIpjGrNsTyjVViT30FW2b85j3uuResz32H6xcX/JcrsxFAHkbSAlzdsGUtLXqw5qrJK0OGkT+ca4+9XA95tfI90Syb0GlIqFrGb7oSEe9kKpeHLxssEu3n4d6GlybzMUsSZpg4qxOBTRL9tAhuQUUr2fxj6278/nPtzthuKCak1ImmZ7s6rjGE6J0gx5w5DtaRmKsF18amjV8kYlz5KmPza2gTy7D0pDj0ntSnH0qjxHzHetjxsl9VNxrsGU6E28mJPXvKEI5lfjG1NsP237JdtzbE+2/Z3mxC7phirjG4PmSlqncSBpbXq0p20My9THW4GPSPozaYpVY4ZI0Z2Y+kQMRYzcksM/JHTRZ4CrJN1Peo2uCRzUi4ZiWKYmcrmBBbg/dnyfR9IvbO89/CM7aqPdUMRZvZhLPNqNpY1L+kVeJdyYcHC37eeb7uvahINI7jUiaSPg/+XD39m+raI4Ki23K+nrtj833LkQyb3fRMnfsABJhwNnAyvn21l50+zScfRDud0d25x7R+EYRovYlqm/dO33ET33msi7qr/J9tP5eBlSfZmiY+6SZlBRuV1J/wZ8jLTa709Ndy0LXG/7gNIxVU3Sx0mzYx4f5P4NbN9ROKwwiG723OOCan2IgVfd51JNr6zKcrvnABcDxwFHNp2fPYbH218N/FHSzcAZwKXNb7yR2Osreu41kTfKPhA4P5/aE/ih7W8XjqMx7bDScruSFgVWYeC4/1hcodpY5PZ20qyMScAU4PRcez/0kW5OOIjkXiOSNiWNdQu41vYtTfeNG+yjeZdjqLzcbh6KOAZ4FHhpfghjclooMO9i+0Gk8r9XkerNXG77s5UGNkZIGjJh2/5F19uM5D42jKVZEZLuA7aMVZgg6TDSJ7q/AT8ALrD9Yt444l7b6wz5A0JXSDozf7sysDXw23z8VuDqXkwPjjH3saOn4+99Vm73IQqVGR4FViJt/ThgvYPtlyTtVlFMY47tgwAkXUSacPBIPl4V+J9etBk99zFijPXcTyctEvk1A8f9v1VZUIVJWnGo+8fwBeZKSbrD9gZNx4sAtzef65bouYeeqLjc7l/y7RX5NhZNY/4nqNZPbT3ZHCKMyNWSLgV+Qvo97Ee6BtJ10XMfI0pUY8zttC23a3ssltsNYQH54mpjJfm1ts8f6vEL3U4k93qQ9E3gzME2wZa0YomP4lWW25X0bdtHSPoVbapPlp6OWaU8c2pQtm8uFUuoRgzL1MfdwKmSFgPOBH5ie95FxYJjrFXu/PPj/PWbhdrrZ8cPcZ9Jb8ChsNxr/zpp1ozo4YSD6LnXjKT1SPOZ9weuB07L9cxLtX8FaQHVcaSZGo8Bm9veulQMIfSrPE13d9szet5WJPf6yKsydyMl99VJKxG3BZ62vV+hGCovtytpXdKby0QGXtQdkxcRJW3Agv8Xxap0hvkkXW97myJtRXKvB0nfAnYnLY443fZNTffdY3u9QnFUXm5X0nXA0cAJpP+Tg0h/60eXiqFfSDoa2I6U3H9Dqo55ne19qoxrrJJ0IqnezwUMnKbb9RWqUfJ3lJO0b/72DmAj2x9pTuzZFgVD6odyu0vZvpKU0P9s+xjG7hjzPsAOwP/lhTQbAUtUG9KYthzwDKnWz+751pPFZHFBdfQ7CvgZ8HHbZ7R7QPOF1V5pLrebyw83LEsa+y/pucby+lxn5q+kC1hj0bN5NeocScuRroGMyeGpftBYqVpCJPfR7++5EuMESRe23llw+l8/lds9AlgaOAz4T1L9jg8UjqFfTJW0AnAaaWHTP0kVO0MFJL0OOAVYxfYGkjYE3mn7q11vK8bcRzdJrwA2JU0D/HDr/SWrMTbFVGm5XUmTgC+QFlAtPj+EsVsVEkDSWsBytm8f7rGhNyRdQ9ok+/uNRYWtJQm6JXruo5ztF4AbJW1te1bV8QxWbhcomVjPJr2ApjfFMCZJ2gv4re0nbT8oaQVJe9q+oOrYxqilbd+USuzPM6cXDUVyH+UaqzKBMyT1w6rMI4D1Ki63O8v2AkNUY9TRzcvbbT+RZ9BcUF1IY9rfJK1DXkEtaR96tGtZJPfRr99WZfZDud2jJf0AuJIeTzcbBdrNiIvXfXUOBU4FXi/pr8ADQE/29o0x99BV/VBuV9JZwOuBOxm4E9PBpWLoF5LOAJ4g1Qw38AlgnO0PVhjWmJcX+y1ie3av2oh38FFO0nTaFMlqqOAiYj+U293I9hsrarvffAL4EvBTUh2Ty0i9x1ABSYeTaj/NBk7LBd6OtH1Z19uKnvvoJmnN/G3jBdsYpnkf8Iztr5SPqlqSTgNOsH1X1bGE0EzSbbY3krQT6TX7JVI1165vpBPJvSba1awoWseij8rtSpoBrEMaz3ye+ZX3xsxUyH76fYT5JN1ue8NchuBq2+f3aq+FGJapj2UkbWv7OgBJWwPLFGy/ny7s7lx1AH2gn34fYb5pki4DJgBHSVqWHk3XjZ57TUjaDDiDVIUR0kW0g2NThtAgaRyweixiqk4ui7ExcH+elvoq4LWN34mk9QfbcOdltxXJvV5y/RCVqCczSPtRbrePSLoaeCfpU/qtwCzgGtufqjCsMIhubmQfVSFrQtLyuezvb4ErJR0vafnhntcDZ5JqZ8wh1XT5EfOHCEJ5y9t+CtibdOFuM+BtFccUBte6mflCi+ReH2eQple9O9+eIiXa0qLcbn9ZTNKqpL+Ji6oOJgyra0MpcUG1Ptax/a6m4y9LurWCOKLcbn/5CnApaYOOP0paG7i34phCAdFzr49nJW3bOJC0DWm7u9KOYH653c1IS6vHarndfnCl7Q1tfwzA9v0tnYDQX17o1g+KC6o1IWljYDJptoyAfwAftH1b4Tii3G4fkXQv6ULqmcDFjhd8pZTKQb4PWNv2VyStAby6ze5pnbcVv+t6ybNlyBfRqmj/HtqU27X95yriGetyMnkbcDBpu8WfAj+0/b+VBjZGSTqF9LrY3vYb8vTUy2xv3vW2IrnXQ95t5wPAWgzcJOOwwnFcZ3vb4R8ZSpP0VuAs0uK220g1TW6oNqqxpTHVsXlVaqMkQbfbiguq9fEb4Eaq36Aiyu32kbxI5gDg/aQNVD4BXEhaSPMz0krJUM6LeaeyRj338fTo9RrJvT6W7JOFKQeRyu0uzsCdmCK5V+MG0jqDPW3PbDo/VdL3KoppLPsOcD6wsqRjgX2AL/aioRiWqQlJnyRtfnwRA3vMRTenljQ9yu32D0mKi6j9RdLrgR1IEx+utD2jF+1Ez70+XgD+mzRTpfFiNlB62f+NkiZGud2+sa6kT7PgtZhYWFaBvMXeA7b/R9J2wI6SHrH9RNfbijf1epD0J2BL23+rOI4xX263n0i6DfgeMA2Y2zhve1plQY1heWHhJNKb7SXAr0h7Du/S7bai514fdwLPVB0EUW6338yxfUrVQYR5XrI9R9LewIm2T5J0Sy8aiuReH3OBWyVdxcAx96JTIWM+e3+QtGL+9leSDiVd0K7sWkyY50VJ+5OmLe+ezy0+xOMXWgzL1ISkA9udtz25dCyhepIeIF1zaVQZHPBCjxLM1ZA0EfgocIPtn0iaALzH9te63lYk97FB0nlRU2TskbQU8DFgW1KC/x3wPdtV1B0KBUVyHyN6tU9j6G+SppDKP5+dT+0PrGD73dVFNXblgn7HkGovLcb8CQdd/yQVY+5jR7yLj03rtSxtvyrPoAnVOB34JC2zl3ohknsI9XaLpK1s3wggaUvg+opjGsuetH1xiYZiWGaMiGGZsSmvO1gP+Es+tQYwg1QaItYfFCbpa8CiLDh7qesb2UdyrwlJuwG/sd22CJGkt9u+rHBYoWKS1hzq/pi6WlaeqtzKvVgxHMm9JiSdBbwJOI+0EXJP6lWEEEaH2GavJmwfAGwC/Ak4U9INkg6RtGzFoYUQMkmrSDpd0sX5eKKkD/WirUjuNZJ3XzoPOBdYFdgLuFnSJyoNLITQ8EPShuWvycf/S9p3uOsiudeEpN0lnQ/8lrSceQvb7wA2Aj5daXAhhIaVbE8h73Vgew49mhIZUyHrY1/gBNvXNp+0/YykgyuKKYQw0NN5d6zGTkxbAU/2oqG4oBpCCIVI2hQ4CdgAuAMYD+xj+/ZutxU991FO0mwGrj4V8wtG2fZylQQWQmhnHeAdwOrAu4At6VEejjH3Uc72sraXa7ot2/y16vhCCAN8KU98GAe8DTgV6Em9/ei5j3JNdbvbirrdIfSVxsXTXUnVOX8p6ZheNBRj7qNcm7rdzXpSbS6EsHAkXQT8ldRr3wx4Frippbhbd9qK5B5CCGVIWpq0FeV02/dKWhV4Yy9Kg0RyrxFJ44B1gSUb51qnRoYQxoYYc68JSR8GDgdWA24FtgJuALpekCiE0P9itkx9HA5sDvzZ9ltJdWZmVRtSCKEqkdzr4znbzwFIWsL23aQ63iGEMSiGZepjpqQVgAuAyyU9DjxcaUQhhMrEBdUakvQWYHngEtsvVB1PCKG8SO41kQsQ3Wl7dj5eFpho+w/VRhZCqEIk95qQdAuwqfMvVNIiwFTbm1YbWQihCnFBtT7kpnfqvJdqXFMJYYyK5F4f90s6TNLi+XY4cH/VQYUQqhHJvT4+CmxNqlsxk1RK9JBKIwohVCbG3EMIoYZiTHaUk/RZ29+QdBIDN+0AwPZhFYQVQqhYJPfRb0b+OrXSKEIIfSWGZUIIoYai514TkiYBXwDWpOn3anvDyoIKIVQmeu41Ieke4DPAdOClxnnbf64sqBBCZaLnXh+zbF9YdRAhhP4QPfeakLQDsD9wJfB847ztX1QWVAihMtFzr4+DgNcDizN/WMZAJPcQxqBI7vWxke03Vh1ECKE/RPmB+rhR0sSqgwgh9IcYc68JSTOAdYAHSGPuAhxTIUMYmyK514SkNdudj6mQIYxNMSxTEzmJrw5sn79/hvj9hjBmRc+9JiQdDUwC1rP9OkmvAX5me5uKQwshVCB6dvWxF/BO4GkA2w8Dy1YaUQihMpHc6+OFvM1eYw/VZSqOJ4RQoUju9TFF0veBFST9K3AFcFrFMYUQKhJj7jUiaUfg7aRpkJfavrzikEIIFYnkHkIINRTDMjUhaW9J90p6UtJTkmZLeqrquEII1Yiee01Iug/Y3faMYR8cQqi96LnXx6OR2EMIDdFzH+Uk7Z2/fQvwauACop57CGNeJPdRTtKZ+VuTZsk0s+2DC4cUQugDUc99lLN9EICkycDhtp/Ix+OA4ysMLYRQoRhzr48NG4kdwPbjwCbVhRNCqFIk9/pYJPfWAZC0IvHJLIQxK1789XE88HtJPyeNv78bOLbakEIIVYkLqjWSt9nbnnRh9Urbd1UcUgihIpHcQwihhmLMPYQQaiiSewgh1FAk9xBCqKFI7iGEUEP/H74V6+pLvhK+AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "testing = df.loc[:, ~df.columns.isin(['abstract', 'title'])]\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "testing.sum().plot.bar()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "id": "5baee1c8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(961,)\n",
      "(170,)\n"
     ]
    }
   ],
   "source": [
    "# split into train and test\n",
    "train, test = train_test_split(sample_df, random_state=76, test_size=0.15, shuffle=True)\n",
    "\n",
    "X_train = train.abstract\n",
    "X_test = test.abstract\n",
    "\n",
    "print(X_train.shape)\n",
    "print(X_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "742b88fc",
   "metadata": {},
   "outputs": [],
   "source": [
    "def feature_importance(pipeline):\n",
    "    '''\n",
    "    Extract feature importances from pipeline. \n",
    "    Since I am using CalibratedClassifierCV I will average the coefficients over calibrated classifiers.\n",
    "    '''\n",
    "    # average coefficients over all calibrated classifiers\n",
    "    coef_avg = 0\n",
    "    classifiers = pipeline[1].estimators_[0].calibrated_classifiers_\n",
    "    for i in classifiers:\n",
    "        coef_avg = coef_avg + i.base_estimator.coef_\n",
    "    coef_avg  = (coef_avg/len(classifiers)).tolist()[0]\n",
    "    # get feature names from tf-idf vectorizer\n",
    "    features = pipeline[0].get_feature_names()\n",
    "    # get 10 most important features\n",
    "    top_f = pd.DataFrame(list(zip(features,coef_avg)), columns = ['token','coef']) \\\n",
    "        .nlargest(10,'coef').to_dict(orient = 'records')\n",
    "    return top_f"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "id": "df63faf1",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "... Processing attach\n",
      "Accuracy : 0.9588 . Area under the ROC curve : 0.9114\n",
      "\n",
      "... Processing change_size_or_color\n",
      "Accuracy : 0.9529 . Area under the ROC curve : 0.857\n",
      "\n",
      "... Processing chemically_modify_or_change_energy_state\n",
      "Accuracy : 0.9588 . Area under the ROC curve : 0.9237\n",
      "\n",
      "... Processing maintain_ecological_community\n",
      "Accuracy : 0.9294 . Area under the ROC curve : 0.828\n",
      "\n",
      "... Processing maintain_structural_integrity\n",
      "Accuracy : 0.8824 . Area under the ROC curve : 0.903\n",
      "\n",
      "... Processing move\n",
      "Accuracy : 0.9294 . Area under the ROC curve : 0.9038\n",
      "\n",
      "... Processing physically_assemble/disassemble\n",
      "Accuracy : 0.9235 . Area under the ROC curve : 0.8274\n",
      "\n",
      "... Processing process_resources\n",
      "Accuracy : 0.9118 . Area under the ROC curve : 0.8977\n",
      "\n",
      "... Processing protect_from_harm\n",
      "Accuracy : 0.8588 . Area under the ROC curve : 0.8637\n",
      "\n",
      "... Processing sense_send_or_process_information\n",
      "Accuracy : 0.8941 . Area under the ROC curve : 0.8386\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# define the pipeline\n",
    "classifier = CalibratedClassifierCV(LinearSVC()) \n",
    "\n",
    "# for each category train the model and get accuracy, auc\n",
    "models = {}\n",
    "features = {}\n",
    "preds = {}\n",
    "for category in categories:\n",
    "    # give pipelines unique names. important!  \n",
    "    SVC_pipeline = Pipeline([\n",
    "                (f'tfidf_{category}', TfidfVectorizer(stop_words=stop_words)),\n",
    "                (f'clf_{category}', OneVsRestClassifier(classifier, n_jobs=1)),\n",
    "            ])\n",
    "    print('... Processing {}'.format(category))\n",
    "    # train the model using X_dtm & y\n",
    "    SVC_pipeline.fit(X_train.values.astype('U'), train[category])\n",
    "    models[category] = SVC_pipeline\n",
    "    # compute the testing accuracy\n",
    "    prediction = SVC_pipeline.predict(X_test.values.astype('U'))\n",
    "    preds[category] = prediction\n",
    "    accuracy = accuracy_score(test[category], prediction)\n",
    "    # compute auc\n",
    "    probas_ = SVC_pipeline.predict_proba(X_test.values.astype('U'))\n",
    "    fpr, tpr, thresholds = roc_curve(test[category], probas_[:, 1])\n",
    "    roc_auc = auc(fpr, tpr)\n",
    "    print(\"Accuracy : {} . Area under the ROC curve : {}\".format(round(accuracy,4), round(roc_auc,4)))\n",
    "    print()\n",
    "    # get most predictive features\n",
    "    features[category] = feature_importance(SVC_pipeline)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "id": "cea63d60",
   "metadata": {},
   "outputs": [],
   "source": [
    "def predict_tags(X, labels = None):\n",
    "    '''\n",
    "    Predict tags for a given abstract.\n",
    "    \n",
    "    Args:\n",
    "      - X (list): an iterable with text.\n",
    "      - labels (pandas.Dataframe): label indicators for an abstract\n",
    "    '''\n",
    "    preds = []\n",
    "    if type(X) is str: # convert into iterable if string\n",
    "        X = [X]\n",
    "    \n",
    "    # get prediction from each model\n",
    "    for c in models.keys():\n",
    "        preds.append(models[c].predict(X))\n",
    "    \n",
    "    # print original labels if given\n",
    "    if labels is not None:\n",
    "        assert len(X) == 1, 'Only one extract at a time.'\n",
    "        predicted_tags = [k for k,v in zip(list(models.keys()),preds) if v[0] > 0]\n",
    "        original_tags = list(labels.index[labels.map(lambda x: x>0)])\n",
    "        print('Original Tags: {}'.format(str(original_tags)))\n",
    "        print(\"Predicted Tags: {}\".format(str(predicted_tags)))\n",
    "        \n",
    "    return preds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "id": "dc66b928",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Title: The role of granules within viscous capture threads of orb-weaving spiders\n",
      "Original Tags: ['attach', 'maintain_structural_integrity']\n",
      "Predicted Tags: ['attach', 'maintain_structural_integrity']\n",
      "\n",
      "Title: Evidence that halogenated furanones from Delisea pulchra inhibit acylated homoserine lactone (AHL)-mediated gene expression by displacing the AHL signal from its receptor protein.\n",
      "Original Tags: ['protect_from_harm']\n",
      "Predicted Tags: ['protect_from_harm']\n",
      "\n",
      "Title: Orange/lemon-scented beetles: opposite enantiomers of limonene as major constituents in the defensive secretion of related carabids\n",
      "Original Tags: ['protect_from_harm']\n",
      "Predicted Tags: ['protect_from_harm']\n",
      "\n",
      "Title: Polyphosphoprotein-Containing Marine Adhesives\n",
      "Original Tags: ['attach']\n",
      "Predicted Tags: ['attach']\n",
      "\n",
      "Title: Convergent evolution of metabolic roles in bacterial co-symbionts of insects\n",
      "Original Tags: ['maintain_ecological_community']\n",
      "Predicted Tags: []\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# predict tags for 20 abstracts\n",
    "for i in range(0,5):\n",
    "    print(\"Title: \" + test.title.iloc[i])\n",
    "    predict_tags(test['abstract'].iloc[i], labels = test.iloc[i,2:])\n",
    "    print()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "id": "3a376b1d",
   "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>attach</th>\n",
       "      <th>change_size_or_color</th>\n",
       "      <th>chemically_modify_or_change_energy_state</th>\n",
       "      <th>maintain_ecological_community</th>\n",
       "      <th>maintain_structural_integrity</th>\n",
       "      <th>move</th>\n",
       "      <th>physically_assemble/disassemble</th>\n",
       "      <th>process_resources</th>\n",
       "      <th>protect_from_harm</th>\n",
       "      <th>sense_send_or_process_information</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>adhesive</td>\n",
       "      <td>microtubule</td>\n",
       "      <td>pathway</td>\n",
       "      <td>ecosystem</td>\n",
       "      <td>mechanical</td>\n",
       "      <td>muscle</td>\n",
       "      <td>blades</td>\n",
       "      <td>transport</td>\n",
       "      <td>temperature</td>\n",
       "      <td>sensitivity</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>attachment</td>\n",
       "      <td>folding</td>\n",
       "      <td>gold</td>\n",
       "      <td>enac</td>\n",
       "      <td>bone</td>\n",
       "      <td>swimming</td>\n",
       "      <td>slime</td>\n",
       "      <td>networks</td>\n",
       "      <td>heat</td>\n",
       "      <td>light</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>adhesion</td>\n",
       "      <td>blades</td>\n",
       "      <td>microbial</td>\n",
       "      <td>fertilization</td>\n",
       "      <td>properties</td>\n",
       "      <td>drag</td>\n",
       "      <td>microtubule</td>\n",
       "      <td>aqp</td>\n",
       "      <td>temperatures</td>\n",
       "      <td>magnetic</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>glue</td>\n",
       "      <td>proboscis</td>\n",
       "      <td>enzyme</td>\n",
       "      <td>microtubule</td>\n",
       "      <td>wood</td>\n",
       "      <td>vortex</td>\n",
       "      <td>web</td>\n",
       "      <td>water</td>\n",
       "      <td>resistance</td>\n",
       "      <td>visual</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>proboscis</td>\n",
       "      <td>dermis</td>\n",
       "      <td>bacteria</td>\n",
       "      <td>reproductive</td>\n",
       "      <td>strength</td>\n",
       "      <td>microtubule</td>\n",
       "      <td>decorations</td>\n",
       "      <td>uptake</td>\n",
       "      <td>loss</td>\n",
       "      <td>scent</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>pads</td>\n",
       "      <td>multilayer</td>\n",
       "      <td>oil</td>\n",
       "      <td>algal</td>\n",
       "      <td>erosion</td>\n",
       "      <td>locomotion</td>\n",
       "      <td>matrix</td>\n",
       "      <td>wetlands</td>\n",
       "      <td>bacterial</td>\n",
       "      <td>colour</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>proteins</td>\n",
       "      <td>colour</td>\n",
       "      <td>energy</td>\n",
       "      <td>beaver</td>\n",
       "      <td>tissue</td>\n",
       "      <td>wake</td>\n",
       "      <td>rostrum</td>\n",
       "      <td>prey</td>\n",
       "      <td>soil</td>\n",
       "      <td>response</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>dopa</td>\n",
       "      <td>scattering</td>\n",
       "      <td>muscles</td>\n",
       "      <td>tree</td>\n",
       "      <td>stiffness</td>\n",
       "      <td>penguins</td>\n",
       "      <td>skeletons</td>\n",
       "      <td>leaf</td>\n",
       "      <td>enac</td>\n",
       "      <td>pollination</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>fertilization</td>\n",
       "      <td>humidity</td>\n",
       "      <td>enzymes</td>\n",
       "      <td>rules</td>\n",
       "      <td>wetlands</td>\n",
       "      <td>force</td>\n",
       "      <td>dopa</td>\n",
       "      <td>fog</td>\n",
       "      <td>lymphatic</td>\n",
       "      <td>sensory</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>cement</td>\n",
       "      <td>canopy</td>\n",
       "      <td>leg</td>\n",
       "      <td>ant</td>\n",
       "      <td>collagen</td>\n",
       "      <td>awn</td>\n",
       "      <td>formation</td>\n",
       "      <td>blubber</td>\n",
       "      <td>anti</td>\n",
       "      <td>hairs</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          attach change_size_or_color  \\\n",
       "0       adhesive          microtubule   \n",
       "1     attachment              folding   \n",
       "2       adhesion               blades   \n",
       "3           glue            proboscis   \n",
       "4      proboscis               dermis   \n",
       "5           pads           multilayer   \n",
       "6       proteins               colour   \n",
       "7           dopa           scattering   \n",
       "8  fertilization             humidity   \n",
       "9         cement               canopy   \n",
       "\n",
       "  chemically_modify_or_change_energy_state maintain_ecological_community  \\\n",
       "0                                  pathway                     ecosystem   \n",
       "1                                     gold                          enac   \n",
       "2                                microbial                 fertilization   \n",
       "3                                   enzyme                   microtubule   \n",
       "4                                 bacteria                  reproductive   \n",
       "5                                      oil                         algal   \n",
       "6                                   energy                        beaver   \n",
       "7                                  muscles                          tree   \n",
       "8                                  enzymes                         rules   \n",
       "9                                      leg                           ant   \n",
       "\n",
       "  maintain_structural_integrity         move physically_assemble/disassemble  \\\n",
       "0                    mechanical       muscle                          blades   \n",
       "1                          bone     swimming                           slime   \n",
       "2                    properties         drag                     microtubule   \n",
       "3                          wood       vortex                             web   \n",
       "4                      strength  microtubule                     decorations   \n",
       "5                       erosion   locomotion                          matrix   \n",
       "6                        tissue         wake                         rostrum   \n",
       "7                     stiffness     penguins                       skeletons   \n",
       "8                      wetlands        force                            dopa   \n",
       "9                      collagen          awn                       formation   \n",
       "\n",
       "  process_resources protect_from_harm sense_send_or_process_information  \n",
       "0         transport       temperature                       sensitivity  \n",
       "1          networks              heat                             light  \n",
       "2               aqp      temperatures                          magnetic  \n",
       "3             water        resistance                            visual  \n",
       "4            uptake              loss                             scent  \n",
       "5          wetlands         bacterial                            colour  \n",
       "6              prey              soil                          response  \n",
       "7              leaf              enac                       pollination  \n",
       "8               fog         lymphatic                           sensory  \n",
       "9           blubber              anti                             hairs  "
      ]
     },
     "execution_count": 50,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 10 most important features by category\n",
    "features_df = pd.DataFrame(features)\n",
    "features_df.apply(lambda x: [d['token'] for d in x], axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "id": "add1ee12",
   "metadata": {},
   "outputs": [],
   "source": [
    "# get all predictions\n",
    "y_pred = np.array(predict_tags(test.abstract.values.astype('U'))).T\n",
    "\n",
    "# get true labels in the same order\n",
    "y_true = test[list(models.keys())].to_numpy()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "id": "a2a43b43",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.08"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "hamming_loss(y_true,y_pred)\n",
    "# fraction of labels assgined incorrectly.\n",
    "# the lower the better"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "id": "303e0710",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.4411764705882353"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(y_true, y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "id": "04c392cc",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Hamming Score : 0.518235294117647\n"
     ]
    }
   ],
   "source": [
    "def hamming_score(y_true, y_pred, normalize=True, sample_weight=None):\n",
    "    '''\n",
    "    Compute the Hamming score (a.k.a. label-based accuracy) for the multi-label case\n",
    "    http://stackoverflow.com/q/32239577/395857\n",
    "    '''\n",
    "    acc_list = []\n",
    "    for i in range(y_true.shape[0]):\n",
    "        set_true = set( np.where(y_true[i])[0] )\n",
    "        set_pred = set( np.where(y_pred[i])[0] )\n",
    "        #print('\\nset_true: {0}'.format(set_true))\n",
    "        #print('set_pred: {0}'.format(set_pred))\n",
    "        tmp_a = None\n",
    "        if len(set_true) == 0 and len(set_pred) == 0:\n",
    "            tmp_a = 1\n",
    "        else:\n",
    "            tmp_a = len(set_true.intersection(set_pred))/\\\n",
    "                    float( len(set_true.union(set_pred)) )\n",
    "        #print('tmp_a: {0}'.format(tmp_a))\n",
    "        acc_list.append(tmp_a)\n",
    "    return np.mean(acc_list)\n",
    "\n",
    "print('Hamming Score : {}'.format(hamming_score(y_true,y_pred)))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.9.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}