MLProject.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# News Classification with Pegasos"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's import the required libraries. We also use the `imbalanced-learn` library, that can be installed by executing the following command in the terminal:\n",
    "\n",
    "    conda install -c conda-forge imbalanced-learn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "import timeit\n",
    "\n",
    "from imblearn.over_sampling import RandomOverSampler\n",
    "from imblearn.under_sampling import RandomUnderSampler\n",
    "\n",
    "from sklearn.base import BaseEstimator\n",
    "from sklearn.feature_extraction import DictVectorizer\n",
    "from sklearn.metrics import accuracy_score, f1_score, precision_score, recall_score, confusion_matrix\n",
    "from sklearn.model_selection import train_test_split, GridSearchCV, cross_val_score, validation_curve, learning_curve\n",
    "\n",
    "from IPython.display import display"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data pre-processing"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Read the dataset file and create a list where each element corresponds to a line of the file (that is, to a document news)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "news = []\n",
    "with open(\"SingerSplit_sorted_20k.data\", \"r\") as file:\n",
    "    news = file.readlines()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Initialize dictionaries and lists"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "labelsFreq, toRemove, y = {}, [], []"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's iterate through each news. Create the list of dictionaries that will contain the features and the list of labels.\n",
    "Remember that the format of each line is:\n",
    "\n",
    "    docId labList = feature:val feature:val feature:val ...\n",
    "    \n",
    "What we do is the following:\n",
    "    1. Iterate through each line\n",
    "    2. Remove the characters '\\n' at the end of each line\n",
    "    3. Remove the part of the 'docId', that we can identify by removing all the characters before the very first space charcater\n",
    "    4. Find the position of the equal sign in order to split labels and features\n",
    "    5. If a row does not have labels, add its position to the list containing the indexes to be removed and skip it.\n",
    "    6. Otherwise, get the labels. They are all characters before the equal sign.\n",
    "    7. Split the string of labels into a list and append them to the labels list\n",
    "    8. Update the frequency count of each label. This means looking into the dictionary that has the label as key and the times it appears as value. If the label is present, add +1 to the count. Otherwise, add the new label and assign value 1.\n",
    "    9. Get the features. They are all the characters after the equal sign.\n",
    "    10. Split the string of features first by each space sign, in order to get a list of strings like 'feature:value'.\n",
    "    11. Then split each element of the list again by the ':' sign, and add it to the dictionary of the features."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "for i in range(0,len(news)):\n",
    "    # We need this to treat the row as a string\n",
    "    s = news[i]\n",
    "    \n",
    "    # Remove trailing '\\n'\n",
    "    s = s.replace(\"\\n\", \"\") \n",
    "    \n",
    "    # Remove doc000000 \n",
    "    s = s[s.find(\" \") + 1:] \n",
    "    \n",
    "    # Find the position of the '=' sign\n",
    "    ind = s.find(\"=\")\n",
    "    \n",
    "    # Some rows do not have labels. We'll skip them\n",
    "    # and later remove them\n",
    "    if ind == 0:\n",
    "        toRemove.append(i)\n",
    "    else:\n",
    "        # Get the labels\n",
    "        lbl = s[:ind - 1]\n",
    "        lblList = lbl.split(\" \")\n",
    "        lblList = [np.int32(x) for x in lblList]\n",
    "        y.append(lblList)\n",
    "        \n",
    "        # Update the labels frequency count\n",
    "        for key in lblList:\n",
    "            if key in labelsFreq:\n",
    "                labelsFreq[key] += 1\n",
    "            else:\n",
    "                labelsFreq[key] = 1\n",
    "        \n",
    "        # Get the features\n",
    "        feats = s[ind+2:]\n",
    "        featsList1 = feats.split(\" \")\n",
    "        finalFeat = {}\n",
    "        for x in featsList1:\n",
    "            elem = x.split(\":\")\n",
    "            a = elem[0]\n",
    "            b = elem[1]\n",
    "            d = {a: b}\n",
    "            finalFeat.update(d)\n",
    "                \n",
    "        news[i] = finalFeat"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Remove the rows with no label. The `reverse` param is needed to avoid a nasty problem."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'372': '0.0322298670393407',\n",
       " '412': '0.0399666607009284',\n",
       " '1191': '0.0208579883334745',\n",
       " '1398': '0.03469461250916',\n",
       " '1696': '0.0313742944440639',\n",
       " '1788': '0.101860242861076',\n",
       " '1946': '0.0482992561844891',\n",
       " '2007': '0.0608106298756862',\n",
       " '2051': '0.0685800346592637',\n",
       " '2118': '0.0460935142795857',\n",
       " '2357': '0.064574545251984',\n",
       " '2358': '0.0679112351082636',\n",
       " '3010': '0.0302992811495597',\n",
       " '3223': '0.0535774602927466',\n",
       " '3332': '0.0528445944684083',\n",
       " '3525': '0.124761568327326',\n",
       " '3678': '0.114124916097942',\n",
       " '3821': '0.0374500120062826',\n",
       " '3943': '0.0491044772013921',\n",
       " '4582': '0.0423139885170287',\n",
       " '4660': '0.0338201197562019',\n",
       " '4754': '0.0905414297162012',\n",
       " '5065': '0.0546179961978313',\n",
       " '5152': '0.0379625776601642',\n",
       " '5371': '0.0563250723470629',\n",
       " '5747': '0.114623751792671',\n",
       " '6292': '0.0515429194037677',\n",
       " '6391': '0.0653704764924526',\n",
       " '6392': '0.0915777427420058',\n",
       " '6539': '0.0333395456778519',\n",
       " '6896': '0.0485553861036169',\n",
       " '6926': '0.00535893643734539',\n",
       " '7096': '0.049637668327603',\n",
       " '7592': '0.155827159788231',\n",
       " '7593': '0.0990756780868573',\n",
       " '7644': '0.0901770315474647',\n",
       " '7929': '0.0632968759100219',\n",
       " '7986': '0.0575839380629727',\n",
       " '7987': '0.0459485518834876',\n",
       " '8207': '0.0481563686167347',\n",
       " '8536': '0.0823745648244692',\n",
       " '8888': '0.039668902059409',\n",
       " '8889': '0.0547061916569871',\n",
       " '8937': '0.0587565120859794',\n",
       " '8938': '0.0522478792341445',\n",
       " '8975': '0.0920709637470903',\n",
       " '9058': '0.092935315117032',\n",
       " '9203': '0.0507013601536046',\n",
       " '9399': '0.0457182568877099',\n",
       " '9400': '0.0954234042373778',\n",
       " '9524': '0.0510182407654905',\n",
       " '9629': '0.0426970693955248',\n",
       " '9700': '0.149843264114026',\n",
       " '9753': '0.0351134729849058',\n",
       " '9754': '0.0263149515698',\n",
       " '9934': '0.0522879360160611',\n",
       " '10450': '0.15883772666081',\n",
       " '10565': '0.0525806787322141',\n",
       " '10665': '0.0550494618883384',\n",
       " '10666': '0.125312322958824',\n",
       " '10714': '0.0765043559014316',\n",
       " '10896': '0.037570139125719',\n",
       " '10937': '0.0346169114500481',\n",
       " '11506': '0.0830094423241606',\n",
       " '12683': '0.0494789476387753',\n",
       " '12781': '0.0537035704976116',\n",
       " '12824': '0.0941906332260423',\n",
       " '12992': '0.0423902210112844',\n",
       " '13205': '0.0597870676120206',\n",
       " '13493': '0.0381178107783271',\n",
       " '13578': '0.0219571973590385',\n",
       " '13653': '0.0534708128702748',\n",
       " '13654': '0.0477795799745177',\n",
       " '13740': '0.0927077098418291',\n",
       " '13741': '0.0405960894721612',\n",
       " '14130': '0.0570752918529302',\n",
       " '14348': '0.0482657181104484',\n",
       " '14459': '0.0152930884102708',\n",
       " '14599': '0.0574289027515529',\n",
       " '15314': '0.0472007082380623',\n",
       " '15367': '0.0409386132192889',\n",
       " '15404': '0.128169321028094',\n",
       " '15984': '0.0557406816149261',\n",
       " '16051': '0.108653626104296',\n",
       " '16087': '0.15883772666081',\n",
       " '16251': '0.0701635216131481',\n",
       " '16348': '0.0687276237301238',\n",
       " '16390': '0.0575802259663889',\n",
       " '16443': '0.0534579856964782',\n",
       " '16997': '0.0388847656672758',\n",
       " '17792': '0.16539616524162',\n",
       " '18029': '0.0496531279545364',\n",
       " '18084': '0.0481656012284332',\n",
       " '18145': '0.0373883797300678',\n",
       " '18248': '0.092783152433275',\n",
       " '18497': '0.0632998092564077',\n",
       " '18694': '0.108171731487129',\n",
       " '18771': '0.139073409713813',\n",
       " '18909': '0.105036969996568',\n",
       " '18978': '0.199301809805273',\n",
       " '19235': '0.03780568161255',\n",
       " '19488': '0.0684294467539398',\n",
       " '19838': '0.0682546323339885',\n",
       " '19870': '0.1541316980211',\n",
       " '20177': '0.0564880493729895',\n",
       " '20266': '0.0531953207720503',\n",
       " '20375': '0.0486910581854204',\n",
       " '20415': '0.0500944156817878',\n",
       " '20540': '0.0542578063870181',\n",
       " '20599': '0.0448511168146785',\n",
       " '20721': '0.132584256205316',\n",
       " '20849': '0.101374880630975',\n",
       " '20908': '0.0302008803866204',\n",
       " '20976': '0.0731187497727314',\n",
       " '21424': '0.0758417866843585',\n",
       " '21425': '0.0543334550623888',\n",
       " '21456': '0.042467492094287',\n",
       " '21892': '0.0593611170944157',\n",
       " '21958': '0.0682650087538082',\n",
       " '22002': '0.0818957374622157',\n",
       " '22402': '0.0600007526487241',\n",
       " '22711': '0.0523193133383312',\n",
       " '22757': '0.052289189708851',\n",
       " '22817': '0.0265905641212986',\n",
       " '22878': '0.0587423455348181',\n",
       " '23622': '0.0561292489810793',\n",
       " '23761': '0.0820270582567047',\n",
       " '23876': '0.0555717224541103',\n",
       " '24051': '0.101222084890645',\n",
       " '24134': '0.0539182956923293',\n",
       " '24135': '0.0530937992300409',\n",
       " '24190': '0.0510353671778295',\n",
       " '24191': '0.119472287435434',\n",
       " '24661': '0.0965056723121174',\n",
       " '24791': '0.0743665302277051',\n",
       " '25032': '0.0428787168576463',\n",
       " '25090': '0.0720415543572573',\n",
       " '25112': '0.111644796892333',\n",
       " '25143': '0.0807011998427138',\n",
       " '25351': '0.150997657369086',\n",
       " '25414': '0.062270942731703',\n",
       " '25415': '0.0855691328130595',\n",
       " '25464': '0.0548100722607414',\n",
       " '25704': '0.0603243511981',\n",
       " '26109': '0.0510811441831462',\n",
       " '26345': '0.0614108918256086',\n",
       " '26834': '0.0328191217718412',\n",
       " '26901': '0.0646261337776922',\n",
       " '26959': '0.0310263060676379',\n",
       " '27258': '0.08309525933739',\n",
       " '27375': '0.0981280272470351',\n",
       " '27450': '0.0530194796994856',\n",
       " '27950': '0.038085395843702',\n",
       " '28227': '0.156257152986428',\n",
       " '28512': '0.248004554322686',\n",
       " '28665': '0.0519157725701903',\n",
       " '28666': '0.031366918705922',\n",
       " '28753': '0.0371169095058771',\n",
       " '29108': '0.0559249092963845',\n",
       " '29109': '0.038857867759734',\n",
       " '29110': '0.0675373343490071',\n",
       " '29193': '0.0446425182353309',\n",
       " '29615': '0.079660964032846',\n",
       " '29616': '0.0515346179659248',\n",
       " '29932': '0.0111804497768424',\n",
       " '30864': '0.0912361646253409',\n",
       " '31184': '0.0317588430378785',\n",
       " '31416': '0.108410527366106',\n",
       " '31642': '0.0812155282702178'}"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "[news.pop(i) for i in sorted(toRemove, reverse=True)]\n",
    "news[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's get the 4 most present labels. Dict comprehension taken from: https://stackoverflow.com/a/40777805"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[33, 70, 101, 4]"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "highest = list(({key: labelsFreq[key] for key in sorted(labelsFreq, key=labelsFreq.get, reverse=True)[:4]}).keys())\n",
    "highest"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Create the 4 list containing the labels:\n",
    "- 1 if the element contains the most frequent label\n",
    "- 0 otherwise"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "y1, y2, y3, y4 = [], [], [], []\n",
    "\n",
    "for i in range(1, 5):\n",
    "    for row in y:\n",
    "        if highest[i-1] in row:\n",
    "            eval('y'+str(i)).append(int(1))\n",
    "        else:\n",
    "            eval('y'+str(i)).append(int(-1))\n",
    "            \n",
    "y1 = np.asarray(y1)\n",
    "y2 = np.asarray(y2)\n",
    "y3 = np.asarray(y3)\n",
    "y4 = np.asarray(y4)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Check the distribution of the labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYcAAAE7CAYAAAA2Fk1wAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAHd9JREFUeJzt3Xu0nXV95/H3h2tElEAIIDlgoqaO4FiqEbDtdPBSwEsNzpIWl0qqaGwnts70olBXByvSYsdVLPVWRlJArRHxQpbGpqmKjktuQbwB0qSo5BiQkHDxBgJ+54/9xNmcZ5+ck7M32eecvF9r7ZVnf5/f8zzfZ4v7c57L3jtVhSRJ3fYYdgOSpOnHcJAktRgOkqQWw0GS1GI4SJJaDAdJUovhoBklyZVJXrerlx2znu8leUEz/RdJPtjvOrvW/eMkT2qmL07yjgGu+wNJ/nJQ69PsZjhoKLrfYGeyqvrrqpowcCYbTFW1f1Xd2m9fSX4/yVfGrPsPquqcftet3YPhIE0DSfYadg9SN8NB00qSA5N8JsmWJHc30yNjhj05ybVJ7k1yRZKDupY/PslXk9yT5BtJThhnO09J8qVmHXcl+dgOenp1ku8n2ZrkrWPmvS3Jh5vpOUk+3Iy7J8l1SQ5Nci7wX4D3NKeN3tOMryQrkmwANnTVntK1iYOTrEvyo6bfJzbjFjZj9+rq5cokr0vyNOADwHOa7d3TzH/Eaaokr0+yMcm2JKuTHN41r5L8QZINzf8O702S8V4jzT6Gg6abPYB/Ap4IHAn8DHjPmDGnA68FDgceAi4ASLIA+CzwDuAg4M+ATySZ32M75wD/ChwIjAD/0KuZJEcB7wde3WxvXjO+l2XAAcARzbg/AH5WVW8F/i/wxua00Ru7ljkFOA44apx1vrLp9WDg68BHxhn3S1V1c7Ptq5rtze2xX88D/gb4XeAJwPeBVWOGvQR4NvCrzbiTJtq2Zg/DQdNKVW2tqk9U1U+r6kfAucB/HTPsQ1X17ar6CfCXwO8m2RN4FbCmqtZU1S+qah2wHnhRj009SCeADq+q+6vqKz3GALwc+ExVfbmqHmi294txxj5IJxSeUlUPV9X1VXXfBLv8N1W1rap+Ns78z3Zt+610jgaOmGCdk/FKYGVVfa1Z91nNuhd2jTmvqu6pqtuALwLHDGC7miEMB00rSfZL8o/NaZz7gC8Dc5s3/+02dU1/H9ibzl/WTwRObU7p3NOcTvlNOn8Zj/VmIMC1SW5M8tpxWjq8e3tNIG0dZ+yHgLXAqiSbk/xtkr0n2OVNk51fVT8GtjU99etwOq9d97q3Agu6xtzRNf1TYP8BbFczhOGg6eZPgacCx1XV44Hfaurd57u7/3I+ks5f7HfReSP9UFXN7Xo8tqrOG7uRqrqjql5fVYcDbwDeN+Zc/3a3d28vyX50jg5aqurBqvqrqjoK+HU6p2VO3z57nP2d6GuRu7e9P53TZZuBnzTl/brGHrYT691MJ0y3r/uxdPbrBxMsp92E4aBh2ru5iLv9sRfwODrXGe5pLjSf3WO5VyU5qnmjfjtweVU9DHwY+J0kJyXZs1nnCT0uaJPk1K763XTeTB/usa3LgZck+c0k+zTb6/n/myTPTfKfm6Oc++iE1vZ1/hB40qRelUd6Ude2zwGuqapNVbWFzhv5q5p9fS3w5K7lfgiMNMv18s/Aa5Ick2Rf4K+bdX9vCj1qFjIcNExr6ATB9sfbgHcDj6FzJHA18C89lvsQcDGd0x5zgD8GqKpNwFLgL4AtdI4k/pze/50/G7gmyY+B1cCbquq7YwdV1Y3ACjpvprfTCZLRcfbnMDphch9wM/AlOoEF8PfAy5s7fy4YZ/le/plOQG4DnkXnWsF2r2/2bytwNPDVrnlfAG4E7khyV4/9+jyd6yefaPbrycBpO9GXZrn4Yz+SpLE8cpAktRgOkqQWw0GS1GI4SJJaDAdJUsuM/SbIgw8+uBYuXDjsNiRpRrn++uvvqqpe3zf2CDM2HBYuXMj69euH3YYkzShJvj/xKE8rSZJ6MBwkSS2GgySpZcZec+jlwQcfZHR0lPvvv3/YrUzZnDlzGBkZYe+9J/qmZ0l69MyqcBgdHeVxj3scCxcuZCb+omFVsXXrVkZHR1m0aNGw25G0G5tVp5Xuv/9+5s2bNyODASAJ8+bNm9FHPpJmh1kVDsCMDYbtZnr/kmaHWRcO09nHP/5xjj76aPbYYw8/oyFpWptV1xzGWnjmZwe6vu+d9+JJj73yyiu5+OKLufjii39Ze/rTn84nP/lJ3vCGNwy0L0katFkdDtPN0572tGG3IEmTYjhI0iC87YBHYZ33Dn6dkzThNYckK5PcmeTbY+p/lOSWJDcm+duu+llJNjbzTuqqn9zUNiY5s6u+KMk1STYk+dgOfhB9RjjuuOM45phjeN3rXsfq1as55phjOOaYY1i7du2wW5OkSZvMkcPFwHuAS7cXkjyXzg+5P6OqHkhySFM/is6PlB8NHA78W5JfaRZ7L/DbdH6c/bokq6vqJuCdwPlVtSrJB4AzgPcPYueG4ZprrgF6X3OQpJliwiOHqvoysG1M+Q+B86rqgWbMnU19KbCqqh6oqu8CG4Fjm8fGqrq1qn4OrAKWpnPf5vOAy5vlLwFO6XOfJEl9muqtrL8C/JfmdNCXkjy7qS8ANnWNG21q49XnAfdU1UNj6rPSpz71KUZGRrjqqqt48YtfzEknnTTxQpI0BFO9IL0XcCBwPPBs4LIkTwJ6fYKr6B1CtYPxPSVZDiwHOPLIIydscmduPR20E044gRNOOOERtZe97GW87GUvG05DkrQTpnrkMAp8sjquBX4BHNzUj+gaNwJs3kH9LmBukr3G1HuqqguraklVLZk/f8IfMpIkTdFUw+HTdK4V0Fxw3ofOG/1q4LQk+yZZBCwGrgWuAxY3dybtQ+ei9eqqKuCLwMub9S4DrpjqzkiSBmPC00pJPgqcABycZBQ4G1gJrGxub/05sKx5o78xyWXATcBDwIqqerhZzxuBtcCewMqqurHZxFuAVUneAdwAXDTA/ZMkTcGE4VBVrxhn1qvGGX8ucG6P+hpgTY/6rXTuZpIkTRN+8Z4kqcVwkCS1GA6Pku985zs85znPYd999+Vd73rXsNuRpJ0yu794b9BfhLUTX4J10EEHccEFF/DpT396sD1I0i7gkcOj5JBDDuHZz342e++997BbkaSdZjhIkloMB0lSi+EwQO9973t/+fsNmzeP+y0gkjTtze4L0rvYihUrWLFixbDbkKS+GQ6PkjvuuIMlS5Zw3333sccee/Dud7+bm266icc//vHDbk2SJjS7w2GIv7962GGHMTo6OrTtSxOaZb95rMHymoMkqcVwkCS1GA6SpJZZFw6dn5WYuWZ6/5Jmh1kVDnPmzGHr1q0z9g22qti6dStz5swZdiuSdnOz6m6lkZERRkdH2bJly7BbmbI5c+YwMjIy7DYk7eZmVTjsvffeLFq0aNhtSNKMN6tOK0mSBsNwkCS1TBgOSVYmuTPJt3vM+7MkleTg5nmSXJBkY5JvJnlm19hlSTY0j2Vd9Wcl+VazzAVJMqidkyRNzWSuOVwMvAe4tLuY5Ajgt4HbusovBBY3j+OA9wPHJTkIOBtYAhRwfZLVVXV3M2Y5cDWwBjgZ+NzUd0mzll/3IO0yEx45VNWXgW09Zp0PvJnOm/12S4FLq+NqYG6SJwAnAeuqalsTCOuAk5t5j6+qq6pz/+mlwCn97ZIkqV9TuuaQ5KXAD6rqG2NmLQA2dT0fbWo7qo/2qEuShminb2VNsh/wVuDEXrN71GoK9fG2vZzOKSiOPPLICXuVJE3NVI4cngwsAr6R5HvACPC1JIfR+cv/iK6xI8DmCeojPeo9VdWFVbWkqpbMnz9/Cq1LkiZjp8Ohqr5VVYdU1cKqWkjnDf6ZVXUHsBo4vblr6Xjg3qq6HVgLnJjkwCQH0jnqWNvM+1GS45u7lE4HrhjQvkmSpmgyt7J+FLgKeGqS0SRn7GD4GuBWYCPwf4D/DlBV24BzgOuax9ubGsAfAh9slvkPvFNJkoZuwmsOVfWKCeYv7JouoOePKFfVSmBlj/p64OkT9SFJ2nX8hLQkqcVwkCS1GA6SpBbDQZLUYjhIkloMB0lSi+EgSWoxHCRJLYaDJKnFcJAktRgOkqQWw0GS1GI4SJJaDAdJUovhIElqMRwkSS2GgySpxXCQJLUYDpKklgnDIcnKJHcm+XZX7X8n+U6Sbyb5VJK5XfPOSrIxyS1JTuqqn9zUNiY5s6u+KMk1STYk+ViSfQa5g5KknTeZI4eLgZPH1NYBT6+qZwD/DpwFkOQo4DTg6GaZ9yXZM8mewHuBFwJHAa9oxgK8Ezi/qhYDdwNn9LVHkqS+TRgOVfVlYNuY2r9W1UPN06uBkWZ6KbCqqh6oqu8CG4Fjm8fGqrq1qn4OrAKWJgnwPODyZvlLgFP63CdJUp8Gcc3htcDnmukFwKaueaNNbbz6POCerqDZXpckDVFf4ZDkrcBDwEe2l3oMqynUx9ve8iTrk6zfsmXLzrYrSZqkKYdDkmXAS4BXVtX2N/RR4IiuYSPA5h3U7wLmJtlrTL2nqrqwqpZU1ZL58+dPtXVJ0gSmFA5JTgbeAry0qn7aNWs1cFqSfZMsAhYD1wLXAYubO5P2oXPRenUTKl8EXt4svwy4Ymq7IkkalMncyvpR4CrgqUlGk5wBvAd4HLAuydeTfACgqm4ELgNuAv4FWFFVDzfXFN4IrAVuBi5rxkInZP4kyUY61yAuGugeSpJ22l4TDaiqV/Qoj/sGXlXnAuf2qK8B1vSo30rnbiZJ0jThJ6QlSS2GgySpxXCQJLUYDpKkFsNBktRiOEiSWgwHSVKL4SBJajEcJEkthoMkqWXCr89Qn952wIDXd+9g1ydJPXjkIElqMRwkSS2GgySpxXCQJLUYDpKkFsNBktRiOEiSWgwHSVKL4SBJapkwHJKsTHJnkm931Q5Ksi7JhubfA5t6klyQZGOSbyZ5Ztcyy5rxG5Is66o/K8m3mmUuSJJB76QkaedM5sjhYuDkMbUzgc9X1WLg881zgBcCi5vHcuD90AkT4GzgOOBY4OztgdKMWd613NhtSZJ2sQnDoaq+DGwbU14KXNJMXwKc0lW/tDquBuYmeQJwErCuqrZV1d3AOuDkZt7jq+qqqirg0q51SZKGZKrXHA6tqtsBmn8PaeoLgE1d40ab2o7qoz3qkqQhGvQF6V7XC2oK9d4rT5YnWZ9k/ZYtW6bYoiRpIlMNhx82p4Ro/r2zqY8CR3SNGwE2T1Af6VHvqaourKolVbVk/vz5U2xdkjSRqYbDamD7HUfLgCu66qc3dy0dD9zbnHZaC5yY5MDmQvSJwNpm3o+SHN/cpXR617okSUMy4Y/9JPkocAJwcJJROncdnQdcluQM4Dbg1Gb4GuBFwEbgp8BrAKpqW5JzgOuacW+vqu0Xuf+Qzh1RjwE+1zwkSUM0YThU1SvGmfX8HmMLWDHOelYCK3vU1wNPn6gPSdKu4yekJUkthoMkqcVwkCS1GA6SpBbDQZLUYjhIkloMB0lSi+EgSWoxHCRJLYaDJKllwq/P2J0sPPOzA1/n9+YMfJWS9KjzyEGS1GI4SJJaDAdJUovhIElqMRwkSS2GgySpxXCQJLUYDpKklr7CIcn/THJjkm8n+WiSOUkWJbkmyYYkH0uyTzN23+b5xmb+wq71nNXUb0lyUn+7JEnq15TDIckC4I+BJVX1dGBP4DTgncD5VbUYuBs4o1nkDODuqnoKcH4zjiRHNcsdDZwMvC/JnlPtS5LUv35PK+0FPCbJXsB+wO3A84DLm/mXAKc000ub5zTzn58kTX1VVT1QVd8FNgLH9tmXJKkPUw6HqvoB8C7gNjqhcC9wPXBPVT3UDBsFFjTTC4BNzbIPNePnddd7LCNJGoJ+TisdSOev/kXA4cBjgRf2GFrbFxln3nj1XttcnmR9kvVbtmzZ+aYlSZPSz2mlFwDfraotVfUg8Eng14G5zWkmgBFgczM9ChwB0Mw/ANjWXe+xzCNU1YVVtaSqlsyfP7+P1iVJO9JPONwGHJ9kv+bawfOBm4AvAi9vxiwDrmimVzfPaeZ/oaqqqZ/W3M20CFgMXNtHX5KkPk359xyq6poklwNfAx4CbgAuBD4LrEryjqZ2UbPIRcCHkmykc8RwWrOeG5NcRidYHgJWVNXDU+1LktS/vn7sp6rOBs4eU76VHncbVdX9wKnjrOdc4Nx+epEkDY6fkJYktRgOkqQWw0GS1GI4SJJaDAdJUovhIElqMRwkSS2GgySpxXCQJLUYDpKkFsNBktRiOEiSWgwHSVKL4SBJajEcJEkthoMkqcVwkCS1GA6SpBbDQZLUYjhIklr6Cockc5NcnuQ7SW5O8pwkByVZl2RD8++BzdgkuSDJxiTfTPLMrvUsa8ZvSLKs352SJPWn3yOHvwf+par+E/CrwM3AmcDnq2ox8PnmOcALgcXNYznwfoAkBwFnA8cBxwJnbw8USdJwTDkckjwe+C3gIoCq+nlV3QMsBS5phl0CnNJMLwUurY6rgblJngCcBKyrqm1VdTewDjh5qn1JkvrXz5HDk4AtwD8luSHJB5M8Fji0qm4HaP49pBm/ANjUtfxoUxuvLkkakn7CYS/gmcD7q+rXgJ/w/08h9ZIetdpBvb2CZHmS9UnWb9myZWf7lSRNUj/hMAqMVtU1zfPL6YTFD5vTRTT/3tk1/oiu5UeAzTuot1TVhVW1pKqWzJ8/v4/WJUk7MuVwqKo7gE1JntqUng/cBKwGtt9xtAy4opleDZze3LV0PHBvc9ppLXBikgObC9EnNjVJ0pDs1efyfwR8JMk+wK3Aa+gEzmVJzgBuA05txq4BXgRsBH7ajKWqtiU5B7iuGff2qtrWZ1+SpD70FQ5V9XVgSY9Zz+8xtoAV46xnJbCyn14kSYPT75GD1NPCMz878HV+b87AVylpHH59hiSpxXCQJLUYDpKkFsNBktRiOEiSWgwHSVKL4SBJajEcJEkthoMkqcVwkCS1+PUZ0gzg15FoV/PIQZLUYjhIkloMB0lSi+EgSWoxHCRJLYaDJKnFcJAktfQdDkn2THJDks80zxcluSbJhiQfS7JPU9+3eb6xmb+wax1nNfVbkpzUb0+SpP4M4sjhTcDNXc/fCZxfVYuBu4EzmvoZwN1V9RTg/GYcSY4CTgOOBk4G3pdkzwH0JUmaor7CIckI8GLgg83zAM8DLm+GXAKc0kwvbZ7TzH9+M34psKqqHqiq7wIbgWP76UuS1J9+jxzeDbwZ+EXzfB5wT1U91DwfBRY00wuATQDN/Hub8b+s91hGkjQEUw6HJC8B7qyq67vLPYbWBPN2tMzYbS5Psj7J+i1btuxUv5KkyevnyOE3gJcm+R6wis7ppHcDc5Ns/0K/EWBzMz0KHAHQzD8A2NZd77HMI1TVhVW1pKqWzJ8/v4/WJUk7MuVwqKqzqmqkqhbSuaD8hap6JfBF4OXNsGXAFc306uY5zfwvVFU19dOau5kWAYuBa6falySpf4/GV3a/BViV5B3ADcBFTf0i4ENJNtI5YjgNoKpuTHIZcBPwELCiqh5+FPqSJE3SQMKhqq4Ermymb6XH3UZVdT9w6jjLnwucO4heJEn98xPSkqQWw0GS1GI4SJJa/A1pSbsdf5N7Yh45SJJaDAdJUovhIElqMRwkSS2GgySpxXCQJLUYDpKkFsNBktRiOEiSWgwHSVKL4SBJajEcJEkthoMkqcVwkCS1GA6SpJYph0OSI5J8McnNSW5M8qamflCSdUk2NP8e2NST5IIkG5N8M8kzu9a1rBm/Icmy/ndLktSPfo4cHgL+tKqeBhwPrEhyFHAm8PmqWgx8vnkO8EJgcfNYDrwfOmECnA0cBxwLnL09UCRJwzHlcKiq26vqa830j4CbgQXAUuCSZtglwCnN9FLg0uq4Gpib5AnAScC6qtpWVXcD64CTp9qXJKl/A7nmkGQh8GvANcChVXU7dAIEOKQZtgDY1LXYaFMbry5JGpK+wyHJ/sAngP9RVfftaGiPWu2g3mtby5OsT7J+y5YtO9+sJGlS+gqHJHvTCYaPVNUnm/IPm9NFNP/e2dRHgSO6Fh8BNu+g3lJVF1bVkqpaMn/+/H5alyTtQD93KwW4CLi5qv6ua9ZqYPsdR8uAK7rqpzd3LR0P3NucdloLnJjkwOZC9IlNTZI0JHv1sexvAK8GvpXk603tL4DzgMuSnAHcBpzazFsDvAjYCPwUeA1AVW1Lcg5wXTPu7VW1rY++JEl9mnI4VNVX6H29AOD5PcYXsGKcda0EVk61F0nSYPkJaUlSi+EgSWoxHCRJLYaDJKnFcJAktRgOkqQWw0GS1GI4SJJaDAdJUovhIElqMRwkSS2GgySpxXCQJLUYDpKkFsNBktRiOEiSWgwHSVKL4SBJajEcJEkthoMkqWXahEOSk5PckmRjkjOH3Y8k7c6mRTgk2RN4L/BC4CjgFUmOGm5XkrT7mhbhABwLbKyqW6vq58AqYOmQe5Kk3dZ0CYcFwKau56NNTZI0BHsNu4FGetSqNShZDixvnv44yS2PalcDEDgYuGtgK/yrXi/V7mHgryX4evp6DswMej2fOJlB0yUcRoEjup6PAJvHDqqqC4ELd1VTg5BkfVUtGXYfs4Gv5WD5eg7WbHs9p8tppeuAxUkWJdkHOA1YPeSeJGm3NS2OHKrqoSRvBNYCewIrq+rGIbclSbutaREOAFW1Blgz7D4eBTPqNNg052s5WL6egzWrXs9Uta77SpJ2c9PlmoMkaRoxHCRJLYaDJPUpyaXD7mHQps0F6ZkuyX7AG+l8eO8f6NyO+9+A7wBvr6ofD7G9GSfJAcBZwCnA/KZ8J3AFcF5V3TOs3rR7SzL2NvsAz00yF6CqXrrruxo8L0gPSJLL6HwFyGOApwI3A5cBvwMcVlWvHmJ7M06StcAXgEuq6o6mdhiwDHhBVf32MPubTZLs7x8vk5fka8BNwAfp/DEY4KN0/iCkqr40vO4Gx3AYkCRfr6pjkgS4HXhCVVXz/BtV9YwhtzijJLmlqp66s/O085LcVlVHDruPmSLJHsCbgBcBf15VX09ya1U9acitDZSnlQasCYQ11aRu89wE3nnfT/JmOkcOPwRIcijw+zzySxo1CUn+ZLxZwP67speZrqp+AZyf5OPNvz9kFr6XekF6cNYn2R+gql67vZjkycCPhtbVzPV7wDzgS0nuTrINuBI4CPjdYTY2Q/01cCDwuDGP/fF9YEqqarSqTgU+B3x42P0MmqeVBijJsXQOFq5rfqzoZOAW4JdHEpqcJH8MfKqqPEoYgCRfBf6oqq7vMW9TVR3RYzHtxgyHAUlyNp1fstsLWAccR+cv3RcAa6vq3OF1N/MkuRf4CfAfwD8DH6+qwX4d8m4kyVOBbVW1pce8Q7efupO2MxwGJMm3gGOAfYE7gJGqui/JY4BrvCC9c5LcADyLTrj+HvBS4Ho6d4V8sqo8VSc9ijzXODgPVdXDVfVT4D+q6j6AqvoZ8IvhtjYjVVX9oqr+tarOAA4H3kfnVN2tw21t5klyQJLzknwnydbmcXNTmzvs/jT9GA6D8/Pmg3DQ+YsX+OWHuQyHnfeIn8CqqgeranVVvQLwtsuddxlwN3BCVc2rqnnAc5vax4famaYlTysNSJJ9q+qBHvWD6Xzm4VtDaGvGSvIrVfXvw+5jtvBzI9pZHjkMSK9gaOp3GQw7z2AYuO8neXPzWRGgcyE6yVvwcyPqwXCQdg/dnxvZNuZzI6cOszFNT55WknZzSV5TVf807D40vRgO0m7O71ZSL7Pu+0AktSX55nizgEPHmafdmOEg7R4OBU6ic+tqtwBf3fXtaLozHKTdw2eA/avq62NnJLly17ej6c5rDpKkFm9llSS1GA6SpBbDQZpAkkn/vnKStyX5s0dr/dKuYjhIkloMB2kKkvxOkmuS3JDk37q/swj41SRfSLIhyeu7lvnzJNcl+WaSvxpC29KkGQ7S1HwFOL6qfg1YBby5a94zgBcDzwH+V5LDk5wILAaOpfOjUM9K8lu7uGdp0vycgzQ1I8DHkjwB2Af4bte8K5ofefpZki/SCYTfBE4EbmjG7E8nLL6861qWJs9wkKbmH4C/q6rVSU4A3tY1b+yHh4rOJ5H/pqr+cde0J/XH00rS1BwA/KCZXjZm3tIkc5LMA04ArgPWAq9Nsj9AkgVJDtlVzUo7yyMHaWL7JRntev53dI4UPp7kB8DVwKKu+dcCn6Xzc6bnVNVmYHOSpwFXJQH4MfAq4M5Hv31p5/n1GZKkFk8rSZJaDAdJUovhIElqMRwkSS2GgySpxXCQJLUYDpKkFsNBktTy/wAlrPr4i+V5WgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x324 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "pos, neg = [], []\n",
    "for i in range(1,5):\n",
    "    index, count = np.unique(eval('y'+str(i)), return_counts=True)\n",
    "    pos.append(np.int32(count[1]))\n",
    "    neg.append(np.int32(count[0]))\n",
    "\n",
    "df = pd.DataFrame(np.column_stack([highest, pos, neg]), columns=['Label','+1','-1'])\n",
    "fig = df.plot(x=\"Label\", y=[\"+1\", \"-1\"], kind=\"bar\", title='Labels distribution', figsize=(6, 4.5)).get_figure()\n",
    "fig.savefig(\"label_distribution.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "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>Label</th>\n",
       "      <th>+1</th>\n",
       "      <th>-1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>33</td>\n",
       "      <td>8899</td>\n",
       "      <td>10961</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>70</td>\n",
       "      <td>5634</td>\n",
       "      <td>14226</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>101</td>\n",
       "      <td>4874</td>\n",
       "      <td>14986</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>3465</td>\n",
       "      <td>16395</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label    +1     -1\n",
       "0     33  8899  10961\n",
       "1     70  5634  14226\n",
       "2    101  4874  14986\n",
       "3      4  3465  16395"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can clearly see that we are dealing with inbalanced classes, except for the first label. Thus, we'll use _F1_ as our main metric. We'll also check other metrics just to be sure for comparison. Moreover, we'll upsample and downsample to see if this helps the classifier"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The **F1 score** can be interpreted as a weighted average of the precision and recall:\n",
    "\n",
    "$$F1 = 2 \\cdot \\frac{(precision \\cdot recall)}{(precision + recall)} $$\n",
    "\n",
    "where **precision** can be seen as \"the ability of the classifier not to label as positive a sample that is negative\":\n",
    "\n",
    "$$ precision=\\frac{true\\ positive}{true\\ postive + false\\ positive} $$\n",
    "\n",
    "and **recall** can be seen as \"the ability of the classifier to find all the positive samples\":\n",
    "\n",
    "$$ recall=\\frac{true\\ positive}{true\\ postive + false\\ negative} $$"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We'll deal with this while implementing the `MyBaseClassifier` from which both `Pegasos` and the `Perceptron` will inherit."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's do the last steps of our data pre-processing. Convert features to `int` and values to `float`. Dict comprehension taken from: https://stackoverflow.com/a/5317657"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "news = [dict([int(a), float(x)] for a, x in elem.items()) for elem in news]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Transforms lists of feature-value mappings to a CSR matrix. Setting the `dtype` param to `np.float32` cuts memory usage by 50%!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<19860x80362 sparse matrix of type '<class 'numpy.float32'>'\n",
       "\twith 1858673 stored elements in Compressed Sparse Row format>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "v = DictVectorizer(sparse=True, dtype=np.float32)\n",
    "X = v.fit_transform(news)\n",
    "X"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Since our data is very sparse, using a CSR matrix instead of a dense one (or a classical Pandas DataFrame) allows us to save a lot of memory, as we can see below.\n",
    "\n",
    "**Warning**: the execution of this cell requires more than 6GB of memory (either RAM or swap). If you don't have them, you can simply skip this cell."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dense:  6383.96 MB\n",
      "Sparse:  14.95 MB\n"
     ]
    }
   ],
   "source": [
    "dense_size = np.round(np.array(X.toarray()).nbytes/1e6, 2)\n",
    "sparse_size = np.round((X.data.nbytes + X.indptr.nbytes + X.indices.nbytes)/1e6, 2)\n",
    "\n",
    "print(\"Dense: \", dense_size, \"MB\")\n",
    "print(\"Sparse: \", sparse_size, \"MB\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Free up some memory"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "del news\n",
    "del toRemove\n",
    "del labelsFreq"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Pegasos"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here there's our own Pegasos implementation. It inherits from the `BaseEstimator` so that we don't need to define the `get_params` and `set_params` methods. Moreover, we created a class `MyBaseClassifier` that contains the `predict` and `score` methods. Both our Pegagos and Perceptron implementations will inherit from this class so that, in the end, we can use them in the `GridSearchCV`. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MyBaseClassifier():\n",
    "    def sign(self, X):\n",
    "        return -1. if X <= 0 else 1.\n",
    "    \n",
    "    def predict(self, X):\n",
    "        pred = X.dot(self.w.T)\n",
    "        return [self.sign(pred[i]) for i in range(len(pred))]\n",
    "    \n",
    "    def score(self, X, y):\n",
    "        y_pred = self.predict(X)\n",
    "        return f1_score(y, y_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Pegasos(BaseEstimator, MyBaseClassifier):\n",
    "    def __init__(self, lmbd=0.001, T=1000):\n",
    "        self.T = T\n",
    "        self.lmbd = lmbd\n",
    "        \n",
    "    def fit(self, X, y):\n",
    "        # Initialize w vector to zero\n",
    "        self.w = np.zeros(X.shape[1])\n",
    "    \n",
    "        for t in range(1, self.T + 1):\n",
    "            # We pick a random position i in the training set.\n",
    "            i = np.random.randint(0, X.shape[0]-1)\n",
    "\n",
    "            # Step length in gradient descent\n",
    "            eta = 1/(self.lmbd*t)\n",
    "\n",
    "            # Compute the score, that is the inner product\n",
    "            # between w and the current training example\n",
    "            score = X[i].dot(self.w.T)\n",
    "\n",
    "            # Check if we got it right\n",
    "            if y[i]*score < 1:\n",
    "                self.w = (1 - eta*self.lmbd)*self.w + (eta*y[i])*X[i]\n",
    "            else:\n",
    "                self.w = (1 - eta*self.lmbd)*self.w"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Label 1: 33"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Since we have to train 4 classifiers for the 4 most frequent labels, create this variable where we set what label we're training our classifier to recognize."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "currentLabel = y1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's now get a first training/test split, where we allocate a third of the dataset as test set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(X,currentLabel,test_size=0.33,random_state=42,stratify=currentLabel)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Run Pegasos for the first time. We use the default values for `lmbd=0.001` and `T=1000` that we set in our implementation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "classifier = Pegasos()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "classifier.fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.8711993617587093, 0.8516624040920716)"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "classifier.score(X_train, y_train), classifier.score(X_test, y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now check other metrics:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "True positive:  2331\n",
      "False positive:  206\n",
      "FP ratio:  0.08119826566811195 ( 8.12 %)\n",
      "True negative:  3411\n",
      "False negative:  606\n",
      "FN ratio:  0.1508588498879761 ( 15.09 %)\n",
      "Precision:  0.918801734331888 ( 91.88 %)\n",
      "Recall:  0.7936670071501533 ( 79.37 %)\n",
      "Accuracy:  0.8761061946902655 ( 87.61 %)\n"
     ]
    }
   ],
   "source": [
    "TPs, FPs, TNs, FNs, precisions, recalls, accs = [], [], [], [], [], [], []\n",
    "y_pred = classifier.predict(X_test)\n",
    "\n",
    "tn, fp, fn, tp = confusion_matrix(y_test, y_pred).ravel()\n",
    "precision = precision_score(y_test, y_pred)\n",
    "recall = recall_score(y_test, y_pred)\n",
    "acc = accuracy_score(y_test, y_pred)\n",
    "    \n",
    "TNs.append(tn)\n",
    "TPs.append(tp)\n",
    "FPs.append(fp)\n",
    "FNs.append(fn)\n",
    "precisions.append(precision)\n",
    "recalls.append(recall)\n",
    "accs.append(acc)\n",
    "    \n",
    "print(\"True positive: \", tp)\n",
    "print(\"False positive: \", fp)\n",
    "print(\"FP ratio: \", fp/(tp+fp),\"(\",np.round(fp/(tp+fp)*100, 2),\"%)\")\n",
    "print(\"True negative: \", tn)\n",
    "print(\"False negative: \", fn)\n",
    "print(\"FN ratio: \", fn/(tn+fn),\"(\",np.round(fn/(tn+fn)*100, 2),\"%)\")\n",
    "print(\"Precision: \", precision,\"(\",np.round(precision*100, 2),\"%)\")\n",
    "print(\"Recall: \", recall,\"(\",np.round(recall*100, 2),\"%)\")\n",
    "print(\"Accuracy: \", acc,\"(\",np.round(acc*100, 2),\"%)\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Pegasos doesn't always deliver the same results because it depends on the order of which the training examples are presented. Therefore, we run it 100 times and we average the results."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.8768452910151163, 0.8654436843812745)"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results_train, results_test = [], []\n",
    "for i in range(100):\n",
    "    classifier.fit(X_train, y_train)\n",
    "    results_train.append(classifier.score(X_train, y_train))\n",
    "    results_test.append(classifier.score(X_test, y_test))\n",
    "\n",
    "np.mean(results_train), np.mean(results_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We don't want to stick to a single split. Let's see the learning curve. We use a 10-fold cross validation because we read a paper that suggests so :)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[learning_curve] Training set sizes: [ 1000  2000  3000  4000  5000  6000  7000  8000  9000 10000 11000 12000\n",
      " 13000]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   5 out of 130 | elapsed:    2.6s remaining:  1.1min\n",
      "[Parallel(n_jobs=-1)]: Done  19 out of 130 | elapsed:    5.0s remaining:   29.1s\n",
      "[Parallel(n_jobs=-1)]: Done  33 out of 130 | elapsed:    7.3s remaining:   21.5s\n",
      "[Parallel(n_jobs=-1)]: Done  47 out of 130 | elapsed:    9.2s remaining:   16.3s\n",
      "[Parallel(n_jobs=-1)]: Done  61 out of 130 | elapsed:   11.5s remaining:   13.0s\n",
      "[Parallel(n_jobs=-1)]: Done  75 out of 130 | elapsed:   13.9s remaining:   10.2s\n",
      "[Parallel(n_jobs=-1)]: Done  89 out of 130 | elapsed:   16.1s remaining:    7.4s\n",
      "[Parallel(n_jobs=-1)]: Done 103 out of 130 | elapsed:   18.3s remaining:    4.8s\n",
      "[Parallel(n_jobs=-1)]: Done 117 out of 130 | elapsed:   20.6s remaining:    2.3s\n",
      "[Parallel(n_jobs=-1)]: Done 130 out of 130 | elapsed:   22.2s finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[0.82259887, 0.82635399, 0.84102564, 0.83686319, 0.83544304,\n",
       "        0.81397006, 0.82045322, 0.82388359, 0.76092233, 0.81333333],\n",
       "       [0.82436428, 0.84130688, 0.8658726 , 0.8730065 , 0.88998156,\n",
       "        0.82094082, 0.84533648, 0.86915888, 0.83473054, 0.83949045],\n",
       "       [0.84270953, 0.85431235, 0.86015831, 0.84441733, 0.88847584,\n",
       "        0.83370619, 0.85697875, 0.86197779, 0.83610451, 0.82490272],\n",
       "       [0.84982538, 0.83775811, 0.89406994, 0.85645356, 0.89783654,\n",
       "        0.82281369, 0.84542212, 0.86475196, 0.85681293, 0.85215221],\n",
       "       [0.87185104, 0.86021505, 0.86515071, 0.87478754, 0.89444783,\n",
       "        0.8324568 , 0.87681971, 0.85287846, 0.88690808, 0.82398957],\n",
       "       [0.84324324, 0.85859729, 0.87098473, 0.85495283, 0.89663462,\n",
       "        0.82370593, 0.87023744, 0.87018984, 0.87364208, 0.83967561],\n",
       "       [0.85454545, 0.86157381, 0.87247609, 0.85765125, 0.89263421,\n",
       "        0.82809071, 0.88403361, 0.86446886, 0.85913043, 0.86112796],\n",
       "       [0.85447107, 0.84540787, 0.86645299, 0.85749235, 0.90072639,\n",
       "        0.81030445, 0.86927744, 0.86448848, 0.87421742, 0.85939394],\n",
       "       [0.8507109 , 0.82465589, 0.88186528, 0.87571922, 0.86666667,\n",
       "        0.83898917, 0.84985163, 0.86018075, 0.8377724 , 0.86946521],\n",
       "       [0.87197232, 0.80943513, 0.84496124, 0.8495356 , 0.87844612,\n",
       "        0.83658171, 0.88565022, 0.871903  , 0.88790233, 0.83987539],\n",
       "       [0.8587963 , 0.85398487, 0.88694758, 0.82412698, 0.88261142,\n",
       "        0.81537302, 0.86247086, 0.8488746 , 0.85647335, 0.87583283],\n",
       "       [0.86238532, 0.86023959, 0.87656904, 0.84773151, 0.9010989 ,\n",
       "        0.84776536, 0.86783626, 0.87901991, 0.84223155, 0.83640993],\n",
       "       [0.85959885, 0.86135371, 0.89113924, 0.85108959, 0.89676617,\n",
       "        0.82878788, 0.86464411, 0.87147335, 0.88248848, 0.85890327]])"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sizes = range(1000, 14000, 1000)\n",
    "train_size, train_score, val_score = learning_curve(Pegasos(), X, currentLabel, train_sizes=sizes, cv=10,\n",
    "                                                   n_jobs=-1, verbose=10)\n",
    "val_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(6, 5))\n",
    "plt.title('Pegasos vs. training size (Label 33)')\n",
    "plt.plot(train_size, np.mean(val_score, 1), label='Validation')\n",
    "plt.plot(train_size, np.mean(train_score, 1), label='Training')\n",
    "plt.legend()\n",
    "plt.xlabel('Training size')\n",
    "plt.ylabel('Score')\n",
    "\n",
    "plt.savefig('pegasos_learning_curve_y1.png')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now want to check what are the best values for `lmbd` and `T`. Therefore, we use `GridSearchCV` to estimate them. Again, we run a 10-fold cross validation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 42 candidates, totalling 420 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   2 tasks      | elapsed:    0.3s\n",
      "[Parallel(n_jobs=-1)]: Done   9 tasks      | elapsed:    0.5s\n",
      "[Parallel(n_jobs=-1)]: Done  16 tasks      | elapsed:    0.6s\n",
      "[Parallel(n_jobs=-1)]: Done  25 tasks      | elapsed:    1.1s\n",
      "[Parallel(n_jobs=-1)]: Done  34 tasks      | elapsed:    1.4s\n",
      "[Parallel(n_jobs=-1)]: Done  45 tasks      | elapsed:    1.8s\n",
      "[Parallel(n_jobs=-1)]: Done  56 tasks      | elapsed:    2.1s\n",
      "[Parallel(n_jobs=-1)]: Done  69 tasks      | elapsed:    2.7s\n",
      "[Parallel(n_jobs=-1)]: Done  82 tasks      | elapsed:    3.3s\n",
      "[Parallel(n_jobs=-1)]: Done  97 tasks      | elapsed:    3.9s\n",
      "[Parallel(n_jobs=-1)]: Done 112 tasks      | elapsed:    4.6s\n",
      "[Parallel(n_jobs=-1)]: Done 129 tasks      | elapsed:    6.4s\n",
      "[Parallel(n_jobs=-1)]: Done 146 tasks      | elapsed:    9.6s\n",
      "[Parallel(n_jobs=-1)]: Done 165 tasks      | elapsed:   13.3s\n",
      "[Parallel(n_jobs=-1)]: Done 184 tasks      | elapsed:   17.5s\n",
      "[Parallel(n_jobs=-1)]: Done 205 tasks      | elapsed:   23.4s\n",
      "[Parallel(n_jobs=-1)]: Done 226 tasks      | elapsed:   31.3s\n",
      "[Parallel(n_jobs=-1)]: Done 249 tasks      | elapsed:   43.4s\n",
      "[Parallel(n_jobs=-1)]: Done 272 tasks      | elapsed:  1.0min\n",
      "[Parallel(n_jobs=-1)]: Done 297 tasks      | elapsed:  1.4min\n",
      "[Parallel(n_jobs=-1)]: Done 322 tasks      | elapsed:  2.0min\n",
      "[Parallel(n_jobs=-1)]: Done 349 tasks      | elapsed:  2.9min\n",
      "[Parallel(n_jobs=-1)]: Done 376 tasks      | elapsed:  3.8min\n",
      "[Parallel(n_jobs=-1)]: Done 405 tasks      | elapsed:  5.6min\n",
      "[Parallel(n_jobs=-1)]: Done 420 out of 420 | elapsed:  6.4min finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[{'T': 19860, 'lmbd': 0.0001}, 0.9119869516392145]"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "params = []\n",
    "param_grid = [{'T':[10,100,1000,2000,5000,10000,X.shape[0]],\n",
    "               'lmbd': [0.00001, 0.0001, 0.001, 0.01, 0.1, 1]}]\n",
    "\n",
    "svc = Pegasos()\n",
    "clf = GridSearchCV(svc, param_grid, cv = 10, verbose=10, n_jobs=-1)\n",
    "\n",
    "clf.fit(X, currentLabel)\n",
    "\n",
    "params.append([clf.best_params_, clf.best_score_])\n",
    "params[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Other labels: 70, 101, 4"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now repeat the same workflow for the other labels.\n",
    "However, here we also upsample and undersample data in order to check if the algorithm prediction improves."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "LABEL  70\n",
      "==============\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>Metric</th>\n",
       "      <th>Not resampled</th>\n",
       "      <th>Oversampled</th>\n",
       "      <th>Undersampled</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>True positive</td>\n",
       "      <td>1406</td>\n",
       "      <td>1733</td>\n",
       "      <td>1727</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>False positive</td>\n",
       "      <td>55</td>\n",
       "      <td>342</td>\n",
       "      <td>360</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>True negative</td>\n",
       "      <td>4640</td>\n",
       "      <td>4353</td>\n",
       "      <td>4335</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>False negative</td>\n",
       "      <td>453</td>\n",
       "      <td>126</td>\n",
       "      <td>132</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>FP ratio</td>\n",
       "      <td>0.03764544832306639 (3.76%)</td>\n",
       "      <td>0.16481927710843375 (16.48%)</td>\n",
       "      <td>0.17249640632486823 (17.25%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>FN ratio</td>\n",
       "      <td>0.08894561162379737 (8.89%)</td>\n",
       "      <td>0.02813127930341594 (2.81%)</td>\n",
       "      <td>0.029550033579583614 (2.96%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>Training score</td>\n",
       "      <td>0.861642625846335 (86.16%)</td>\n",
       "      <td>0.938092991560518 (93.81%)</td>\n",
       "      <td>0.9330859944348748 (93.31%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>Test score</td>\n",
       "      <td>0.846987951807229 (84.7%)</td>\n",
       "      <td>0.8810371123538384 (88.1%)</td>\n",
       "      <td>0.8753167764825139 (87.53%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>Precision</td>\n",
       "      <td>0.9623545516769336 (96.24%)</td>\n",
       "      <td>0.8351807228915663 (83.52%)</td>\n",
       "      <td>0.8275035936751318 (82.75%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>Recall</td>\n",
       "      <td>0.7563206024744487 (75.63%)</td>\n",
       "      <td>0.9322216245293168 (75.63%)</td>\n",
       "      <td>0.9289940828402367 (75.63%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>Accuracy</td>\n",
       "      <td>0.9224900823924321 (92.25%)</td>\n",
       "      <td>0.9285932255111382 (92.86%)</td>\n",
       "      <td>0.9249313396399146 (92.49%)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            Metric                Not resampled                   Oversampled  \\\n",
       "0    True positive                         1406                          1733   \n",
       "1   False positive                           55                           342   \n",
       "2    True negative                         4640                          4353   \n",
       "3   False negative                          453                           126   \n",
       "4         FP ratio  0.03764544832306639 (3.76%)  0.16481927710843375 (16.48%)   \n",
       "5         FN ratio  0.08894561162379737 (8.89%)   0.02813127930341594 (2.81%)   \n",
       "6   Training score   0.861642625846335 (86.16%)    0.938092991560518 (93.81%)   \n",
       "7       Test score    0.846987951807229 (84.7%)    0.8810371123538384 (88.1%)   \n",
       "8        Precision  0.9623545516769336 (96.24%)   0.8351807228915663 (83.52%)   \n",
       "9           Recall  0.7563206024744487 (75.63%)   0.9322216245293168 (75.63%)   \n",
       "10        Accuracy  0.9224900823924321 (92.25%)   0.9285932255111382 (92.86%)   \n",
       "\n",
       "                    Undersampled  \n",
       "0                           1727  \n",
       "1                            360  \n",
       "2                           4335  \n",
       "3                            132  \n",
       "4   0.17249640632486823 (17.25%)  \n",
       "5   0.029550033579583614 (2.96%)  \n",
       "6    0.9330859944348748 (93.31%)  \n",
       "7    0.8753167764825139 (87.53%)  \n",
       "8    0.8275035936751318 (82.75%)  \n",
       "9    0.9289940828402367 (75.63%)  \n",
       "10   0.9249313396399146 (92.49%)  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "LABEL  101\n",
      "==============\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>Metric</th>\n",
       "      <th>Not resampled</th>\n",
       "      <th>Oversampled</th>\n",
       "      <th>Undersampled</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>True positive</td>\n",
       "      <td>1116</td>\n",
       "      <td>1496</td>\n",
       "      <td>1511</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>False positive</td>\n",
       "      <td>86</td>\n",
       "      <td>318</td>\n",
       "      <td>371</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>True negative</td>\n",
       "      <td>4860</td>\n",
       "      <td>4628</td>\n",
       "      <td>4575</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>False negative</td>\n",
       "      <td>492</td>\n",
       "      <td>112</td>\n",
       "      <td>97</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>FP ratio</td>\n",
       "      <td>0.07154742096505824 (7.15%)</td>\n",
       "      <td>0.175303197353914 (17.53%)</td>\n",
       "      <td>0.1971307120085016 (19.71%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>FN ratio</td>\n",
       "      <td>0.09192825112107623 (9.19%)</td>\n",
       "      <td>0.02362869198312236 (2.36%)</td>\n",
       "      <td>0.020761986301369863 (2.08%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>Training score</td>\n",
       "      <td>0.808414464534075 (80.84%)</td>\n",
       "      <td>0.942035464163314 (94.2%)</td>\n",
       "      <td>0.9410334346504559 (94.1%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>Test score</td>\n",
       "      <td>0.794306049822064 (79.43%)</td>\n",
       "      <td>0.8743424897720632 (87.43%)</td>\n",
       "      <td>0.8659025787965616 (86.59%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>Precision</td>\n",
       "      <td>0.9284525790349417 (92.85%)</td>\n",
       "      <td>0.824696802646086 (82.47%)</td>\n",
       "      <td>0.8028692879914984 (80.29%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>Recall</td>\n",
       "      <td>0.6940298507462687 (69.4%)</td>\n",
       "      <td>0.9303482587064676 (69.4%)</td>\n",
       "      <td>0.9396766169154229 (69.4%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>Accuracy</td>\n",
       "      <td>0.9118095819346964 (91.18%)</td>\n",
       "      <td>0.934391211473909 (93.44%)</td>\n",
       "      <td>0.9285932255111382 (92.86%)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            Metric                Not resampled                  Oversampled  \\\n",
       "0    True positive                         1116                         1496   \n",
       "1   False positive                           86                          318   \n",
       "2    True negative                         4860                         4628   \n",
       "3   False negative                          492                          112   \n",
       "4         FP ratio  0.07154742096505824 (7.15%)   0.175303197353914 (17.53%)   \n",
       "5         FN ratio  0.09192825112107623 (9.19%)  0.02362869198312236 (2.36%)   \n",
       "6   Training score   0.808414464534075 (80.84%)    0.942035464163314 (94.2%)   \n",
       "7       Test score   0.794306049822064 (79.43%)  0.8743424897720632 (87.43%)   \n",
       "8        Precision  0.9284525790349417 (92.85%)   0.824696802646086 (82.47%)   \n",
       "9           Recall   0.6940298507462687 (69.4%)   0.9303482587064676 (69.4%)   \n",
       "10        Accuracy  0.9118095819346964 (91.18%)   0.934391211473909 (93.44%)   \n",
       "\n",
       "                    Undersampled  \n",
       "0                           1511  \n",
       "1                            371  \n",
       "2                           4575  \n",
       "3                             97  \n",
       "4    0.1971307120085016 (19.71%)  \n",
       "5   0.020761986301369863 (2.08%)  \n",
       "6     0.9410334346504559 (94.1%)  \n",
       "7    0.8659025787965616 (86.59%)  \n",
       "8    0.8028692879914984 (80.29%)  \n",
       "9     0.9396766169154229 (69.4%)  \n",
       "10   0.9285932255111382 (92.86%)  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "LABEL  4\n",
      "==============\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>Metric</th>\n",
       "      <th>Not resampled</th>\n",
       "      <th>Oversampled</th>\n",
       "      <th>Undersampled</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>True positive</td>\n",
       "      <td>821</td>\n",
       "      <td>1018</td>\n",
       "      <td>1041</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>False positive</td>\n",
       "      <td>66</td>\n",
       "      <td>186</td>\n",
       "      <td>313</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>True negative</td>\n",
       "      <td>5345</td>\n",
       "      <td>5225</td>\n",
       "      <td>5098</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>False negative</td>\n",
       "      <td>322</td>\n",
       "      <td>125</td>\n",
       "      <td>102</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>FP ratio</td>\n",
       "      <td>0.07440811724915446 (7.44%)</td>\n",
       "      <td>0.15448504983388706 (15.45%)</td>\n",
       "      <td>0.2311669128508124 (23.12%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>FN ratio</td>\n",
       "      <td>0.056820187047820714 (5.68%)</td>\n",
       "      <td>0.02336448598130841 (2.34%)</td>\n",
       "      <td>0.019615384615384614 (1.96%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>Training score</td>\n",
       "      <td>0.8309178743961352 (83.09%)</td>\n",
       "      <td>0.9359837957510952 (93.6%)</td>\n",
       "      <td>0.9328358208955225 (93.28%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>Test score</td>\n",
       "      <td>0.8088669950738916 (80.89%)</td>\n",
       "      <td>0.8674904132935664 (86.75%)</td>\n",
       "      <td>0.8338005606728074 (83.38%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>Precision</td>\n",
       "      <td>0.9255918827508456 (92.56%)</td>\n",
       "      <td>0.845514950166113 (84.55%)</td>\n",
       "      <td>0.7688330871491876 (76.88%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>Recall</td>\n",
       "      <td>0.7182852143482065 (71.83%)</td>\n",
       "      <td>0.8906386701662292 (71.83%)</td>\n",
       "      <td>0.910761154855643 (71.83%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>Accuracy</td>\n",
       "      <td>0.9407995117485505 (94.08%)</td>\n",
       "      <td>0.9525480622520598 (95.25%)</td>\n",
       "      <td>0.9366798901434239 (93.67%)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            Metric                 Not resampled  \\\n",
       "0    True positive                           821   \n",
       "1   False positive                            66   \n",
       "2    True negative                          5345   \n",
       "3   False negative                           322   \n",
       "4         FP ratio   0.07440811724915446 (7.44%)   \n",
       "5         FN ratio  0.056820187047820714 (5.68%)   \n",
       "6   Training score   0.8309178743961352 (83.09%)   \n",
       "7       Test score   0.8088669950738916 (80.89%)   \n",
       "8        Precision   0.9255918827508456 (92.56%)   \n",
       "9           Recall   0.7182852143482065 (71.83%)   \n",
       "10        Accuracy   0.9407995117485505 (94.08%)   \n",
       "\n",
       "                     Oversampled                  Undersampled  \n",
       "0                           1018                          1041  \n",
       "1                            186                           313  \n",
       "2                           5225                          5098  \n",
       "3                            125                           102  \n",
       "4   0.15448504983388706 (15.45%)   0.2311669128508124 (23.12%)  \n",
       "5    0.02336448598130841 (2.34%)  0.019615384615384614 (1.96%)  \n",
       "6     0.9359837957510952 (93.6%)   0.9328358208955225 (93.28%)  \n",
       "7    0.8674904132935664 (86.75%)   0.8338005606728074 (83.38%)  \n",
       "8     0.845514950166113 (84.55%)   0.7688330871491876 (76.88%)  \n",
       "9    0.8906386701662292 (71.83%)    0.910761154855643 (71.83%)  \n",
       "10   0.9525480622520598 (95.25%)   0.9366798901434239 (93.67%)  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "lbl_idx = { '2': '70', '3': '101', '4': '4' }\n",
    "TPs_resampled, FPs_resampled, TNs_resampled, FNs_resampled, precisions_resampled, recalls_resampled, accs_resampled = [], [], [], [], [], [], []\n",
    "TPs_unders, FPs_unders, TNs_unders, FNs_unders, precisions_unders, recalls_unders, accs_unders = [], [], [], [], [], [], []\n",
    "\n",
    "for i in range(2,5):\n",
    "    currentLabel = eval('y'+str(i))\n",
    "    \n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, currentLabel, test_size=0.33,\n",
    "                                                        random_state=42, stratify=currentLabel)\n",
    "    \n",
    "    classifier = Pegasos()\n",
    "    classifier.fit(X_train, y_train)\n",
    "    train_score = classifier.score(X_train, y_train)\n",
    "    \n",
    "    y_pred = classifier.predict(X_test)  \n",
    "    f1 = f1_score(y_test, y_pred)\n",
    "    tn, fp, fn, tp = confusion_matrix(y_test, y_pred).ravel()\n",
    "    precision = precision_score(y_test, y_pred)\n",
    "    recall = recall_score(y_test, y_pred)\n",
    "    acc = accuracy_score(y_test, y_pred)\n",
    "    \n",
    "    TNs.append(tn)\n",
    "    TPs.append(tp)\n",
    "    FPs.append(fp)\n",
    "    FNs.append(fn)\n",
    "    precisions.append(precision)\n",
    "    recalls.append(recall)\n",
    "    accs.append(acc)\n",
    "    \n",
    "    # Up-sample\n",
    "    ros = RandomOverSampler(random_state=42)\n",
    "    X_train_resampled, y_train_resampled = ros.fit_resample(X_train, y_train)\n",
    "    \n",
    "    classifier = Pegasos()\n",
    "    classifier.fit(X_train_resampled, y_train_resampled)\n",
    "    train_score_resampled = classifier.score(X_train_resampled, y_train_resampled)\n",
    "    \n",
    "    y_pred = classifier.predict(X_test)  \n",
    "    f1_resampled = f1_score(y_test, y_pred)\n",
    "    tn_resampled, fp_resampled, fn_resampled, tp_resampled = confusion_matrix(y_test, y_pred).ravel()\n",
    "    precision_resampled = precision_score(y_test, y_pred)\n",
    "    recall_resampled = recall_score(y_test, y_pred)\n",
    "    acc_resampled = accuracy_score(y_test, y_pred)\n",
    "    \n",
    "    TNs_resampled.append(tn_resampled)\n",
    "    TPs_resampled.append(tp_resampled)\n",
    "    FPs_resampled.append(fp_resampled)\n",
    "    FNs_resampled.append(fn_resampled)\n",
    "    precisions_resampled.append(precision_resampled)\n",
    "    recalls_resampled.append(recall_resampled)\n",
    "    accs_resampled.append(acc_resampled)\n",
    "    \n",
    "    # Under-sample\n",
    "    rus = RandomUnderSampler(random_state=42, replacement=True)\n",
    "    X_train_unders, y_train_unders = rus.fit_resample(X_train, y_train)\n",
    "    \n",
    "    classifier = Pegasos()\n",
    "    classifier.fit(X_train_unders, y_train_unders)\n",
    "    train_score_unders = classifier.score(X_train_unders, y_train_unders)\n",
    "    \n",
    "    y_pred = classifier.predict(X_test)  \n",
    "    f1_unders = f1_score(y_test, y_pred)\n",
    "    tn_unders, fp_unders, fn_unders, tp_unders = confusion_matrix(y_test, y_pred).ravel()\n",
    "    precision_unders = precision_score(y_test, y_pred)\n",
    "    recall_unders = recall_score(y_test, y_pred)\n",
    "    acc_unders = accuracy_score(y_test, y_pred)\n",
    "    \n",
    "    TNs_unders.append(tn_unders)\n",
    "    TPs_unders.append(tp_unders)\n",
    "    FPs_unders.append(fp_unders)\n",
    "    FNs_unders.append(fn_unders)\n",
    "    precisions_unders.append(precision_unders)\n",
    "    recalls_unders.append(recall_unders)\n",
    "    accs_unders.append(acc_unders)\n",
    "    \n",
    "    print(\"\\nLABEL \", lbl_idx.get(str(i)))\n",
    "    print(\"==============\")\n",
    "    \n",
    "    metrics_name = [\"True positive\", \"False positive\", \"True negative\", \"False negative\", \"FP ratio\", \"FN ratio\",\n",
    "                    \"Training score\", \"Test score\", \"Precision\", \"Recall\", \"Accuracy\"]\n",
    "    nr = [tp,fp,tn,fn,\n",
    "          str(fp/(tp+fp))+\" (\"+str(np.round(fp/(tp+fp)*100, 2))+\"%)\",\n",
    "          str(fn/(tn+fn))+\" (\"+str(np.round(fn/(tn+fn)*100, 2))+\"%)\",\n",
    "          str(train_score)+\" (\"+str(np.round(train_score*100, 2))+\"%)\",\n",
    "          str(f1)+\" (\"+str(np.round(f1*100, 2))+\"%)\",\n",
    "          str(precision)+\" (\"+str(np.round(precision*100, 2))+\"%)\",\n",
    "          str(recall)+\" (\"+str(np.round(recall*100, 2))+\"%)\",\n",
    "          str(acc)+\" (\"+str(np.round(acc*100, 2))+\"%)\"]\n",
    "    \n",
    "    ovs = [tp_resampled,fp_resampled, tn_resampled,fn_resampled,\n",
    "          str(fp_resampled/(tp_resampled+fp_resampled))+\" (\"+str(np.round(fp_resampled/(tp_resampled+fp_resampled)*100, 2))+\"%)\",\n",
    "          str(fn_resampled/(tn_resampled+fn_resampled))+\" (\"+str(np.round(fn_resampled/(tn_resampled+fn_resampled)*100, 2))+\"%)\",\n",
    "          str(train_score_resampled)+\" (\"+str(np.round(train_score_resampled*100, 2))+\"%)\",\n",
    "          str(f1_resampled)+\" (\"+str(np.round(f1_resampled*100, 2))+\"%)\",\n",
    "          str(precision_resampled)+\" (\"+str(np.round(precision_resampled*100, 2))+\"%)\",\n",
    "          str(recall_resampled)+\" (\"+str(np.round(recall*100, 2))+\"%)\",\n",
    "          str(acc_resampled)+\" (\"+str(np.round(acc_resampled*100, 2))+\"%)\"]\n",
    "    \n",
    "    uds = [tp_unders,fp_unders, tn_unders,fn_unders,\n",
    "          str(fp_unders/(tp_unders+fp_unders))+\" (\"+str(np.round(fp_unders/(tp_unders+fp_unders)*100, 2))+\"%)\",\n",
    "          str(fn_unders/(tn_unders+fn_unders))+\" (\"+str(np.round(fn_unders/(tn_unders+fn_unders)*100, 2))+\"%)\",\n",
    "          str(train_score_unders)+\" (\"+str(np.round(train_score_unders*100, 2))+\"%)\",\n",
    "          str(f1_unders)+\" (\"+str(np.round(f1_unders*100, 2))+\"%)\",\n",
    "          str(precision_unders)+\" (\"+str(np.round(precision_unders*100, 2))+\"%)\",\n",
    "          str(recall_unders)+\" (\"+str(np.round(recall*100, 2))+\"%)\",\n",
    "          str(acc_unders)+\" (\"+str(np.round(acc_unders*100, 2))+\"%)\"]\n",
    "    \n",
    "    df1 = pd.DataFrame(np.column_stack([metrics_name, nr, ovs, uds]), columns=[\"Metric\", \"Not resampled\", \"Oversampled\", \"Undersampled\"])\n",
    "    display(df1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[learning_curve] Training set sizes: [ 1000  2000  3000  4000  5000  6000  7000  8000  9000 10000 11000 12000\n",
      " 13000]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   5 out of 130 | elapsed:    1.3s remaining:   33.2s\n",
      "[Parallel(n_jobs=-1)]: Done  19 out of 130 | elapsed:    3.8s remaining:   22.0s\n",
      "[Parallel(n_jobs=-1)]: Done  33 out of 130 | elapsed:    6.1s remaining:   17.9s\n",
      "[Parallel(n_jobs=-1)]: Done  47 out of 130 | elapsed:    8.1s remaining:   14.3s\n",
      "[Parallel(n_jobs=-1)]: Done  61 out of 130 | elapsed:   10.4s remaining:   11.7s\n",
      "[Parallel(n_jobs=-1)]: Done  75 out of 130 | elapsed:   12.7s remaining:    9.3s\n",
      "[Parallel(n_jobs=-1)]: Done  89 out of 130 | elapsed:   14.5s remaining:    6.7s\n",
      "[Parallel(n_jobs=-1)]: Done 103 out of 130 | elapsed:   16.8s remaining:    4.4s\n",
      "[Parallel(n_jobs=-1)]: Done 117 out of 130 | elapsed:   18.7s remaining:    2.1s\n",
      "[Parallel(n_jobs=-1)]: Done 130 out of 130 | elapsed:   20.4s finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[learning_curve] Training set sizes: [ 1000  2000  3000  4000  5000  6000  7000  8000  9000 10000 11000 12000\n",
      " 13000]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Done   5 out of 130 | elapsed:    1.3s remaining:   33.1s\n",
      "[Parallel(n_jobs=-1)]: Done  19 out of 130 | elapsed:    3.6s remaining:   21.1s\n",
      "[Parallel(n_jobs=-1)]: Done  33 out of 130 | elapsed:    5.8s remaining:   17.1s\n",
      "[Parallel(n_jobs=-1)]: Done  47 out of 130 | elapsed:    7.4s remaining:   13.1s\n",
      "[Parallel(n_jobs=-1)]: Done  61 out of 130 | elapsed:    9.6s remaining:   10.9s\n",
      "[Parallel(n_jobs=-1)]: Done  75 out of 130 | elapsed:   11.7s remaining:    8.6s\n",
      "[Parallel(n_jobs=-1)]: Done  89 out of 130 | elapsed:   13.7s remaining:    6.3s\n",
      "[Parallel(n_jobs=-1)]: Done 103 out of 130 | elapsed:   15.7s remaining:    4.1s\n",
      "[Parallel(n_jobs=-1)]: Done 117 out of 130 | elapsed:   17.9s remaining:    2.0s\n",
      "[Parallel(n_jobs=-1)]: Done 130 out of 130 | elapsed:   19.3s finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[learning_curve] Training set sizes: [ 1000  2000  3000  4000  5000  6000  7000  8000  9000 10000 11000 12000\n",
      " 13000]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Done   5 out of 130 | elapsed:    1.2s remaining:   30.1s\n",
      "[Parallel(n_jobs=-1)]: Done  19 out of 130 | elapsed:    3.5s remaining:   20.3s\n",
      "[Parallel(n_jobs=-1)]: Done  33 out of 130 | elapsed:    5.4s remaining:   15.9s\n",
      "[Parallel(n_jobs=-1)]: Done  47 out of 130 | elapsed:    7.0s remaining:   12.4s\n",
      "[Parallel(n_jobs=-1)]: Done  61 out of 130 | elapsed:    9.2s remaining:   10.4s\n",
      "[Parallel(n_jobs=-1)]: Done  75 out of 130 | elapsed:   11.1s remaining:    8.1s\n",
      "[Parallel(n_jobs=-1)]: Done  89 out of 130 | elapsed:   12.9s remaining:    5.9s\n",
      "[Parallel(n_jobs=-1)]: Done 103 out of 130 | elapsed:   14.8s remaining:    3.9s\n",
      "[Parallel(n_jobs=-1)]: Done 117 out of 130 | elapsed:   16.7s remaining:    1.9s\n",
      "[Parallel(n_jobs=-1)]: Done 130 out of 130 | elapsed:   18.0s finished\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(20, 5))\n",
    "\n",
    "for i in range(2,5):\n",
    "    currentLabel = eval('y'+str(i))\n",
    "\n",
    "    sizes = range(1000, 14000, 1000)\n",
    "    train_size, train_score, val_score = learning_curve(Pegasos(), X, currentLabel, train_sizes=sizes, cv=10,\n",
    "                                                   n_jobs=-1, verbose=10)\n",
    "    \n",
    "    plt.subplot(1, 3, i-1)\n",
    "    plt.title('Pegasos vs. training size (Label ' + lbl_idx.get(str(i)) + ')')\n",
    "    plt.plot(train_size, np.mean(val_score, 1), label='Validation')\n",
    "    plt.plot(train_size, np.mean(train_score, 1), label='Training')\n",
    "    plt.legend()\n",
    "    plt.xlabel('Training size')\n",
    "    plt.ylabel('Score')\n",
    "\n",
    "plt.savefig('pegasos_learning_curve_other.png')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We run `GridSearchCV` to get the best parameters of Pegasos for every label."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 42 candidates, totalling 420 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   2 tasks      | elapsed:    0.4s\n",
      "[Parallel(n_jobs=-1)]: Done   9 tasks      | elapsed:    0.7s\n",
      "[Parallel(n_jobs=-1)]: Done  16 tasks      | elapsed:    0.8s\n",
      "[Parallel(n_jobs=-1)]: Done  25 tasks      | elapsed:    1.6s\n",
      "[Parallel(n_jobs=-1)]: Done  34 tasks      | elapsed:    2.0s\n",
      "[Parallel(n_jobs=-1)]: Done  45 tasks      | elapsed:    2.5s\n",
      "[Parallel(n_jobs=-1)]: Done  56 tasks      | elapsed:    3.0s\n",
      "[Parallel(n_jobs=-1)]: Done  69 tasks      | elapsed:    3.8s\n",
      "[Parallel(n_jobs=-1)]: Done  82 tasks      | elapsed:    4.6s\n",
      "[Parallel(n_jobs=-1)]: Done  97 tasks      | elapsed:    5.5s\n",
      "[Parallel(n_jobs=-1)]: Done 112 tasks      | elapsed:    6.3s\n",
      "[Parallel(n_jobs=-1)]: Done 129 tasks      | elapsed:    8.3s\n",
      "[Parallel(n_jobs=-1)]: Done 146 tasks      | elapsed:   10.9s\n",
      "[Parallel(n_jobs=-1)]: Done 165 tasks      | elapsed:   15.0s\n",
      "[Parallel(n_jobs=-1)]: Done 184 tasks      | elapsed:   19.1s\n",
      "[Parallel(n_jobs=-1)]: Done 205 tasks      | elapsed:   24.2s\n",
      "[Parallel(n_jobs=-1)]: Done 226 tasks      | elapsed:   32.3s\n",
      "[Parallel(n_jobs=-1)]: Done 249 tasks      | elapsed:   44.0s\n",
      "[Parallel(n_jobs=-1)]: Done 272 tasks      | elapsed:  1.0min\n",
      "[Parallel(n_jobs=-1)]: Done 297 tasks      | elapsed:  1.4min\n",
      "[Parallel(n_jobs=-1)]: Done 322 tasks      | elapsed:  1.9min\n",
      "[Parallel(n_jobs=-1)]: Done 349 tasks      | elapsed:  2.7min\n",
      "[Parallel(n_jobs=-1)]: Done 376 tasks      | elapsed:  3.7min\n",
      "[Parallel(n_jobs=-1)]: Done 405 tasks      | elapsed:  5.4min\n",
      "[Parallel(n_jobs=-1)]: Done 420 out of 420 | elapsed:  6.2min finished\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 42 candidates, totalling 420 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   2 tasks      | elapsed:    0.3s\n",
      "[Parallel(n_jobs=-1)]: Done   9 tasks      | elapsed:    0.7s\n",
      "[Parallel(n_jobs=-1)]: Done  16 tasks      | elapsed:    0.9s\n",
      "[Parallel(n_jobs=-1)]: Done  25 tasks      | elapsed:    1.5s\n",
      "[Parallel(n_jobs=-1)]: Done  34 tasks      | elapsed:    1.9s\n",
      "[Parallel(n_jobs=-1)]: Done  45 tasks      | elapsed:    2.5s\n",
      "[Parallel(n_jobs=-1)]: Done  56 tasks      | elapsed:    3.0s\n",
      "[Parallel(n_jobs=-1)]: Done  69 tasks      | elapsed:    3.7s\n",
      "[Parallel(n_jobs=-1)]: Done  82 tasks      | elapsed:    4.5s\n",
      "[Parallel(n_jobs=-1)]: Done  97 tasks      | elapsed:    5.4s\n",
      "[Parallel(n_jobs=-1)]: Done 112 tasks      | elapsed:    6.2s\n",
      "[Parallel(n_jobs=-1)]: Done 129 tasks      | elapsed:    8.6s\n",
      "[Parallel(n_jobs=-1)]: Done 146 tasks      | elapsed:   11.2s\n",
      "[Parallel(n_jobs=-1)]: Done 165 tasks      | elapsed:   14.9s\n",
      "[Parallel(n_jobs=-1)]: Done 184 tasks      | elapsed:   18.6s\n",
      "[Parallel(n_jobs=-1)]: Done 205 tasks      | elapsed:   24.6s\n",
      "[Parallel(n_jobs=-1)]: Done 226 tasks      | elapsed:   32.4s\n",
      "[Parallel(n_jobs=-1)]: Done 249 tasks      | elapsed:   43.5s\n",
      "[Parallel(n_jobs=-1)]: Done 272 tasks      | elapsed:  1.1min\n",
      "[Parallel(n_jobs=-1)]: Done 297 tasks      | elapsed:  1.5min\n",
      "[Parallel(n_jobs=-1)]: Done 322 tasks      | elapsed:  2.0min\n",
      "[Parallel(n_jobs=-1)]: Done 349 tasks      | elapsed:  2.9min\n",
      "[Parallel(n_jobs=-1)]: Done 376 tasks      | elapsed:  3.8min\n",
      "[Parallel(n_jobs=-1)]: Done 405 tasks      | elapsed:  5.6min\n",
      "[Parallel(n_jobs=-1)]: Done 420 out of 420 | elapsed:  6.3min finished\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 42 candidates, totalling 420 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   2 tasks      | elapsed:    0.3s\n",
      "[Parallel(n_jobs=-1)]: Done   9 tasks      | elapsed:    0.7s\n",
      "[Parallel(n_jobs=-1)]: Done  16 tasks      | elapsed:    0.9s\n",
      "[Parallel(n_jobs=-1)]: Done  25 tasks      | elapsed:    1.5s\n",
      "[Parallel(n_jobs=-1)]: Done  34 tasks      | elapsed:    1.9s\n",
      "[Parallel(n_jobs=-1)]: Done  45 tasks      | elapsed:    2.4s\n",
      "[Parallel(n_jobs=-1)]: Done  56 tasks      | elapsed:    2.9s\n",
      "[Parallel(n_jobs=-1)]: Done  69 tasks      | elapsed:    3.6s\n",
      "[Parallel(n_jobs=-1)]: Done  82 tasks      | elapsed:    4.4s\n",
      "[Parallel(n_jobs=-1)]: Done  97 tasks      | elapsed:    5.3s\n",
      "[Parallel(n_jobs=-1)]: Done 112 tasks      | elapsed:    6.1s\n",
      "[Parallel(n_jobs=-1)]: Done 129 tasks      | elapsed:    8.2s\n",
      "[Parallel(n_jobs=-1)]: Done 146 tasks      | elapsed:   10.6s\n",
      "[Parallel(n_jobs=-1)]: Done 165 tasks      | elapsed:   14.4s\n",
      "[Parallel(n_jobs=-1)]: Done 184 tasks      | elapsed:   18.5s\n",
      "[Parallel(n_jobs=-1)]: Done 205 tasks      | elapsed:   23.9s\n",
      "[Parallel(n_jobs=-1)]: Done 226 tasks      | elapsed:   31.3s\n",
      "[Parallel(n_jobs=-1)]: Done 249 tasks      | elapsed:   42.7s\n",
      "[Parallel(n_jobs=-1)]: Done 272 tasks      | elapsed:  1.0min\n",
      "[Parallel(n_jobs=-1)]: Done 297 tasks      | elapsed:  1.4min\n",
      "[Parallel(n_jobs=-1)]: Done 322 tasks      | elapsed:  1.9min\n",
      "[Parallel(n_jobs=-1)]: Done 349 tasks      | elapsed:  2.7min\n",
      "[Parallel(n_jobs=-1)]: Done 376 tasks      | elapsed:  3.7min\n",
      "[Parallel(n_jobs=-1)]: Done 405 tasks      | elapsed:  5.4min\n",
      "[Parallel(n_jobs=-1)]: Done 420 out of 420 | elapsed:  6.1min finished\n"
     ]
    }
   ],
   "source": [
    "for i in range(2,5):\n",
    "    currentLabel = eval('y'+str(i))\n",
    "    \n",
    "    param_grid = [{'T':[10,100,1000,2000,5000,10000,X.shape[0]],\n",
    "               'lmbd': [0.00001, 0.0001, 0.001, 0.01, 0.1, 1]}]\n",
    "\n",
    "    svc = Pegasos()\n",
    "    clf = GridSearchCV(svc, param_grid, cv = 10, verbose=10, n_jobs=-1)\n",
    "\n",
    "    clf.fit(X, currentLabel)\n",
    "\n",
    "    params.append([clf.best_params_, clf.best_score_])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Results: Pegasos"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now analyze the results."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### No resampling"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These are for the random split where the test set is a third of the dataset, with no resampling."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "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>Label</th>\n",
       "      <th>True Positive</th>\n",
       "      <th>False Positive</th>\n",
       "      <th>True Negative</th>\n",
       "      <th>False Negative</th>\n",
       "      <th>Precision</th>\n",
       "      <th>Recall</th>\n",
       "      <th>Accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>33</td>\n",
       "      <td>2331</td>\n",
       "      <td>206</td>\n",
       "      <td>3411</td>\n",
       "      <td>606</td>\n",
       "      <td>0.918802</td>\n",
       "      <td>0.793667</td>\n",
       "      <td>0.876106</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>70</td>\n",
       "      <td>1406</td>\n",
       "      <td>55</td>\n",
       "      <td>4640</td>\n",
       "      <td>453</td>\n",
       "      <td>0.962355</td>\n",
       "      <td>0.756321</td>\n",
       "      <td>0.922490</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>101</td>\n",
       "      <td>1116</td>\n",
       "      <td>86</td>\n",
       "      <td>4860</td>\n",
       "      <td>492</td>\n",
       "      <td>0.928453</td>\n",
       "      <td>0.694030</td>\n",
       "      <td>0.911810</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>821</td>\n",
       "      <td>66</td>\n",
       "      <td>5345</td>\n",
       "      <td>322</td>\n",
       "      <td>0.925592</td>\n",
       "      <td>0.718285</td>\n",
       "      <td>0.940800</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label  True Positive  False Positive  True Negative  False Negative  \\\n",
       "0     33           2331             206           3411             606   \n",
       "1     70           1406              55           4640             453   \n",
       "2    101           1116              86           4860             492   \n",
       "3      4            821              66           5345             322   \n",
       "\n",
       "   Precision    Recall  Accuracy  \n",
       "0   0.918802  0.793667  0.876106  \n",
       "1   0.962355  0.756321  0.922490  \n",
       "2   0.928453  0.694030  0.911810  \n",
       "3   0.925592  0.718285  0.940800  "
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_metrics = pd.DataFrame(np.column_stack([highest, TPs, FPs, TNs, FNs, precisions, recalls, accs]),\n",
    "                  columns=['Label',\n",
    "                           'True Positive',\n",
    "                           'False Positive',\n",
    "                           'True Negative', \n",
    "                           'False Negative',\n",
    "                           'Precision',\n",
    "                           'Recall',\n",
    "                           'Accuracy'])\n",
    "\n",
    "df_metrics = df_metrics.astype({'Label': int,\n",
    "                                'True Positive': int,\n",
    "                                'False Positive': int,\n",
    "                                'True Negative': int,\n",
    "                                'False Negative': int})\n",
    "df_metrics"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x324 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = df_metrics.plot(x='Label', \n",
    "                      y=['Precision', 'Recall', 'Accuracy'],\n",
    "                      kind=\"bar\",\n",
    "                      title='Pegasos metrics',\n",
    "                      figsize=(6, 4.5)).get_figure()\n",
    "\n",
    "fig.savefig(\"pegasos_all_metrics.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean Precision:  0.9338001869486523\n",
      "Mean Recall:  0.7405756686797693\n",
      "Mean Accuracy:  0.912801342691486\n"
     ]
    }
   ],
   "source": [
    "print(\"Mean Precision: \", df_metrics[\"Precision\"].mean())\n",
    "print(\"Mean Recall: \", df_metrics[\"Recall\"].mean())\n",
    "print(\"Mean Accuracy: \", df_metrics[\"Accuracy\"].mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Upsampling"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These are for the random split where the test set is a third of the dataset, with upsampling."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "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>Label</th>\n",
       "      <th>True Positive</th>\n",
       "      <th>False Positive</th>\n",
       "      <th>True Negative</th>\n",
       "      <th>False Negative</th>\n",
       "      <th>Precision</th>\n",
       "      <th>Recall</th>\n",
       "      <th>Accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>70</td>\n",
       "      <td>1733</td>\n",
       "      <td>342</td>\n",
       "      <td>4353</td>\n",
       "      <td>126</td>\n",
       "      <td>0.835181</td>\n",
       "      <td>0.932222</td>\n",
       "      <td>0.928593</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>101</td>\n",
       "      <td>1496</td>\n",
       "      <td>318</td>\n",
       "      <td>4628</td>\n",
       "      <td>112</td>\n",
       "      <td>0.824697</td>\n",
       "      <td>0.930348</td>\n",
       "      <td>0.934391</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4</td>\n",
       "      <td>1018</td>\n",
       "      <td>186</td>\n",
       "      <td>5225</td>\n",
       "      <td>125</td>\n",
       "      <td>0.845515</td>\n",
       "      <td>0.890639</td>\n",
       "      <td>0.952548</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label  True Positive  False Positive  True Negative  False Negative  \\\n",
       "0     70           1733             342           4353             126   \n",
       "1    101           1496             318           4628             112   \n",
       "2      4           1018             186           5225             125   \n",
       "\n",
       "   Precision    Recall  Accuracy  \n",
       "0   0.835181  0.932222  0.928593  \n",
       "1   0.824697  0.930348  0.934391  \n",
       "2   0.845515  0.890639  0.952548  "
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_metrics_resampled = pd.DataFrame(np.column_stack([highest[1:], TPs_resampled, FPs_resampled, TNs_resampled, FNs_resampled, precisions_resampled, recalls_resampled, accs_resampled]),\n",
    "                  columns=['Label',\n",
    "                           'True Positive',\n",
    "                           'False Positive',\n",
    "                           'True Negative', \n",
    "                           'False Negative',\n",
    "                           'Precision',\n",
    "                           'Recall',\n",
    "                           'Accuracy'])\n",
    "\n",
    "df_metrics_resampled = df_metrics_resampled.astype({'Label': int,\n",
    "                                'True Positive': int,\n",
    "                                'False Positive': int,\n",
    "                                'True Negative': int,\n",
    "                                'False Negative': int})\n",
    "df_metrics_resampled"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x324 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = df_metrics_resampled.plot(x='Label', \n",
    "                      y=['Precision', 'Recall', 'Accuracy'],\n",
    "                      kind=\"bar\",\n",
    "                      title='Pegasos metrics (with resampling)',\n",
    "                      figsize=(6, 4.5)).get_figure()\n",
    "\n",
    "fig.savefig(\"pegasos_all_metrics_upsampled.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean Precision:  0.8351308252345885\n",
      "Mean Recall:  0.917736184467338\n",
      "Mean Accuracy:  0.9385108330790356\n"
     ]
    }
   ],
   "source": [
    "print(\"Mean Precision: \", df_metrics_resampled[\"Precision\"].mean())\n",
    "print(\"Mean Recall: \", df_metrics_resampled[\"Recall\"].mean())\n",
    "print(\"Mean Accuracy: \", df_metrics_resampled[\"Accuracy\"].mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Undersampling"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These are for the random split where the test set is a third of the dataset, with undersampling."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "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>Label</th>\n",
       "      <th>True Positive</th>\n",
       "      <th>False Positive</th>\n",
       "      <th>True Negative</th>\n",
       "      <th>False Negative</th>\n",
       "      <th>Precision</th>\n",
       "      <th>Recall</th>\n",
       "      <th>Accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>70</td>\n",
       "      <td>1727</td>\n",
       "      <td>360</td>\n",
       "      <td>4335</td>\n",
       "      <td>132</td>\n",
       "      <td>0.827504</td>\n",
       "      <td>0.928994</td>\n",
       "      <td>0.924931</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>101</td>\n",
       "      <td>1511</td>\n",
       "      <td>371</td>\n",
       "      <td>4575</td>\n",
       "      <td>97</td>\n",
       "      <td>0.802869</td>\n",
       "      <td>0.939677</td>\n",
       "      <td>0.928593</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4</td>\n",
       "      <td>1041</td>\n",
       "      <td>313</td>\n",
       "      <td>5098</td>\n",
       "      <td>102</td>\n",
       "      <td>0.768833</td>\n",
       "      <td>0.910761</td>\n",
       "      <td>0.936680</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label  True Positive  False Positive  True Negative  False Negative  \\\n",
       "0     70           1727             360           4335             132   \n",
       "1    101           1511             371           4575              97   \n",
       "2      4           1041             313           5098             102   \n",
       "\n",
       "   Precision    Recall  Accuracy  \n",
       "0   0.827504  0.928994  0.924931  \n",
       "1   0.802869  0.939677  0.928593  \n",
       "2   0.768833  0.910761  0.936680  "
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_metrics_unders = pd.DataFrame(np.column_stack([highest[1:], TPs_unders, FPs_unders, TNs_unders, FNs_unders, precisions_unders, recalls_unders, accs_unders]),\n",
    "                  columns=['Label',\n",
    "                           'True Positive',\n",
    "                           'False Positive',\n",
    "                           'True Negative', \n",
    "                           'False Negative',\n",
    "                           'Precision',\n",
    "                           'Recall',\n",
    "                           'Accuracy'])\n",
    "\n",
    "df_metrics_unders = df_metrics_unders.astype({'Label': int,\n",
    "                                'True Positive': int,\n",
    "                                'False Positive': int,\n",
    "                                'True Negative': int,\n",
    "                                'False Negative': int})\n",
    "df_metrics_unders"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x324 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = df_metrics_unders.plot(x='Label', \n",
    "                      y=['Precision', 'Recall', 'Accuracy'],\n",
    "                      kind=\"bar\",\n",
    "                      title='Pegasos metrics (with resampling)',\n",
    "                      figsize=(6, 4.5)).get_figure()\n",
    "\n",
    "fig.savefig(\"pegasos_all_metrics_undersampled.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean Precision:  0.799735322938606\n",
      "Mean Recall:  0.9264772848704341\n",
      "Mean Accuracy:  0.9300681517648256\n"
     ]
    }
   ],
   "source": [
    "print(\"Mean Precision: \", df_metrics_unders[\"Precision\"].mean())\n",
    "print(\"Mean Recall: \", df_metrics_unders[\"Recall\"].mean())\n",
    "print(\"Mean Accuracy: \", df_metrics_unders[\"Accuracy\"].mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Best score"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These are the results of the `GridSearchCV`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "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>Label</th>\n",
       "      <th>Iterations</th>\n",
       "      <th>Lambda</th>\n",
       "      <th>F1 Test score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>33</td>\n",
       "      <td>19860</td>\n",
       "      <td>0.0001</td>\n",
       "      <td>0.911987</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>70</td>\n",
       "      <td>19860</td>\n",
       "      <td>0.0001</td>\n",
       "      <td>0.916777</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>101</td>\n",
       "      <td>19860</td>\n",
       "      <td>0.0001</td>\n",
       "      <td>0.907211</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>19860</td>\n",
       "      <td>0.0001</td>\n",
       "      <td>0.884570</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label  Iterations  Lambda  F1 Test score\n",
       "0     33       19860  0.0001       0.911987\n",
       "1     70       19860  0.0001       0.916777\n",
       "2    101       19860  0.0001       0.907211\n",
       "3      4       19860  0.0001       0.884570"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Ts, lmbds, scores = [], [], []\n",
    "\n",
    "for i in params:\n",
    "    Ts.append(i[0]['T'])\n",
    "    lmbds.append(i[0]['lmbd'])\n",
    "    scores.append(i[1])\n",
    "\n",
    "df_results = pd.DataFrame(np.column_stack([highest, Ts, lmbds, scores]),\n",
    "                          columns=['Label', 'Iterations', 'Lambda', 'F1 Test score'])\n",
    "\n",
    "df_results = df_results.astype({\"Label\": int, \"Iterations\": int})\n",
    "df_results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see that only the test score changes.\n",
    "\n",
    "Let's do some plotting"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x324 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = df_results.plot(x='Label', y=['F1 Test score'], kind=\"bar\",\n",
    "                      title='Pegasos F1 score', figsize=(6, 4.5)).get_figure()\n",
    "\n",
    "fig.savefig(\"pegasos_test_scores.png\")\n",
    "\n",
    "if len(df_results['Iterations'].unique()) != 1:\n",
    "    fig = df_results.plot(x='Label', y=['Iterations'], title='Pegasos iterations',\n",
    "                          kind=\"bar\", color='g', figsize=(6, 4.5)).get_figure()\n",
    "    \n",
    "    fig.savefig(\"pegasos_iterations.png\")\n",
    "    \n",
    "if len(df_results['Lambda'].unique()) != 1:\n",
    "    fig = df_results.plot(x='Label', y=['Lambda'], kind=\"bar\", title='Pegasos lambdas',\n",
    "                          color='y', figsize=(6, 4.5)).get_figure()\n",
    "    \n",
    "    fig.savefig(\"pegasos_lambda.png\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Our final average F1 test score is:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.9051360658613312"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_results[\"F1 Test score\"].mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Perceptron"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here's our Perceptron. We'll replicate the same workflow as we did for Pegasos and we'll compare their results."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Perceptron(BaseEstimator, MyBaseClassifier):\n",
    "    def __init__(self, T=1):\n",
    "        self.T = T\n",
    "    \n",
    "    def fit(self, X, Y):\n",
    "        # Initialize the weight vector to all zeros.\n",
    "        self.w = np.zeros(X.shape[1])\n",
    "        \n",
    "        # Repeat for each epoch up to T\n",
    "        for t in range(self.T):\n",
    "            \n",
    "            # For each training example\n",
    "            for i in range(X.shape[0]-1):\n",
    "                \n",
    "                #Compute the output score for this instance.\n",
    "                score = X[i].dot(self.w.T)\n",
    "\n",
    "                # If there was an error, update the weights.\n",
    "                if Y[i]*score <= 0:\n",
    "                    self.w += Y[i]*X[i]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Label 1: 33"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "currentLabel = y1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We split the dataset and run the Perceptron for the first time."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = train_test_split(X,currentLabel,test_size=0.33,random_state=42,stratify=currentLabel)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "classifier = Perceptron()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "classifier.fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(0.9345228076081468, 0.8605564089435057)"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "classifier.score(X_train, y_train), classifier.score(X_test, y_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Our perceptron is clearly suffering from overfitting."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now check other metrics:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "True positive:  2521\n",
      "False positive:  401\n",
      "FP ratio:  0.13723477070499657 ( 13.72 %)\n",
      "True negative:  3216\n",
      "False negative:  416\n",
      "FN ratio:  0.1145374449339207 ( 11.45 %)\n",
      "Precision:  0.8627652292950034 ( 86.28 %)\n",
      "Recall:  0.8583588695948247 ( 85.84 %)\n",
      "Accuracy:  0.8753433018004272 ( 87.53 %)\n"
     ]
    }
   ],
   "source": [
    "TPs_perceptron, FPs_perceptron, TNs_perceptron, FNs_perceptron = [], [], [], []\n",
    "precisions_perceptron, recalls_perceptron, accs_perceptron = [], [], []\n",
    "\n",
    "y_pred = classifier.predict(X_test)\n",
    "\n",
    "tn, fp, fn, tp = confusion_matrix(y_test, y_pred).ravel()\n",
    "precision = precision_score(y_test, y_pred)\n",
    "recall = recall_score(y_test, y_pred)\n",
    "acc = accuracy_score(y_test, y_pred)\n",
    "    \n",
    "TNs_perceptron.append(tn)\n",
    "TPs_perceptron.append(tp)\n",
    "FPs_perceptron.append(fp)\n",
    "FNs_perceptron.append(fn)\n",
    "precisions_perceptron.append(precision)\n",
    "recalls_perceptron.append(recall)\n",
    "accs_perceptron.append(acc)\n",
    "    \n",
    "print(\"True positive: \", tp)\n",
    "print(\"False positive: \", fp)\n",
    "print(\"FP ratio: \", fp/(tp+fp),\"(\",np.round(fp/(tp+fp)*100, 2),\"%)\")\n",
    "print(\"True negative: \", tn)\n",
    "print(\"False negative: \", fn)\n",
    "print(\"FN ratio: \", fn/(tn+fn),\"(\",np.round(fn/(tn+fn)*100, 2),\"%)\")\n",
    "print(\"Precision: \", precision,\"(\",np.round(precision*100, 2),\"%)\")\n",
    "print(\"Recall: \", recall,\"(\",np.round(recall*100, 2),\"%)\")\n",
    "print(\"Accuracy: \", acc,\"(\",np.round(acc*100, 2),\"%)\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We plot the learning curve"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[learning_curve] Training set sizes: [ 1000  2000  3000  4000  5000  6000  7000  8000  9000 10000 11000 12000\n",
      " 13000]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   5 out of 130 | elapsed:    1.4s remaining:   34.5s\n",
      "[Parallel(n_jobs=-1)]: Done  19 out of 130 | elapsed:    4.9s remaining:   28.4s\n",
      "[Parallel(n_jobs=-1)]: Done  33 out of 130 | elapsed:    8.2s remaining:   24.2s\n",
      "[Parallel(n_jobs=-1)]: Done  47 out of 130 | elapsed:   12.1s remaining:   21.3s\n",
      "[Parallel(n_jobs=-1)]: Done  61 out of 130 | elapsed:   16.6s remaining:   18.8s\n",
      "[Parallel(n_jobs=-1)]: Done  75 out of 130 | elapsed:   21.1s remaining:   15.5s\n",
      "[Parallel(n_jobs=-1)]: Done  89 out of 130 | elapsed:   26.1s remaining:   12.0s\n",
      "[Parallel(n_jobs=-1)]: Done 103 out of 130 | elapsed:   30.7s remaining:    8.1s\n",
      "[Parallel(n_jobs=-1)]: Done 117 out of 130 | elapsed:   35.6s remaining:    4.0s\n",
      "[Parallel(n_jobs=-1)]: Done 130 out of 130 | elapsed:   40.2s finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[0.62365591, 0.80645161, 0.8262863 , 0.78139535, 0.82923977,\n",
       "        0.77031802, 0.81485527, 0.82907268, 0.77870216, 0.80550775],\n",
       "       [0.81813031, 0.86085544, 0.86203616, 0.84952804, 0.8807543 ,\n",
       "        0.80816327, 0.84150156, 0.84352431, 0.83958333, 0.84527064],\n",
       "       [0.86153846, 0.85211655, 0.78764897, 0.73921442, 0.76714101,\n",
       "        0.76394194, 0.79528024, 0.78864009, 0.75274056, 0.8045977 ],\n",
       "       [0.8346915 , 0.81362007, 0.84576613, 0.81747457, 0.85061728,\n",
       "        0.79407407, 0.84049774, 0.84400826, 0.81524249, 0.84241706],\n",
       "       [0.85934066, 0.85092127, 0.87317073, 0.86023835, 0.89608939,\n",
       "        0.832315  , 0.871903  , 0.86367969, 0.87539599, 0.86684932],\n",
       "       [0.82790698, 0.80282519, 0.80487805, 0.83273165, 0.88019266,\n",
       "        0.80783723, 0.86023959, 0.84806776, 0.82714617, 0.84463107],\n",
       "       [0.86053077, 0.83642496, 0.86595852, 0.83828775, 0.90086957,\n",
       "        0.84550562, 0.88791689, 0.86886887, 0.86636971, 0.87413395],\n",
       "       [0.87119956, 0.86280992, 0.86705202, 0.85020243, 0.86221678,\n",
       "        0.80358476, 0.84693878, 0.8168398 , 0.79832936, 0.83099451],\n",
       "       [0.84745763, 0.80817799, 0.84068891, 0.83129584, 0.86637663,\n",
       "        0.84620939, 0.88549618, 0.85568587, 0.85262564, 0.86966686],\n",
       "       [0.88840501, 0.86302112, 0.88461538, 0.86535211, 0.90973036,\n",
       "        0.84319527, 0.88173237, 0.85817432, 0.86541738, 0.87440758],\n",
       "       [0.84167157, 0.81058328, 0.86054951, 0.83687943, 0.87819732,\n",
       "        0.81907646, 0.85138249, 0.82154516, 0.81953544, 0.83261272],\n",
       "       [0.87689713, 0.86314594, 0.875965  , 0.85467128, 0.89761905,\n",
       "        0.84678016, 0.88876712, 0.87048193, 0.88973799, 0.8847032 ],\n",
       "       [0.88683351, 0.85620915, 0.90014903, 0.86483455, 0.90102389,\n",
       "        0.85434174, 0.90753608, 0.85458377, 0.85648415, 0.86211031]])"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sizes = range(1000, 14000, 1000)\n",
    "train_size, train_score, val_score = learning_curve(Perceptron(), X, currentLabel, train_sizes=sizes, cv=10,\n",
    "                                                   n_jobs=-1, verbose=10)\n",
    "val_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(6, 5))\n",
    "plt.title('Perceptron vs. training size (Label 33)')\n",
    "plt.plot(train_size, np.mean(val_score, 1), label='Validation')\n",
    "plt.plot(train_size, np.mean(train_score, 1), label='Training')\n",
    "plt.legend()\n",
    "plt.xlabel('Training size')\n",
    "plt.ylabel('Score')\n",
    "\n",
    "plt.savefig('perceptron_learning_curve_y1.png')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's see the validation curve"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   2 out of 100 | elapsed:   11.9s remaining:  9.7min\n",
      "[Parallel(n_jobs=-1)]: Done  13 out of 100 | elapsed:   56.4s remaining:  6.3min\n",
      "[Parallel(n_jobs=-1)]: Done  24 out of 100 | elapsed:  1.6min remaining:  5.1min\n",
      "[Parallel(n_jobs=-1)]: Done  35 out of 100 | elapsed:  2.4min remaining:  4.4min\n",
      "[Parallel(n_jobs=-1)]: Done  46 out of 100 | elapsed:  3.0min remaining:  3.6min\n",
      "[Parallel(n_jobs=-1)]: Done  57 out of 100 | elapsed:  3.7min remaining:  2.8min\n",
      "[Parallel(n_jobs=-1)]: Done  68 out of 100 | elapsed:  4.4min remaining:  2.1min\n",
      "[Parallel(n_jobs=-1)]: Done  79 out of 100 | elapsed:  5.2min remaining:  1.4min\n",
      "[Parallel(n_jobs=-1)]: Done  90 out of 100 | elapsed:  5.9min remaining:   39.4s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:  6.6min finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "array([[0.89422028, 0.88324324, 0.90873786, 0.87416481, 0.91925466,\n",
       "        0.8654105 , 0.9118896 , 0.89029126, 0.89736701, 0.8838008 ],\n",
       "       [0.8989845 , 0.88381963, 0.91250618, 0.88472459, 0.92573402,\n",
       "        0.8719346 , 0.91318328, 0.88803089, 0.89685125, 0.90249433],\n",
       "       [0.89314195, 0.87919106, 0.91144279, 0.88611898, 0.92281304,\n",
       "        0.87733888, 0.91190603, 0.89936181, 0.88131868, 0.90602138],\n",
       "       [0.90447922, 0.87919106, 0.91411649, 0.89113356, 0.92562929,\n",
       "        0.88054608, 0.90987124, 0.89805825, 0.89397975, 0.89624724],\n",
       "       [0.90287575, 0.88463583, 0.91248771, 0.8836425 , 0.93333333,\n",
       "        0.87777778, 0.91257996, 0.89985344, 0.89428263, 0.90454545],\n",
       "       [0.90584416, 0.88197425, 0.91015625, 0.88463703, 0.92914286,\n",
       "        0.88070893, 0.91483664, 0.89970501, 0.89875474, 0.9068323 ],\n",
       "       [0.90118153, 0.88142934, 0.91420118, 0.88342857, 0.92795389,\n",
       "        0.8744884 , 0.91121744, 0.89882122, 0.88767123, 0.90457369],\n",
       "       [0.90010858, 0.87969516, 0.91295747, 0.88433048, 0.93249428,\n",
       "        0.8778157 , 0.91017316, 0.90410959, 0.89264305, 0.90326209],\n",
       "       [0.89813801, 0.88203463, 0.91687407, 0.88598979, 0.93271995,\n",
       "        0.87939352, 0.91045576, 0.89970501, 0.89256198, 0.90056497],\n",
       "       [0.90623306, 0.87588846, 0.91508492, 0.88776086, 0.92873563,\n",
       "        0.8773842 , 0.90574215, 0.90303624, 0.89851892, 0.89802817]])"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "T = range(1,11)\n",
    "train_score, val_score = validation_curve(Perceptron(), X, currentLabel, 'T', T, cv=10, verbose=10, n_jobs=-1)\n",
    "val_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x360 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(6, 5))\n",
    "plt.title('Perceptron vs. number of epochs (Label 33)')\n",
    "plt.plot(T, np.mean(val_score, 1), label='Testing')\n",
    "plt.plot(T, np.mean(train_score, 1), label='Training')\n",
    "plt.legend()\n",
    "plt.xlabel('Number of epochs')\n",
    "plt.ylabel('Score')\n",
    "\n",
    "plt.savefig('perceptron_validation_curve_y1.png')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now want to check what's the best number of epochs `T`. Therefore, we use `GridSearchCV` to estimate it. Again, we run a 10-fold cross validation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 9 candidates, totalling 90 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Done   2 tasks      | elapsed:    7.0s\n",
      "[Parallel(n_jobs=-1)]: Done   9 tasks      | elapsed:   13.5s\n",
      "[Parallel(n_jobs=-1)]: Done  16 tasks      | elapsed:   20.3s\n",
      "[Parallel(n_jobs=-1)]: Done  25 tasks      | elapsed:   44.2s\n",
      "[Parallel(n_jobs=-1)]: Done  34 tasks      | elapsed:  1.2min\n",
      "[Parallel(n_jobs=-1)]: Done  45 tasks      | elapsed:  1.8min\n",
      "[Parallel(n_jobs=-1)]: Done  56 tasks      | elapsed:  2.5min\n",
      "[Parallel(n_jobs=-1)]: Done  69 tasks      | elapsed:  3.6min\n",
      "[Parallel(n_jobs=-1)]: Done  85 out of  90 | elapsed:  5.2min remaining:   18.2s\n",
      "[Parallel(n_jobs=-1)]: Done  90 out of  90 | elapsed:  5.4min finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[{'T': 6}, 0.9012592157015665]"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "params_perceptron = []\n",
    "param_grid = [{'T':range(1,10)}]\n",
    "\n",
    "svc = Perceptron()\n",
    "clf = GridSearchCV(svc, param_grid, cv = 10, verbose=10, n_jobs=-1)\n",
    "\n",
    "clf.fit(X, currentLabel)\n",
    "\n",
    "params_perceptron.append([clf.best_params_, clf.best_score_])\n",
    "params_perceptron[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Other labels: 70, 101, 4"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now repeat the same workflow for the other labels.\n",
    "However, here we also upsample data in order to check if it can help the algorithm in the prediction."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "LABEL  70\n",
      "==============\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>Metric</th>\n",
       "      <th>Not resampled</th>\n",
       "      <th>Oversampled</th>\n",
       "      <th>Undersampled</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>True positive</td>\n",
       "      <td>1675</td>\n",
       "      <td>1825</td>\n",
       "      <td>1825</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>False positive</td>\n",
       "      <td>230</td>\n",
       "      <td>1295</td>\n",
       "      <td>1295</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>True negative</td>\n",
       "      <td>4465</td>\n",
       "      <td>3400</td>\n",
       "      <td>3400</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>False negative</td>\n",
       "      <td>184</td>\n",
       "      <td>34</td>\n",
       "      <td>34</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>FP ratio</td>\n",
       "      <td>0.12073490813648294 (12.07%)</td>\n",
       "      <td>0.4150641025641026 (41.51%)</td>\n",
       "      <td>0.4150641025641026 (41.51%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>FN ratio</td>\n",
       "      <td>0.039578403957840395 (3.96%)</td>\n",
       "      <td>0.009900990099009901 (0.99%)</td>\n",
       "      <td>0.009900990099009901 (0.99%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>Training score</td>\n",
       "      <td>0.949758580190526 (94.98%)</td>\n",
       "      <td>0.8978284422252579 (89.78%)</td>\n",
       "      <td>0.8978284422252579 (89.78%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>Test score</td>\n",
       "      <td>0.8900106269925611 (89.0%)</td>\n",
       "      <td>0.7330789315123519 (73.31%)</td>\n",
       "      <td>0.7330789315123519 (73.31%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>Precision</td>\n",
       "      <td>0.8792650918635171 (87.93%)</td>\n",
       "      <td>0.5849358974358975 (58.49%)</td>\n",
       "      <td>0.5849358974358975 (58.49%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>Recall</td>\n",
       "      <td>0.9010220548682087 (90.1%)</td>\n",
       "      <td>0.9817105970952125 (90.1%)</td>\n",
       "      <td>0.9817105970952125 (90.1%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>Accuracy</td>\n",
       "      <td>0.9368324687213915 (93.68%)</td>\n",
       "      <td>0.7972230698809887 (79.72%)</td>\n",
       "      <td>0.7972230698809887 (79.72%)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            Metric                 Not resampled  \\\n",
       "0    True positive                          1675   \n",
       "1   False positive                           230   \n",
       "2    True negative                          4465   \n",
       "3   False negative                           184   \n",
       "4         FP ratio  0.12073490813648294 (12.07%)   \n",
       "5         FN ratio  0.039578403957840395 (3.96%)   \n",
       "6   Training score    0.949758580190526 (94.98%)   \n",
       "7       Test score    0.8900106269925611 (89.0%)   \n",
       "8        Precision   0.8792650918635171 (87.93%)   \n",
       "9           Recall    0.9010220548682087 (90.1%)   \n",
       "10        Accuracy   0.9368324687213915 (93.68%)   \n",
       "\n",
       "                     Oversampled                  Undersampled  \n",
       "0                           1825                          1825  \n",
       "1                           1295                          1295  \n",
       "2                           3400                          3400  \n",
       "3                             34                            34  \n",
       "4    0.4150641025641026 (41.51%)   0.4150641025641026 (41.51%)  \n",
       "5   0.009900990099009901 (0.99%)  0.009900990099009901 (0.99%)  \n",
       "6    0.8978284422252579 (89.78%)   0.8978284422252579 (89.78%)  \n",
       "7    0.7330789315123519 (73.31%)   0.7330789315123519 (73.31%)  \n",
       "8    0.5849358974358975 (58.49%)   0.5849358974358975 (58.49%)  \n",
       "9     0.9817105970952125 (90.1%)    0.9817105970952125 (90.1%)  \n",
       "10   0.7972230698809887 (79.72%)   0.7972230698809887 (79.72%)  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "LABEL  101\n",
      "==============\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>Metric</th>\n",
       "      <th>Not resampled</th>\n",
       "      <th>Oversampled</th>\n",
       "      <th>Undersampled</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>True positive</td>\n",
       "      <td>1355</td>\n",
       "      <td>1580</td>\n",
       "      <td>1580</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>False positive</td>\n",
       "      <td>167</td>\n",
       "      <td>1060</td>\n",
       "      <td>1060</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>True negative</td>\n",
       "      <td>4779</td>\n",
       "      <td>3886</td>\n",
       "      <td>3886</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>False negative</td>\n",
       "      <td>253</td>\n",
       "      <td>28</td>\n",
       "      <td>28</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>FP ratio</td>\n",
       "      <td>0.10972404730617609 (10.97%)</td>\n",
       "      <td>0.4015151515151515 (40.15%)</td>\n",
       "      <td>0.4015151515151515 (40.15%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>FN ratio</td>\n",
       "      <td>0.05027821939586646 (5.03%)</td>\n",
       "      <td>0.007153806847215125 (0.72%)</td>\n",
       "      <td>0.007153806847215125 (0.72%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>Training score</td>\n",
       "      <td>0.9373542670604694 (93.74%)</td>\n",
       "      <td>0.9190351961188155 (91.9%)</td>\n",
       "      <td>0.9190351961188155 (91.9%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>Test score</td>\n",
       "      <td>0.865814696485623 (86.58%)</td>\n",
       "      <td>0.7438794726930321 (74.39%)</td>\n",
       "      <td>0.7438794726930321 (74.39%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>Precision</td>\n",
       "      <td>0.8902759526938239 (89.03%)</td>\n",
       "      <td>0.5984848484848485 (59.85%)</td>\n",
       "      <td>0.5984848484848485 (59.85%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>Recall</td>\n",
       "      <td>0.8426616915422885 (84.27%)</td>\n",
       "      <td>0.9825870646766169 (84.27%)</td>\n",
       "      <td>0.9825870646766169 (84.27%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>Accuracy</td>\n",
       "      <td>0.9359169972535856 (93.59%)</td>\n",
       "      <td>0.8339945071711932 (83.4%)</td>\n",
       "      <td>0.8339945071711932 (83.4%)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            Metric                 Not resampled  \\\n",
       "0    True positive                          1355   \n",
       "1   False positive                           167   \n",
       "2    True negative                          4779   \n",
       "3   False negative                           253   \n",
       "4         FP ratio  0.10972404730617609 (10.97%)   \n",
       "5         FN ratio   0.05027821939586646 (5.03%)   \n",
       "6   Training score   0.9373542670604694 (93.74%)   \n",
       "7       Test score    0.865814696485623 (86.58%)   \n",
       "8        Precision   0.8902759526938239 (89.03%)   \n",
       "9           Recall   0.8426616915422885 (84.27%)   \n",
       "10        Accuracy   0.9359169972535856 (93.59%)   \n",
       "\n",
       "                     Oversampled                  Undersampled  \n",
       "0                           1580                          1580  \n",
       "1                           1060                          1060  \n",
       "2                           3886                          3886  \n",
       "3                             28                            28  \n",
       "4    0.4015151515151515 (40.15%)   0.4015151515151515 (40.15%)  \n",
       "5   0.007153806847215125 (0.72%)  0.007153806847215125 (0.72%)  \n",
       "6     0.9190351961188155 (91.9%)    0.9190351961188155 (91.9%)  \n",
       "7    0.7438794726930321 (74.39%)   0.7438794726930321 (74.39%)  \n",
       "8    0.5984848484848485 (59.85%)   0.5984848484848485 (59.85%)  \n",
       "9    0.9825870646766169 (84.27%)   0.9825870646766169 (84.27%)  \n",
       "10    0.8339945071711932 (83.4%)    0.8339945071711932 (83.4%)  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "LABEL  4\n",
      "==============\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>Metric</th>\n",
       "      <th>Not resampled</th>\n",
       "      <th>Oversampled</th>\n",
       "      <th>Undersampled</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>True positive</td>\n",
       "      <td>966</td>\n",
       "      <td>1117</td>\n",
       "      <td>1117</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>False positive</td>\n",
       "      <td>181</td>\n",
       "      <td>1099</td>\n",
       "      <td>1099</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>True negative</td>\n",
       "      <td>5230</td>\n",
       "      <td>4312</td>\n",
       "      <td>4312</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>False negative</td>\n",
       "      <td>177</td>\n",
       "      <td>26</td>\n",
       "      <td>26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>FP ratio</td>\n",
       "      <td>0.15780296425457715 (15.78%)</td>\n",
       "      <td>0.49593862815884476 (49.59%)</td>\n",
       "      <td>0.49593862815884476 (49.59%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>FN ratio</td>\n",
       "      <td>0.03273534307379323 (3.27%)</td>\n",
       "      <td>0.00599354541263255 (0.6%)</td>\n",
       "      <td>0.00599354541263255 (0.6%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>Training score</td>\n",
       "      <td>0.9219978746014877 (92.2%)</td>\n",
       "      <td>0.9175124253435242 (91.75%)</td>\n",
       "      <td>0.9175124253435242 (91.75%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>Test score</td>\n",
       "      <td>0.8436681222707424 (84.37%)</td>\n",
       "      <td>0.6650788925275379 (66.51%)</td>\n",
       "      <td>0.6650788925275379 (66.51%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>Precision</td>\n",
       "      <td>0.8421970357454228 (84.22%)</td>\n",
       "      <td>0.5040613718411552 (50.41%)</td>\n",
       "      <td>0.5040613718411552 (50.41%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>Recall</td>\n",
       "      <td>0.8451443569553806 (84.51%)</td>\n",
       "      <td>0.9772528433945756 (84.51%)</td>\n",
       "      <td>0.9772528433945756 (84.51%)</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>Accuracy</td>\n",
       "      <td>0.9453768690875801 (94.54%)</td>\n",
       "      <td>0.82834909978639 (82.83%)</td>\n",
       "      <td>0.82834909978639 (82.83%)</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            Metric                 Not resampled  \\\n",
       "0    True positive                           966   \n",
       "1   False positive                           181   \n",
       "2    True negative                          5230   \n",
       "3   False negative                           177   \n",
       "4         FP ratio  0.15780296425457715 (15.78%)   \n",
       "5         FN ratio   0.03273534307379323 (3.27%)   \n",
       "6   Training score    0.9219978746014877 (92.2%)   \n",
       "7       Test score   0.8436681222707424 (84.37%)   \n",
       "8        Precision   0.8421970357454228 (84.22%)   \n",
       "9           Recall   0.8451443569553806 (84.51%)   \n",
       "10        Accuracy   0.9453768690875801 (94.54%)   \n",
       "\n",
       "                     Oversampled                  Undersampled  \n",
       "0                           1117                          1117  \n",
       "1                           1099                          1099  \n",
       "2                           4312                          4312  \n",
       "3                             26                            26  \n",
       "4   0.49593862815884476 (49.59%)  0.49593862815884476 (49.59%)  \n",
       "5     0.00599354541263255 (0.6%)    0.00599354541263255 (0.6%)  \n",
       "6    0.9175124253435242 (91.75%)   0.9175124253435242 (91.75%)  \n",
       "7    0.6650788925275379 (66.51%)   0.6650788925275379 (66.51%)  \n",
       "8    0.5040613718411552 (50.41%)   0.5040613718411552 (50.41%)  \n",
       "9    0.9772528433945756 (84.51%)   0.9772528433945756 (84.51%)  \n",
       "10     0.82834909978639 (82.83%)     0.82834909978639 (82.83%)  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "lbl_idx = { '2': '70', '3': '101', '4': '4' }\n",
    "TPs_resampled_perceptr, FPs_resampled_perceptr, TNs_resampled_perceptr, FNs_resampled_perceptr, precisions_resampled_perceptr, recalls_resampled_perceptr, accs_resampled_perceptr = [], [], [], [], [], [], []\n",
    "TPs_unders_perceptr, FPs_unders_perceptr, TNs_unders_perceptr, FNs_unders_perceptr, precisions_unders_perceptr, recalls_unders_perceptr, accs_unders_perceptr = [], [], [], [], [], [], []\n",
    "\n",
    "for i in range(2,5):    \n",
    "    currentLabel = eval('y'+str(i))\n",
    "    \n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, currentLabel, test_size=0.33,\n",
    "                                                        random_state=42, stratify=currentLabel)\n",
    "    \n",
    "    classifier = Perceptron()\n",
    "    classifier.fit(X_train, y_train)\n",
    "    train_score = classifier.score(X_train, y_train)\n",
    "    \n",
    "    y_pred = classifier.predict(X_test)  \n",
    "    f1 = f1_score(y_test, y_pred)\n",
    "    tn, fp, fn, tp = confusion_matrix(y_test, y_pred).ravel()\n",
    "    precision = precision_score(y_test, y_pred)\n",
    "    recall = recall_score(y_test, y_pred)\n",
    "    acc = accuracy_score(y_test, y_pred)\n",
    "    \n",
    "    TNs_perceptron.append(tn)\n",
    "    TPs_perceptron.append(tp)\n",
    "    FPs_perceptron.append(fp)\n",
    "    FNs_perceptron.append(fn)\n",
    "    precisions_perceptron.append(precision)\n",
    "    recalls_perceptron.append(recall)\n",
    "    accs_perceptron.append(acc)\n",
    "    \n",
    "    # Up-sample\n",
    "    ros = RandomOverSampler(random_state=42)\n",
    "    X_train_resampled, y_train_resampled = ros.fit_resample(X_train, y_train)\n",
    "    \n",
    "    classifier = Perceptron()\n",
    "    classifier.fit(X_train_resampled, y_train_resampled)\n",
    "    train_score_resampled = classifier.score(X_train_resampled, y_train_resampled)\n",
    "    \n",
    "    y_pred = classifier.predict(X_test)  \n",
    "    f1_resampled = f1_score(y_test, y_pred)  \n",
    "    tn_resampled, fp_resampled, fn_resampled, tp_resampled = confusion_matrix(y_test, y_pred).ravel()\n",
    "    precision_resampled = precision_score(y_test, y_pred)\n",
    "    recall_resampled = recall_score(y_test, y_pred)\n",
    "    acc_resampled = accuracy_score(y_test, y_pred)\n",
    "    \n",
    "    TNs_resampled_perceptr.append(tn_resampled)\n",
    "    TPs_resampled_perceptr.append(tp_resampled)\n",
    "    FPs_resampled_perceptr.append(fp_resampled)\n",
    "    FNs_resampled_perceptr.append(fn_resampled)\n",
    "    precisions_resampled_perceptr.append(precision_resampled)\n",
    "    recalls_resampled_perceptr.append(recall_resampled)\n",
    "    accs_resampled_perceptr.append(acc_resampled)\n",
    "    \n",
    "    # Under-sample\n",
    "    rus = RandomUnderSampler(random_state=42, replacement=True)\n",
    "    X_train_unders, y_train_unders = ros.fit_resample(X_train, y_train)\n",
    "    \n",
    "    classifier = Perceptron()\n",
    "    classifier.fit(X_train_unders, y_train_unders)\n",
    "    train_score_unders = classifier.score(X_train_unders, y_train_unders)\n",
    "    \n",
    "    y_pred = classifier.predict(X_test)  \n",
    "    f1_unders = f1_score(y_test, y_pred)\n",
    "    tn_unders, fp_unders, fn_unders, tp_unders = confusion_matrix(y_test, y_pred).ravel()\n",
    "    precision_unders = precision_score(y_test, y_pred)\n",
    "    recall_unders = recall_score(y_test, y_pred)\n",
    "    acc_unders = accuracy_score(y_test, y_pred)\n",
    "    \n",
    "    TNs_unders_perceptr.append(tn_unders)\n",
    "    TPs_unders_perceptr.append(tp_unders)\n",
    "    FPs_unders_perceptr.append(fp_unders)\n",
    "    FNs_unders_perceptr.append(fn_unders)\n",
    "    precisions_unders_perceptr.append(precision_unders)\n",
    "    recalls_unders_perceptr.append(recall_unders)\n",
    "    accs_unders_perceptr.append(acc_unders)\n",
    "    \n",
    "    print(\"\\nLABEL \", lbl_idx.get(str(i)))\n",
    "    print(\"==============\")\n",
    "    \n",
    "    metrics_name = [\"True positive\", \"False positive\", \"True negative\", \"False negative\", \"FP ratio\", \"FN ratio\",\n",
    "                    \"Training score\", \"Test score\", \"Precision\", \"Recall\", \"Accuracy\"]\n",
    "    nr = [tp,fp,tn,fn,\n",
    "          str(fp/(tp+fp))+\" (\"+str(np.round(fp/(tp+fp)*100, 2))+\"%)\",\n",
    "          str(fn/(tn+fn))+\" (\"+str(np.round(fn/(tn+fn)*100, 2))+\"%)\",\n",
    "          str(train_score)+\" (\"+str(np.round(train_score*100, 2))+\"%)\",\n",
    "          str(f1)+\" (\"+str(np.round(f1*100, 2))+\"%)\",\n",
    "          str(precision)+\" (\"+str(np.round(precision*100, 2))+\"%)\",\n",
    "          str(recall)+\" (\"+str(np.round(recall*100, 2))+\"%)\",\n",
    "          str(acc)+\" (\"+str(np.round(acc*100, 2))+\"%)\"]\n",
    "    \n",
    "    ovs = [tp_resampled,fp_resampled, tn_resampled,fn_resampled,\n",
    "          str(fp_resampled/(tp_resampled+fp_resampled))+\" (\"+str(np.round(fp_resampled/(tp_resampled+fp_resampled)*100, 2))+\"%)\",\n",
    "          str(fn_resampled/(tn_resampled+fn_resampled))+\" (\"+str(np.round(fn_resampled/(tn_resampled+fn_resampled)*100, 2))+\"%)\",\n",
    "          str(train_score_resampled)+\" (\"+str(np.round(train_score_resampled*100, 2))+\"%)\",\n",
    "          str(f1_resampled)+\" (\"+str(np.round(f1_resampled*100, 2))+\"%)\",\n",
    "          str(precision_resampled)+\" (\"+str(np.round(precision_resampled*100, 2))+\"%)\",\n",
    "          str(recall_resampled)+\" (\"+str(np.round(recall*100, 2))+\"%)\",\n",
    "          str(acc_resampled)+\" (\"+str(np.round(acc_resampled*100, 2))+\"%)\"]\n",
    "    \n",
    "    uds = [tp_unders,fp_unders, tn_unders,fn_unders,\n",
    "          str(fp_unders/(tp_unders+fp_unders))+\" (\"+str(np.round(fp_unders/(tp_unders+fp_unders)*100, 2))+\"%)\",\n",
    "          str(fn_unders/(tn_unders+fn_unders))+\" (\"+str(np.round(fn_unders/(tn_unders+fn_unders)*100, 2))+\"%)\",\n",
    "          str(train_score_unders)+\" (\"+str(np.round(train_score_unders*100, 2))+\"%)\",\n",
    "          str(f1_unders)+\" (\"+str(np.round(f1_unders*100, 2))+\"%)\",\n",
    "          str(precision_unders)+\" (\"+str(np.round(precision_unders*100, 2))+\"%)\",\n",
    "          str(recall_unders)+\" (\"+str(np.round(recall*100, 2))+\"%)\",\n",
    "          str(acc_unders)+\" (\"+str(np.round(acc_unders*100, 2))+\"%)\"]\n",
    "    \n",
    "    df1 = pd.DataFrame(np.column_stack([metrics_name, nr, ovs, uds]), columns=[\"Metric\", \"Not resampled\", \"Oversampled\", \"Undersampled\"])\n",
    "    display(df1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's now see the Learning curve"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[learning_curve] Training set sizes: [ 1000  2000  3000  4000  5000  6000  7000  8000  9000 10000 11000 12000\n",
      " 13000]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   5 out of 130 | elapsed:    1.3s remaining:   31.4s\n",
      "[Parallel(n_jobs=-1)]: Done  19 out of 130 | elapsed:    4.5s remaining:   26.3s\n",
      "[Parallel(n_jobs=-1)]: Done  33 out of 130 | elapsed:    7.9s remaining:   23.2s\n",
      "[Parallel(n_jobs=-1)]: Done  47 out of 130 | elapsed:   11.4s remaining:   20.1s\n",
      "[Parallel(n_jobs=-1)]: Done  61 out of 130 | elapsed:   16.1s remaining:   18.2s\n",
      "[Parallel(n_jobs=-1)]: Done  75 out of 130 | elapsed:   20.3s remaining:   14.9s\n",
      "[Parallel(n_jobs=-1)]: Done  89 out of 130 | elapsed:   24.8s remaining:   11.4s\n",
      "[Parallel(n_jobs=-1)]: Done 103 out of 130 | elapsed:   28.9s remaining:    7.6s\n",
      "[Parallel(n_jobs=-1)]: Done 117 out of 130 | elapsed:   33.4s remaining:    3.7s\n",
      "[Parallel(n_jobs=-1)]: Done 130 out of 130 | elapsed:   37.5s finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[learning_curve] Training set sizes: [ 1000  2000  3000  4000  5000  6000  7000  8000  9000 10000 11000 12000\n",
      " 13000]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Done   5 out of 130 | elapsed:    1.8s remaining:   45.0s\n",
      "[Parallel(n_jobs=-1)]: Done  19 out of 130 | elapsed:    6.2s remaining:   35.9s\n",
      "[Parallel(n_jobs=-1)]: Done  33 out of 130 | elapsed:   10.2s remaining:   30.1s\n",
      "[Parallel(n_jobs=-1)]: Done  47 out of 130 | elapsed:   14.5s remaining:   25.6s\n",
      "[Parallel(n_jobs=-1)]: Done  61 out of 130 | elapsed:   19.0s remaining:   21.5s\n",
      "[Parallel(n_jobs=-1)]: Done  75 out of 130 | elapsed:   23.0s remaining:   16.9s\n",
      "[Parallel(n_jobs=-1)]: Done  89 out of 130 | elapsed:   27.6s remaining:   12.7s\n",
      "[Parallel(n_jobs=-1)]: Done 103 out of 130 | elapsed:   31.8s remaining:    8.3s\n",
      "[Parallel(n_jobs=-1)]: Done 117 out of 130 | elapsed:   36.3s remaining:    4.0s\n",
      "[Parallel(n_jobs=-1)]: Done 130 out of 130 | elapsed:   40.4s finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[learning_curve] Training set sizes: [ 1000  2000  3000  4000  5000  6000  7000  8000  9000 10000 11000 12000\n",
      " 13000]\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Done   5 out of 130 | elapsed:    1.8s remaining:   44.6s\n",
      "[Parallel(n_jobs=-1)]: Done  19 out of 130 | elapsed:    6.1s remaining:   35.7s\n",
      "[Parallel(n_jobs=-1)]: Done  33 out of 130 | elapsed:   10.4s remaining:   30.7s\n",
      "[Parallel(n_jobs=-1)]: Done  47 out of 130 | elapsed:   14.8s remaining:   26.2s\n",
      "[Parallel(n_jobs=-1)]: Done  61 out of 130 | elapsed:   19.8s remaining:   22.4s\n",
      "[Parallel(n_jobs=-1)]: Done  75 out of 130 | elapsed:   24.0s remaining:   17.6s\n",
      "[Parallel(n_jobs=-1)]: Done  89 out of 130 | elapsed:   28.6s remaining:   13.2s\n",
      "[Parallel(n_jobs=-1)]: Done 103 out of 130 | elapsed:   32.9s remaining:    8.6s\n",
      "[Parallel(n_jobs=-1)]: Done 117 out of 130 | elapsed:   37.7s remaining:    4.2s\n",
      "[Parallel(n_jobs=-1)]: Done 130 out of 130 | elapsed:   41.6s finished\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(20, 5))\n",
    "\n",
    "for i in range(2,5):\n",
    "    currentLabel = eval('y'+str(i))\n",
    "    \n",
    "    sizes = range(1000, 14000, 1000)\n",
    "    train_size, train_score, val_score = learning_curve(Perceptron(), X, currentLabel, train_sizes=sizes, cv=10,\n",
    "                                                   n_jobs=-1, verbose=10)\n",
    "    \n",
    "    plt.subplot(1, 3, i-1)\n",
    "    plt.title('Perceptron vs. training size (Label ' + lbl_idx.get(str(i)) + ')')\n",
    "    plt.plot(train_size, np.mean(val_score, 1), label='Validation')\n",
    "    plt.plot(train_size, np.mean(train_score, 1), label='Training')\n",
    "    plt.legend()\n",
    "    plt.xlabel('Training size')\n",
    "    plt.ylabel('Score')\n",
    "\n",
    "plt.savefig('perceptron_learning_curve_other.png')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now plot the _validation curve_ and we run `GridSearchCV` to get the best number of epochs for the Perceptron for every label."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   2 out of 100 | elapsed:    8.4s remaining:  6.9min\n",
      "[Parallel(n_jobs=-1)]: Done  13 out of 100 | elapsed:   49.2s remaining:  5.5min\n",
      "[Parallel(n_jobs=-1)]: Done  24 out of 100 | elapsed:  1.5min remaining:  4.6min\n",
      "[Parallel(n_jobs=-1)]: Done  35 out of 100 | elapsed:  2.2min remaining:  4.0min\n",
      "[Parallel(n_jobs=-1)]: Done  46 out of 100 | elapsed:  2.8min remaining:  3.3min\n",
      "[Parallel(n_jobs=-1)]: Done  57 out of 100 | elapsed:  3.4min remaining:  2.6min\n",
      "[Parallel(n_jobs=-1)]: Done  68 out of 100 | elapsed:  4.1min remaining:  1.9min\n",
      "[Parallel(n_jobs=-1)]: Done  79 out of 100 | elapsed:  4.9min remaining:  1.3min\n",
      "[Parallel(n_jobs=-1)]: Done  90 out of 100 | elapsed:  5.5min remaining:   36.7s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:  6.1min finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 10 candidates, totalling 100 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Done   2 tasks      | elapsed:    6.5s\n",
      "[Parallel(n_jobs=-1)]: Done   9 tasks      | elapsed:   12.9s\n",
      "[Parallel(n_jobs=-1)]: Done  16 tasks      | elapsed:   18.8s\n",
      "[Parallel(n_jobs=-1)]: Done  25 tasks      | elapsed:   41.0s\n",
      "[Parallel(n_jobs=-1)]: Done  34 tasks      | elapsed:  1.1min\n",
      "[Parallel(n_jobs=-1)]: Done  45 tasks      | elapsed:  1.7min\n",
      "[Parallel(n_jobs=-1)]: Done  56 tasks      | elapsed:  2.3min\n",
      "[Parallel(n_jobs=-1)]: Done  69 tasks      | elapsed:  3.4min\n",
      "[Parallel(n_jobs=-1)]: Done  82 tasks      | elapsed:  4.6min\n",
      "[Parallel(n_jobs=-1)]: Done  96 out of 100 | elapsed:  6.0min remaining:   15.0s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:  6.2min finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   2 out of 100 | elapsed:   11.5s remaining:  9.4min\n",
      "[Parallel(n_jobs=-1)]: Done  13 out of 100 | elapsed:   53.9s remaining:  6.0min\n",
      "[Parallel(n_jobs=-1)]: Done  24 out of 100 | elapsed:  1.5min remaining:  4.8min\n",
      "[Parallel(n_jobs=-1)]: Done  35 out of 100 | elapsed:  2.3min remaining:  4.2min\n",
      "[Parallel(n_jobs=-1)]: Done  46 out of 100 | elapsed:  2.9min remaining:  3.4min\n",
      "[Parallel(n_jobs=-1)]: Done  57 out of 100 | elapsed:  3.5min remaining:  2.7min\n",
      "[Parallel(n_jobs=-1)]: Done  68 out of 100 | elapsed:  4.2min remaining:  2.0min\n",
      "[Parallel(n_jobs=-1)]: Done  79 out of 100 | elapsed:  5.0min remaining:  1.3min\n",
      "[Parallel(n_jobs=-1)]: Done  90 out of 100 | elapsed:  5.6min remaining:   37.5s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:  6.2min finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 10 candidates, totalling 100 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Done   2 tasks      | elapsed:    6.0s\n",
      "[Parallel(n_jobs=-1)]: Done   9 tasks      | elapsed:   12.1s\n",
      "[Parallel(n_jobs=-1)]: Done  16 tasks      | elapsed:   18.1s\n",
      "[Parallel(n_jobs=-1)]: Done  25 tasks      | elapsed:   40.4s\n",
      "[Parallel(n_jobs=-1)]: Done  34 tasks      | elapsed:  1.0min\n",
      "[Parallel(n_jobs=-1)]: Done  45 tasks      | elapsed:  1.7min\n",
      "[Parallel(n_jobs=-1)]: Done  56 tasks      | elapsed:  2.3min\n",
      "[Parallel(n_jobs=-1)]: Done  69 tasks      | elapsed:  3.5min\n",
      "[Parallel(n_jobs=-1)]: Done  82 tasks      | elapsed:  4.6min\n",
      "[Parallel(n_jobs=-1)]: Done  96 out of 100 | elapsed:  6.0min remaining:   15.1s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:  6.3min finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done   2 out of 100 | elapsed:   11.2s remaining:  9.2min\n",
      "[Parallel(n_jobs=-1)]: Done  13 out of 100 | elapsed:   52.7s remaining:  5.9min\n",
      "[Parallel(n_jobs=-1)]: Done  24 out of 100 | elapsed:  1.5min remaining:  4.8min\n",
      "[Parallel(n_jobs=-1)]: Done  35 out of 100 | elapsed:  2.2min remaining:  4.1min\n",
      "[Parallel(n_jobs=-1)]: Done  46 out of 100 | elapsed:  2.9min remaining:  3.4min\n",
      "[Parallel(n_jobs=-1)]: Done  57 out of 100 | elapsed:  3.5min remaining:  2.6min\n",
      "[Parallel(n_jobs=-1)]: Done  68 out of 100 | elapsed:  4.2min remaining:  2.0min\n",
      "[Parallel(n_jobs=-1)]: Done  79 out of 100 | elapsed:  4.9min remaining:  1.3min\n",
      "[Parallel(n_jobs=-1)]: Done  90 out of 100 | elapsed:  5.5min remaining:   36.9s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:  6.1min finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 8 concurrent workers.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 10 folds for each of 10 candidates, totalling 100 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Done   2 tasks      | elapsed:    6.0s\n",
      "[Parallel(n_jobs=-1)]: Done   9 tasks      | elapsed:   11.9s\n",
      "[Parallel(n_jobs=-1)]: Done  16 tasks      | elapsed:   17.6s\n",
      "[Parallel(n_jobs=-1)]: Done  25 tasks      | elapsed:   40.0s\n",
      "[Parallel(n_jobs=-1)]: Done  34 tasks      | elapsed:  1.0min\n",
      "[Parallel(n_jobs=-1)]: Done  45 tasks      | elapsed:  1.7min\n",
      "[Parallel(n_jobs=-1)]: Done  56 tasks      | elapsed:  2.3min\n",
      "[Parallel(n_jobs=-1)]: Done  69 tasks      | elapsed:  3.4min\n",
      "[Parallel(n_jobs=-1)]: Done  82 tasks      | elapsed:  4.5min\n",
      "[Parallel(n_jobs=-1)]: Done  96 out of 100 | elapsed:  5.9min remaining:   14.9s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:  6.2min finished\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABJkAAAFNCAYAAACjXb61AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvnQurowAAIABJREFUeJzs3Xd4VGXax/HvnUaAhBqK9KIiiKiAFRSwl7WBXbGLuqtrYxVXXwvqgru6a0FXUcG6yypYUAFFBLtSlKIggoASmnQINeV5/3hOYBgSSCDJmcn8Ptc115w5Zeaek2TuzH2eYs45RERERERERERE9kZS2AGIiIiIiIiIiEj8U5FJRERERERERET2mopMIiIiIiIiIiKy11RkEhERERERERGRvaYik4iIiIiIiIiI7DUVmUREREREREREZK+pyCSyC2bW3cyyQ3z9c8xsoZnlmNmhYcUREU+ZnQ8zq2dms80sfS+f5yUze6iij93N83Yws6/K+nlFJDYoN+wUj3JDGTCzf5rZ9RX9uiJSsZRDdopHOWT7sRPN7MA9OTaWqMhUhsxsgZltCv5gl5nZUDPLCDuuSEGMJ4Qdh5TYo8CNzrkM59z3YQdTxvoBQ51zmwHMbIKZXRNyTNuY2V+Dv+XC2yYzKzCzrGB7FTMbYmbrzGypmd1WeKxzbjqwxszOCO0NSMxQbpByoNwQEjPbx8xGmtliM3Nm1iJqe7G5wczSzGx48PfmzKx71NP/A7jbzNLK/Y1I3FAOkXKgHBIDgr9lZ2b7Rqx+FOgfVkxlRUWmsneGcy4D6AgcBtxT2icws5QyjyoOXruy28Nz2xz4saxjCZuZVQEuB14LO5biOOf+FiTfjOBv+hFggnNuRbDL/cB++J9RD+AOMzsl4ileB66ryJglpik3SJGUG7aLh9wAFABjgF7FbL+fXeeGL4BLgaXRBzrnlgA/AWeWYbxSOSiHSJGUQ7aLkxwCgJl1BVoXsWkk0MPM9qngkMqUikzlxDm3CBgNtAcws5pm9qKZLTGzRWb2kJklB9uuMLMvzexfZrYK/w8KZnatmc0ys/VmNtPMOgbrG5nZCDNbbmbzzezPha9rZvcHV8n+Fxz3nZkdHGx7FWgGvBdcDbnDzFoEFdSrzew34JNg3zPN7EczWxNUgNtGvMYCM+trZtPNbG3wWjs1SQyu5q0xs/YR6+oFV2Pqm1mWmb0f7LPKzD43s93+TlrQpNLMbjez34NzemXE9h0q1sH5/SLisTOzP5rZnOAcPWhmrc3s6+DK4xvRVxHNt2pZEbz3S6Le46Nm9ltwdelZM6saFeedZrYUGFrEe0kys3vM7NfgvbwS/K5UMbMcIBmYZma/FHMuDjCzscH5m21m50dseymIZ2zwPj81s+YR2482s0nBz3CSmR0dsa2O+er6YjNbbWbvRL1ucef+tOB3dX3we963mB/jEcAa51yJmsaa2ZvmrwivNbPPbOdmpFm7eJ/FnqOSMjMDegMvR6y+DHjQObfaOTcLeB64ImL7BOB48wlPBFBuCPZTblBuiNvc4Jxb5px7BphUzC7F5gbn3Fbn3OPOuS+A/GKOnwCcXpJYJPEohyiHmHJIXOeQ4NgU4CngxuhtQQusKcBJJX2+mOSc062MbsAC4IRguSm+Qvxg8Pgd4DmgOlAfmAhcF2y7AsgDbgJSgKrAecAi/NUKA/bFV52T8L949wJpQCtgHnBy8Fz3A7nAuUAq0BeYD6RGxxg8bgE44JUgtqrA/sAG4MTgOe4A5gJpEc8xEWgE1AFmAdcXc06GAA9HPP4TMCZYHgA8G7xGKnAMYCU4z92D89U/OO40YCNQO9g+AbgmYv8rgC8iHjt8lbgGcCCwBRgXnMuawEzg8qjX+idQBegWnJs2wfbHg+eqA2QC7wEDoo59JDi2ahHv5arg3LYCMoC3gFejYt23mPNQHVgIXBn83nQEVgAHBttfAtYDxwav/0TheQjiXY0vnKQAFwWP6wbbPwD+B9QOznG3Ep77JcAxwXJtoGMxsf8J+CBq3Q4/tyLOU2bwPh4HpkZs29X7LMk5eqgEv3PHAjlARsR7c0CDiH3OBWZEHbcO6BD2Z5Nu4d5QbijqnCg3KDcUFXvc5IbgOAe0iFhXotwQrM8GuhexvifwXdifW7rFzg3lkKLOiXKIckhRscdFDgH+AjxR3M8CeBL4Z9ifPXv1uRV2AJXphv9wzAHWAL8Cz+A/VBvgP2iqRux7ETA+WL4C+C3quT4Ebi7iNY4oYt+78H1PwSeBbyK2JUX9YS6g6CTQKmLd/wFvRD3HIoJ/hoLnuDRi+9+BZ4s5JycA8yIefwlcFiz3B96N/sMqwXnuDmwCUiLW/Q4cGSzv8GFC0UmgS8TjKcCdEY8fAx6PeK08oHrE9jeCc2T4hNA6YttRwPyIY7cC6bt4L+OAP0Y8boNP4ikRsRaXBC4APo9a9xxwX7D8EjAsYlsG/sppU/yH/8SoY78OztU++O4Atffg3P+G7yJWYzc/w7sjYyvq57aLY2sF56VmCd5nSc5RSYpMLwIvRTxuGsSQHrHuRGBB1HGLgGNL8/utW+W7odxQ1DlRblBuKCr2uMkNFF1kKlFuCNYXV2Q6kYi/Dd10QzmkqHOiHKIcUlTsMZ9DguPnRrzOTj8L4GFgSGl+f2Ptpv6xZe9s59zHkSvM7CB8VXaJmRWuTsJXQAtFLoP/BSyqCWNzoJGZrYlYlwx8XtRzOecKzI/W32g3cUe+fiN8Eot8joVA44h9IscS2LiL5/8EqGpmRwTHHAK8HWz7Bz5pfRScl8HOuYG7ibPQSudcXlQMpRkEcVnE8qYiHjeMeLzaObch4vGv+PdbD6gGTIn4uRr+51FouQsGnivGDuc6WE7B/+OwaDfvoTlwRNTvQgrwasTjyN+FHPPNpRsV8bqFr90Y/7u3yjm3upjX3dW574UfJ2CgmU0H+jnnvi7iOVbjrxzslvlm3w/jr77VwycogCxgbbBc3PssyTna3esXXvk7K2J1TnBfA9gcsbw+6vBM/D+FIsoNO1JuUG6I69xQjJLmhl1R3pCiKIfsSDlEOSRec8jjQH/n3Npd7BP3eUBFpoqxEH+lISvqjyeSK+KYogYDW4ivZO+3i9drWrhgvg9yE2BxMa9T1OsvBg6KeA4LnnN3H0o7P6lPIG/gr6wsA953zq0Ptq0HbgduD/rAjjezSc65caV9nSgb8B/OhRoWt2MJ1Taz6hGJoBnwA75Z5CZ808jizk1x57vQYvwHVaFm+Csby4refQcLgU+dcyfuYp/I34UMfDPWxUW8buFrjwmet46Z1XLOleoDzjk3CTjLzFLx/YzfiIwhwnTg1hI+7cX4As8J+KtcNfFJxCL2Ke59luQc7U5PYBX+SggAzrnVZrYEOBgYG6w+mIhBFM2sEb7J+ey9eG2p3JQblBuKo9ywe2Hnhp2UJDeUQFtgWlnGJZWWcohySHGUQ3YvrBxyPNDVzP4ese5rM7vZOfef4HFb4mDw8l3RwN8VwPnZQj4CHjOzGuYHY2ttZt12cdgLQF8z62TevsFgYxOBdeYHe6tqZslm1t7MDos4tpOZ9QwGFbsFn4C+CbYtw/fP3ZU3gNPN7Pjgj/n24Dm+KvWb9/6Db1Z4SbAMgJn9IXhfhh+7Jp/iB8IsjalATzOrZn5KyKvL4DkfMD/18DHAH4A3nXMF+AE9/2Vm9QHMrLGZnVyK5/0vcKuZtQw+vP4G/G8X/yxEeh/Y38x6m1lqcDvMIgZRBE4zs67mBxp8EPjWObcQGBUce7GZpZjZBUA7fJJegh9U8hkzqx0877G7CyY4P5eYWU3nXC7bf6ZFmQjUMrPGUetTzCw94paKr+ZvAVbik/vfini+4t5nSc7R7lwOvOKci07orwD3BOfoAOBafPPYQt2BT5xzW0rxWpJAlBuUG3ZBuWFHMZUbzA9GXDipQxXbcXDiXeYG8wPvFu6fFryfyC813fDnWWSXlEOUQ3ZBOWRHsZRD9sdffDgkuAGcQdASz/yEQZ3YfqEiLqnIVHEuw7dqmImvkg7H900tknPuTXwTvv/gm1m/A9RxzuXjfxEPwQ+4twKfMGpGHP4u/kO3cOC1nsEfJfjB8O4xP+NCkSPzO+dm46fXfSp4/jPwU6duLf3bBufct/jqfyN2/MdpP+BjfPPyr4FnnHMTAMxstJn9dU9eD/gXvq/yMvxsYK/v4fMUWoo/l4uD57reOfdTsO1OfL/ab8xsHf79tCnFcw/BN638DP/z3IwfoHG3gis1JwEXBrEtZfsggIX+A9yHb4nTCZ+Icc6txCez2/EfrncAf3DOrQiO643vu/0Tvk/0LSV8P72BBcG5uB7/e1RU7Fvx/3RHb/83/upN4W0o/h/2X/FXumay/R+aSMW9z5Kco2IFSeq4IIZo9+GbnP8KfAr8wzk3JmL7JfiBJ0V2RblBuaEoyg07iqncEMRQ2DXup+Bxod3lhtnB/o3x4+NsIrjyb37K6nb4v2uRklAOUQ4pinLIjmImhzjnfnfOLS28BatXOOcK88iZwATn3OJiniIu2M4X5yWemdn9+MHDivzjk8RhZi8B2c65e8KOpShmVg/f1//QiA/WSsH8OAmDnXNHhR2LCCg3yHbKDbHLzB4DfnHOPRN2LCKRlEOkkHJI+TKzb4GrnXM/hB3L3tCYTCISCufccuCAsOMoD865GfiZQEREpBQqc27YHefc7WHHICISz+I9hzjnjgg7hrKg7nIiIiIiIiIiIrLX1F1ORERERERERET2mloyiYiIiIiIiIjIXlORSURERERERERE9lqlGfg7KyvLtWjRIuwwRERi0pQpU1Y45+qFHUeYlCdERIqnPKE8ISJSnNLkiEpTZGrRogWTJ08OOwwRkZhkZr+GHUPYlCdERIqnPKE8ISJSnNLkCHWXExERERERERGRvaYik4iIiIiIiIiI7DUVmUREREREREREZK9VmjGZipKbm0t2djabN28OO5S4kZ6eTpMmTUhNTQ07FBGRcqc8UTrKESKSaJQnSkd5QkQqdZEpOzubzMxMWrRogZmFHU7Mc86xcuVKsrOzadmyZdjhiIiUO+WJklOOEJFEpDxRcsoTIgKVvLvc5s2bqVu3rhJCCZkZdevW1ZUaEUkYyhMlpxwhIolIeaLklCdEBCp5kQlQQiglnS8RSTT63Cs5nSsRSUT67Cs5nSsRqfRFpjCtXLmSQw45hEMOOYSGDRvSuHHjbY+3bt1a4ucZMmQIS5cu3fb4yiuvZPbs2eURsoiIVCDlCRER2RXlCRGJN5V6TKaw1a1bl6lTpwJw//33k5GRQd++fUv9PEOGDKFjx440bNgQgKFDh5ZpnCIiEg7lCRER2RXlCRGJNyoyheTll1/m6aefZuvWrRx99NEMGjSIgoICrrzySqZOnYpzjj59+tCgQQOmTp3KBRdcQNWqVZk4cSLHHXccgwYNon379mRlZXH99dczevRoqlWrxrvvvkv9+vWZM2cOl156Kc45Tj75ZJ566inWrFkT9tsWEeegIA/ytkD+Vsjb7JfztkD+FsgL1u2wvNVvb9geGncK+x1IBVGeEJFiFeTDpjWwcSVsWuXvN66EpkdCvf3Djk4qiPKEiGxTkA+5G2HrRsjdENxvhK05EcsbYL+ToGbjcg1FRaYQ/PDDD7z99tt89dVXpKSk0KdPH4YNG0br1q1ZsWIFM2bMAGDNmjXUqlWLp556ikGDBnHIIYfs9Fxr166lW7duDBw4kNtuu40hQ4bQr18/brrpJvr27ct5553HoEGDKvotipSd/DwoyPWFmYJ8cAX+viAPXH6wLrjfYTlv+74uP+L4fCgo2PH4YveNeK3CQk/+lqiiUPTjqMLQtvWb/bb8Lf5598Qxt6vIlCCUJ0QSSEEBbF4DGyOKRdHFo23bgvtNqwG383Od/piKTAlCeUIkDuXnRRWANkTdRxeINpRsn9yN/rtGSVwyXEWmsvLAez8yc/G6Mn3Odo1qcN8ZB5b6uI8//phJkybRuXNnADZt2kTTpk05+eSTmT17NjfffDOnnXYaJ5100m6fq2rVqpx66qkAdOrUic8//xyAb7/9llGjRgFw8cUXc88995Q6TpFyl58H65fA2mxYt8jfRy6vW+T/mY4lSamQkg4paf4+OQ1SqvhbcnBfrU6wHLlPcExylajlKjsfv2056vj0mmG/+0pNeUJ5QmSvFRTAlrU7F4V2KB6t3vlxcRcfkqtAtbrBrQ40PMjfb1tXd8fH1etV7PtNMMoTyhMSZ/JzIed3X7TP3+of52+NWi5iXd6WYvYtarkk++ZuvwhdGslpkFoN0qr7W+FytSyoVQ1Sq0Nate3rU6v5x2kZ25e37RPcV6tbPuc6QsIUmWKJc46rrrqKBx98cKdt06dPZ/To0Tz55JOMGDGCwYMH7/K50tLSti0nJyeTl5dX5vGK7BHnYMPy4gtIaxdBztKd/7GuUtNX12s28a12Mhv6D9ikFEhKBkv295HLlhxsT4pYLlyf5B8Xu29yxPaijk8GS9pe7EnSfAlS/pQnRMpIQYH/ApC7KWhdurmI+83Fr88tbtsWyIt4ztxNQfFolW8RW5Sk1B0LQ/Xb7VwkilyuWsd/adBsXVIE5QlJWM7BlnWwfhnkRNzWL/UFpch1e3ux2pL9///JaZCcGnUftVwlc9fbk1OjikIZRRSIqu9YHEqOz3JNfEa9B/bkCkF5OeGEEzj33HO5+eabycrKYuXKlWzYsIGqVauSnp7OeeedR8uWLbn++usByMzMZP369aV6jcMPP5y3336bXr16MWzYsPJ4G5LoNq/1haK12bAuO2I54j66Wp+SDjUa+yJSq+6+kFSzMdQovG8M6TXCeDciyhMisaygADb8HlykWLj9wsXabP/FIrroU3hf2qvGOzFIrRq0Mk0P7iMfp0N6LUhN90Wh4loZVa3jv4CoYBTXlCdEylF+nv+cz1kWUUD63V+U3mHdsqK7hiVXgYwGkNkA6rSCZkdtf1y1TtBTYBcFoOSgl0FkMSkpueLPQyWQMEWmWHLQQQdx3333ccIJJ1BQUEBqairPPvssycnJXH311TjnMDMeeeQRwE8xes0112wbqK8knnzySXr37s0jjzzCaaedRs2a6mYjpbAlx3djW7e4+FZIW6P+UbFkqNHIF4oad4S2Z0DNptuLRzWb+H+09Q+2yG4pT0jC2bohuFgRVUAqfFzUhYu0TKjV1H+JyGjgCz3bCkHF3UcXjKKOiX6OpBTlLYlJyhMSF5yDLesjWhgFrY22tTqKeLxxJUWONVe1NmQ0hIz60OxIf5/RMPjsr+97PWTU9wV/fV7HBHOuiB9kHOrcubObPHnyDutmzZpF27ZtQ4ooXBs2bKBatWqYGa+99hpvv/02I0aMKNGxiXzeKr38vOBKwJKgiLQE1i/2H+zrFgfrl/omqNGq1/OFosKCUeF94XJmQ1X7Y5iZTXHOdQ47jmhmdgrwBJAMvOCcGxi1vTkwBKgHrAIudc5lB9v+DpwOJAFjgZvdLpKa8sSO9jRPJPI5kz20q1ZIa37z95tW7XiMJUFmo+15Ztut6fbl9Jr6QlGGYjVPVCTliR0pT0iJOOcLRKvmwcpf/P2qwvt5vvdDtOS07UWiwgsFha2OMhpsLypl1PdFfwldaXKEWjJVUpMmTeKWW26hoKCA2rVrM3To0LBDkvLknJ+ZZoeiUREFpJzf2ekKQVIKZO7jb/XbQuvj/HKNRr5wVLOJ/0c/NT2UtyaVl5klA08DJwLZwCQzG+mcmxmx26PAK865l83sOGAA0NvMjga6AB2C/b4AugETKir+eKc8IWWmyFZICyPuF/lZQiMVtkKq2QSadN65gJTZKG7HohCpLJQnZBvnYMOKHYtH2wpK8/2EB4UsCWo1813WDjo/aHHaMKKA1MC3TtJFgkpL2buS6t69O1OnTg07DCkLuZt9U9LiWh2tC9blbdr52Kp1gmLRPn5GmsLlzH2gxj7+n/hqdTWYtYTlcGCuc24egJkNA84CIotM7YBbg+XxwDvBsgPSgTTAgFRgWQXEXGkoTyS4ggLYmhNMj5zjuzNsW87xXaK3bgiWcyLW5+y4buOqXbdCatwZ2p29YxGpVlPNlikSB5QnEkzhpD1FtUZaNX/Hng6WvL2Q1PRwf1+nFdRp7denpBX/OlLpqcgkEiuc8x/iC77wt99n+gJS9D/v4MeJKGxt1Ljjji2PMhv5AlJGQ7U+kljXGFgY8TgbOCJqn2lAL3yXunOATDOr65z72szGA0vwRaZBzrlZFRCzSPicg9XzYc3CnYs+RS5vCIpIhes2QO6Gkr9eajB1cpWMYDacDN+FIa0VVK0V1Qqpqc9JaoUkIhJ7nPM9GwoLSCujCkmRY65aMtRuHhSSjtxeSKrb2n/Wq5AkxdB/ACJhcQ5WzIEFn8OvX8KCL32LJfDjHzXq6K8MZAbFo8KWR5kN1cRUKouifomjx1TqCwwysyuAz4BFQJ6Z7Qu0BZoE+401s2Odc5/t8AJmfYA+AM2aNSvD0EUq0KY1sGgKZE+GRZP9crHTMpsvAlXJ8IWhtAw/q1mNRhHrM3axXN3vX7icVl3j7YmIxJuNq2D5T0ERqbCgFBSTIi8yJKVArebbZ2Or23p7MalWMz/DmkgpqcgkUlGc8x/2hS2Vfv3KD4QK/qpvi67Qogs07wpZ+6mIJIkgG2ga8bgJsDhyB+fcYqAngJllAL2cc2uD4tE3zrmcYNto4Eh8ISry+MHAYPADupbT+xApO/m5sOxHX0zKDm4r5wQbDeodAG1OhSaHQd19IwpCQaEopaq6QIuIJALn/PAZy3+C5T/DitmwPLhtXLF9v6QUqN3CF45adPFd2uq0grqtfIskFZKkjKnIJFJeCgrg9x99C6Vfg6JS4ZXnGo2hdQ9o3sUXl+q0UlFJEtEkYD8za4lvoXQhcHHkDmaWBaxyzhUAd+FnmgP4DbjWzAbgW0R1Ax6vqMBFyoRzfnDsyILSkqmQt9lvr17fD4p98IW+qNToUEivEW7MIiJSsQryYfUCWPHz9iLSitm+sBTZvS295vYLEfXa+OW6raFmM3Vhlgql37ZytHLlSo4//ngAli5dSnJyMvXq1QNg4sSJpKXtvh/rlVdeSb9+/WjTpk2x+zz99NPUqlWLSy65pGwClz1TkA9LZ2zv+vbbV7Bptd9Wsxnsd3LQUqmLv5qgopIkOOdcnpndCHwIJANDnHM/mll/YLJzbiTQHRhgZg7fSulPweHDgeOAGfgudmOcc+9V9HvYW8oTCWbLelj8fdDtbQpkT4KcYLz65CrQ6BDofDU06eQHzK7VTLlCJMEpTySQvC2+e1thAWn5T76wtGIO5G/Zvl9GQ6i3v78AUa+Nv2W18WPlKWdIDDDnKkfvgc6dO7vJkyfvsG7WrFm0bds2pIh2dP/995ORkUHfvn13WO+cwzlHUgw1bY+l8xbT8vNg6bSgpdKX8OvX26fvrN1ye9e3Fl38FwWREJnZFOdc57DjCJPyRNmIpXMW0wry/ReEwnGUsqfA8lngCvz2Oq1966QmnaFxJ2jQXoOoSqiUJ5QnykosnbOYtCUnKB5FtUxaNR9cfrCT+e8P9dpA1v6+VVLhctVaoYYviak0OUItmUIwd+5czj77bLp27cq3337L+++/zwMPPMB3333Hpk2buOCCC7j33nsB6Nq1K4MGDaJ9+/ZkZWVx/fXXM3r0aKpVq8a7775L/fr1ueeee8jKyuKWW26ha9eudO3alU8++YS1a9cydOhQjj76aDZs2MBll13G3LlzadeuHXPmzOGFF17gkEMOCflsxJH8XFg81Xd9W/Al/PbN9iaqdfeFA8/2Xd+ad4GajcONVUTimvJEHFq/NKKgNNm3WNqa47el1/LFpLZn+MJS445QrU648YpIXFOeiAMbV0V0bSssJv0MayMm1k1K8Rcd6reFdmcHxaT9oe5+kFYtvNhF9oKKTCGZOXMmQ4cO5dlnnwVg4MCB1KlTh7y8PHr06MG5555Lu3btdjhm7dq1dOvWjYEDB3LbbbcxZMgQ+vXrt9NzO+eYOHEiI0eOpH///owZM4annnqKhg0bMmLECKZNm0bHjh0r5H3GtbytsPi77QN1L5y4fTaGrDbQ4bztYyplNgw3VhGpdJQnYljuZj92Uvak7V3fCr80JKVAw4Pg4It8YanJYRp3T0TKhfJEjFmbDfM+hfmfwvzP/KDchVKq+ol9mh0J9S733yXqtfH5QQNvSyWTOEWm0f38eDllqeFBcOrAPTq0devWHHbYYdse//e//+XFF18kLy+PxYsXM3PmzJ2SQtWqVTn11FMB6NSpE59//nmRz92zZ89t+yxYsACAL774gjvvvBOAgw8+mAMPPHCP4q70NqyEaf+FOR/CwkmQt8mvr98ODrl4+5hKGfXDjVNEyp7yhPLErjjni0nfvQI/vLW9JWutZr6QdOQNfhylfTpAatVwYxWR8qE8oTwRadNqmP85zJvgC0sr5/r11bKg5bG+1WpWG98yqWYzzfwpCSNxikwxpnr16tuW58yZwxNPPMHEiROpVasWl156KZs3b97pmMiB/ZKTk8nLyyvyuatUqbLTPpVl7K1yUVAACz6HKS/BT+9D/lY/Nkany31BqXkXqF437ChFJMEoT8SIDStg2jD4/jU/plJqNTjwHDjgdF9c0kUHEQmJ8kQFy93kh8soLCotngo4SK3uL0R3uhJadfcXp1VQkgSWOEWmPbxCUBHWrVtHZmYmNWrUYMmSJXz44YeccsopZfoaXbt25Y033uCYY45hxowZzJw5s0yfPy7l/A5TX4cpL8Pq+X7MjM5X++JSfQ1WKJJwlCeUJwoV5MPccfD9qzB7NBTk+lZKZzwBB/aE9BphRyiSMMzsFOAJ/CykLzjnBkZtbw4MAeoBq4BLnXPZwba/A6cDScBY4Ga3N5US5YnEyhP5eb5r9LwJ/rZwop/lLSnFX2To3g9advOTN2jiBpFtEqfIFMM6duxIu3btaN++Pa1ataJLly5l/ho33XQTl112GR06dKBjx460b9+emjVrlvnrxLyCApg33rdamj0KCvJ8S6Uef/UDsqqLg4jEIOWJCrJqvm+xNPWOo74ZAAAgAElEQVQ/sH4xVKsLR1wHh16qiw8iITCzZOBp4EQgG5hkZiOdc5HVjUeBV5xzL5vZccAAoLeZHQ10AToE+30BdAMmVFT8FUl5ogw45wfmnjfBj6204IvtM0c3OAgOv9a3VGp2FFTJCDFQkdhmlaXZY6xPORq2vLw88vLySE9PZ86cOZx00knMmTOHlJSd64yV8rytWwJTX4PvXoU1v0LVOn6MpY6X+37SIpWcpqZWntidkuaJSnfOcjfBrPf8WEsLPgdLgn1P8IWl/U/V1WlJGLGYJ8zsKOB+59zJweO7AJxzAyL2+RE42TmXbWYGrHXO1QiOHQR0BQz4DOjtnJtV3OspT+xapcwTaxf5rm/zPvXFpZylfn2t5r6g1KobtDgWMuqFGKRI+EqTI9SSKUHk5ORw/PHHk5eXh3OO5557rsgCU6VS2N3hu5d9dweX7wfhO+E+OOAPkFIl7AhFRGJGQuUJ53wXiO9ehRnD/ZXq2i3guHvg4IuhZuOwIxQRrzEQMd872cARUftMA3rhu9SdA2SaWV3n3NdmNh5Ygi8yDdpVgUl2r1LkiU2rfQulwtZKK+f49dXq+q5vrbr5+zotQw1TJJ6V66dCTPWhTnC1atViypQpYYdRMdZm++4O370K67Khej04+kbfaqlu67CjExGJSQmRJzaughlv+vywbAakpEPbM6Fjb2jeVQO1isQeK2Jd9PeBvsAgM7sC31ppEZBnZvsCbYEmwX5jzexY59xnO7yAWR+gD0CzZs3KMPTKJy7zROFg3fODlkpLpoEr8IN1Nz8aOl3hC0v1D1QOECkj5VZkUh9qqVD5eTDnI99qac5HPnm0Pg5OfhjanKbuDiIiiaqgAOZP8IWlwhlE9zkETn8M2p8LVWuFHaGIFC8baBrxuAmwOHIH59xioCeAmWUAvZxza4Pi0TfOuZxg22jgSHwhKvL4wcBg8N3lyul9SEVaPBV+GeeLSr99u+Ng3cfe4bvBabBukXJTni2ZDgfmOufmAZjZMOAsILLI1A64NVgeD7wTLDsgHUjDX8FIBZbtSRDOOXz3bCmJuGsstuY3/8Xh+1dh/RLIaABdb4VDe6uZq4iUiPJEycVVjliz0M8g+v3rsPY3qFobOl/lx1pqeFDY0YlIyUwC9jOzlvgWShcCF0fuYGZZwCrnXAFwF76XBMBvwLVmNgD/faIb8PieBKE8UXKh5onsKfBJf19cAmjQ3g/W3bIbND8KqmSGF5tIAinPIlPofajT09NZuXIldevWVWIoAeccK1euJD09PexQdi0/F34e42eImzvOr9vvRDjtUdj/ZEhODTU8EYkfyhMlFxc5Im+Lb630/Wvwy3i/rlV3OPF+aHM6pMZw7CKyE+dcnpndCHyIH35jiHPuRzPrD0x2zo0EugMDzMzhWyn9KTh8OHAcMAN/AXuMc+690sagPFFyoeWJZTNh/MP+879aXTj5b3DQ+RqsWyQk5VlkCr0PdZMmTcjOzmb58uV78z4SSnp6Ok2aNNn9jmFYNd/P/jP1dchZBpmNoNsd/qp0LfWhF5HSU54onZjNEUt/8C1ap//PD+pasyl0uxMOvUT5QSTOOedGAaOi1t0bsTwcX1CKPi4fuG5vX195onQqNE+smgcTBsL0N3wrpR73wJHXq8WSSMjKs8gUeh/q1NRUWrZUl6m4lrcVZn/gWy3Nm+Cnlt7vZD9I374nQHKczWghIjFFeSKObVoDPwz3XaaXTIXkND9zaMfevmtEUnLYEYpIJaA8EYPWLYbP/uEvPielQpeb/a1anbAjExHKt8gUE32oJU6t/MUXlqb+Bzau8Fele9wNh1yiqaVFRBKVc/Drl/6Lxcx3IW+zH3PjlEegw/n6giEiUpltWAlf/gsmPg8F+dDpSji2L2Q2DDsyEYlQbkWmWOhDLXGmoABmjYRJL8CCz8GSoc2pPoG07qGr0iIiiWrLepg2zOeH5T9BlZr+okPH3n6mOI2TIiJSeW1eB988A18NgtwN0OFC6H4n1G4RdmQiUoRy7WsUdh9qiRPOwdyPYdwDsHQG1GoOx9/rv0DoyoSISOL6/SdfWJo2DLauh0aHwlnPwIHnQFq1sKMTEZHylLvJ54DP/wmbVkHbM33PhvoHhB2ZiOyCBrSRcP32rS8u/fqlvxrR83lofy4kJYUdmYiIhCE/z4/FN/F536o1OQ3a94LDroUmncKOTkREylt+rp/M4dO/w/ol0Pp4OO4eaNwx7MhEpARUZJJwLPsRxj0IP4+GjAZw2qPQ8XJISQs7MhERCUPO7zDlZZgyFNYt8mPxHX8fdLwMqmeFHZ2IiJS3gnz4YQSMfxhWL4CmR0CvF6BF17AjE5FSUJFJKtbqBTB+gJ9mukoN3y3uiOshrXrYkYmISEVzDhZ+61stzXwXCnKh9XH+wsP+J2ssPhGRROAczB4FnzwEv8+EhgfBxW/CfidqzD2ROKQik1SMnN/9VKOTh/ovDV3+DF1u0UxAIiKJaOsGmPGmH2tj6Qw/kPfh10LnqyFr37CjExGRijJvAozrD4umQN194dyh0O5sDZ0hEsdUZJLytXktfPkkfPNvP9V0x8ug2x1Qo1HYkYmISEVb+QtMehGmvubzQ4P28IfHocP5atEqIpJIFk6CT/rD/M+gRhM4cxAcfBEk6+upSLzTX7GUj9xNvvvDF/+ETavhwJ5+wL66rcOOTEREKlJBPsz5yOeEX8ZBUgq0O8sP5N3sSHWFEBFJJMt+9N3iZo+C6vXglEeg85WQUiXsyESkjKjIJGUrP89foZ7wCKxfDPue4Mdd2ufgsCMTEZGKtGGlnx1o8ouw5jfI3Ae6/xU6XQ6ZDcOOTkREKtLKX2DCAJgx3I/Letz/+XFZq2SEHZmIlDEVmaRsFBTArHf9lYmVc6HJ4dDrec0GISKSaBZNgYkv+BmC8rdAi2PgxAfhgNMhOTXs6EREpCKtXQSf/R2+e9W3Vup6qx+btWrtsCMTkXKiIpPsHefgl09g3AOwZBrUawsX/hfanKouECIiiSJ3M/z4lu8St/g7SMuAjr3hsGugftuwoxMRkYq2YQV88S+fF1yBzwfH3A6ZDcKOTETKmYpMsueyJ8PH98OCz6FWMzjnOTjoPE05LSKSKFb/CpOHwHevwKZVkLU/nPoPOPhCSK8RdnQiIlLRNq+Fr5/2t9yNcPDFftKf2s3DjkxEKoiKTFJ6v8/y3eJ+eh+qZcGpf4dOV2jAPhGRRFBQAPM+8V3ifh4DlgQHnOYH8m55rFqxiogkqulvwOg7/KQ/7c6GHndDvf3DjkpEKpiKTFJya36D8QNg+jDfFaLHPXDkDRqwT0QkEeRt9a2WJg6GVb/4WYGO7QudroSajcOOTkREwrJpDXxwO/wwHJoe4S9ANzok7KhEJCQqMsnu5SyHzx/zMwRhcOQfoettUL1u2JGJiEhFWPkLjLgaFn/vv0B0vwvanakWrCIiiW7Bl/D2dbBusb8A3fVWSNZXTJFEpk8AKd7mdfD1oO19qg+9FLrdCTWbhB2ZiIhUlGnD/BXqpGQ4/xVod1bYEYmISNjytsKEv8EXj0OdlnD1WGjSKeyoRCQGqMgkO8vdDJNe8K2XNq3yXyh63KM+1SIiiWTzOl9cmvEGNDsaeg6GWk3DjkpERMK2Yg6MuAaWTIVDe8MpAzV8hohsoyKTbJefB9P+CxMGwrpsaNUDjr8XGncMOzIREalI2VNgxFV+LL7uf/VjL2nmUBGRxOYcTBkKH97tu0tf8Bq0PSPsqEQkxqjIJN7vP8GbV8DyWdCoI5z9NLTqHnJQIiJSoQoK4MvHYfzDkLkPXDEKmh8VdlQiIhK2DStg5E0we5S/EH32v6HGPmFHJSIxSEUmgVnvwdvXQ2o1OP9Vf0VCU1CLiCSWdUv84K3zP/VTT5/xOFStHXZUIiIStjkfwzs3wOY1cPIAOOJ6SEoKOyoRiVEqMiWyggI/YN9n/4DGnXyT1xqNwo5KREQq2uwx8O4fIXcTnPEkdLxMFxtERBJd7iYYex9MfA7qt4Peb0PD9mFHJSIxTkWmRLV5LbzVB34e42eNO+0xSE0POyoRSTBmdgrwBJAMvOCcGxi1vTkwBKgHrAIudc5lB9uaAS8ATQEHnOacW1Bx0VcCuZth7L3+C0SDg+DcIZrkQUREYOkMGHGtH0rjiBvghPv1XUFESkRFpkS0fDYMuxhWL4DTHoXDrtEVaxGpcGaWDDwNnAhkA5PMbKRzbmbEbo8CrzjnXjaz44ABQO9g2yvAw865sWaWARRUYPjxb/lsGH4VLPtBXyBERMQrKIBvnoFxD/gu05eOgH1PCDsqEYkjKjIlmp9G+RZMqelw2Uho0SXsiEQkcR0OzHXOzQMws2HAWUBkkakdcGuwPB54J9i3HZDinBsL4JzLqaig455z8N3LMLofpFWDi9+A/U8OOyoREQnbusV+7KV5E6DN6XDmk1A9K+yoRCTOqMiUKAoK4NNH4NOB0OhQP/5SzSZhRyUiia0xsDDicTZwRNQ+04Be+C515wCZZlYX2B9YY2ZvAS2Bj4F+zrn8co86nm1aDSP/DLNG+hlEz3kOMhuGHZWIiIRt5rvw3s2QtwXOeAI6Xq6eDiKyR1RkSgSb1/kZg2aPgoMvhj/8S10iRCQWFPXfq4t63BcYZGZXAJ8Bi4A8fP46BjgU+A34H3AF8OIOL2DWB+gD0KxZs7KLPB79+jWMuAZylsKJ/eGomzQ7kIhIotuSA2PuhO9f8xeie74AWfuGHZWIxDH9d1nZrZgDLxwPP38Ip/4dzn5GBSYRiRXZ+EG7CzUBFkfu4Jxb7Jzr6Zw7FLg7WLc2OPZ759w851wevhtdx+gXcM4Nds51ds51rlevXnm9j9iWnwcTBsJLp0FyKlz9EXS5WQUmEYkLZnaKmc02s7lm1q+I7c3NbJyZTTezCWbWJGJbMzP7yMxmmdlMM2tRkbHHvOzJ8GxX+P51OOZ2uHqsCkwistfUkqkymz3aj7+UnAqXvQstjwk7IhGRSJOA/cysJb6F0oXAxZE7mFkWsMo5VwDchZ9prvDY2mZWzzm3HDgOmFxhkceLNQvhrWvht6/h4IvgtH9AlcywoxIRKRFNEFFO8vPg88f8UBo1GsGVo6D50WFHJSKVhIpMlVFBAXz+KIx/GPY5GC54HWo13f1xIiIVyDmXZ2Y3Ah8CycAQ59yPZtYfmOycGwl0BwaYmcN3l/tTcGy+mfUFxpmZAVOA58N4HzHrx3fgvT/7nNDzeehwftgRiYiUliaIKGurF/iL0Au/hYPO8zNNV60VdlQiUomoyFTZbFkPb18PP70PHS7wA/elVg07KhGRIjnnRgGjotbdG7E8HBhezLFjgQ7lGmA82roRxvTzM8g17gS9XoA6rcKOSkRkT5T7BBEJM3afczBtGIz6ix/Qu+cL0OG8sKMSkUpIRabKZMVcGHYxrJwLJw+AI2/QrBAiIolk6QwYfjWs+Bm63go97vZdpkVE4lO5TxDhnBsMDAbo3Llz9HNXDptWw/u3wo9vQ7Oj4ZxnoXbzsKMSkUpKRabK4ucPYcS1kJQMl70DLY8NOyIREakozsHEwfDR/0HV2j4PtOoedlQiInurRBNEAD0BgnGXejnn1prZtgkigm3vAEcSVWSq9OZ/5ns55CyD4/7PX4BISg47KhGpxFRkinfO+fGXPnkYGrb34y/pyoSISOLYsALe/RP8PAb2PwXOehqqZ4UdlYhIWdAEEXsqbyuMfwi+fNJ3mb56LDTeaRJWEZEypyJTPNuSA+/cALNG+oH7zngS0qqFHZWIiFSUeRPgret8V4hT/wGHX6tu0iJSaWiCiD20fDaMuAaWTodOV8DJf4O06mFHJSIJQkWmeLXyFxh2CayYDSc9BEfdqC8WIiKJIj8XPnkIvnwCsvaHS0f41qwiIpWMJogoBedg8ovw4T1+4p8L/wMHnB52VCKSYFRkikdzPoYRV4ElwaVvQeseYUckIiIVZdU8f4V60ZTgCvUAtWIVEUl0G1f5Hg4/j4HWx8PZz0Bmw7CjEpEEpCJTPHEOvvgXjOsPDQ6EC1+H2i3CjkpERCrK9Dfg/dsgKQnOfwXanRV2RCIiEjbn4K0+MP9TOOUROLyPzxMiIiFQkSlebMnxA7vOfAcO7AlnDVLfahGRRLF1I3xwG0z7r59+uudgqNV098eJiEjlN3kIzB3rx+Y7ok/Y0YhIgivXEreZnWJms81srpn1K2J7czMbZ2bTzWyCmTWJ2NbMzD4ys1lmNtPMWpRnrDFt1Xx48SQ/wPeJ/eHcISowiYgkknH9Ydow6H4XXP6eCkwiIuKt/AU+ugda9YDDrgk7GhGR8mvJZGbJwNPAiUA2MMnMRjrnZkbs9ijwinPuZTM7DhgA9A62vQI87Jwba2YZQEF5xRrT5o6D4Vf55UuGw77HhxuPiIhUrMVTYeJzcNjV0H2n6zUiIpKo8vPg7esgOdWPwaQuciISA8rzk+hwYK5zbp5zbiswDIgePKIdMC5YHl+43czaASnBjBA453KccxvLMdbY45yfNej1c6FGY+gzXgUmEZFEU5AP798C1evB8ffufn8REUkcX/4LsifB6f+EGo3CjkZEBCjfIlNjYGHE4+xgXaRpQK9g+Rwg08zqAvsDa8zsLTP73sz+EbSMSgxbN8CIq2HsvdD2DLj6I6jTKuyoRESkok16ARZ/D6cMgPSaYUcjIiKxYvFUmDDQj9V60LlhRyMisk15FpmsiHUu6nFfoJuZfQ90AxYBefhufMcE2w8DWgFX7PQCZn3MbLKZTV6+fHkZhh6i1QvgxZPhh7fg+PvgvJehSkbYUYmISEVbtxjGPeinoj6wZ9jRiIhIrMjd7LvJVa8Hpz8WdjQiIjsoz9nlsoHIkUmbAIsjd3DOLQZ6AgTjLvVyzq01s2zge+fcvGDbO8CRwItRxw8GBgN07tw5uoAVf34ZD8OvBFfgx1/a74SwIxIRkbCM6QcFuf4LhBV13UZERBLSuP6w/Ce4dARUqxN2NCIiOyjPlkyTgP3MrKWZpQEXAiMjdzCzLDMrjOEuYEjEsbXNrF7w+DggcsDwysU5+GoQvNYTMhrCteNVYBIRSWQ/fwQz34Vj/wJ1WoYdjYiIxIr5n8E3T/uZ5PbV9wURiT3lVmRyzuUBNwIfArOAN5xzP5pZfzM7M9itOzDbzH4GGgAPB8fm47vKjTOzGfiud8+XV6yhm/QCfHQ3HHA6XDMW6rYOOyIREQnL1o0w6naodwAc/eewoxERkVixeS28fQPUaQ0n9g87GhGRIpVndzmcc6OAUVHr7o1YHg4ML+bYsUCH8owvJiyZBh/+FfY9Ec57RVOPiogkuk8fgTW/wZWjISUt7GhERCRWjL4T1i/xkwKlVQ87GhGRIqmiEabN6+DNK6BaFpzznApMIiKJbtlM+HoQHHopND867GhERCRWzBwJ0/4Lx9wOTTqHHY2ISLHKtSWT7IJz8N7NsPpXuOIDqF437IhERCRMBQXw/i2QXhNOfDDsaEREJFasX+a/N+xzCHS7I+xoRER2SUWmsEwZCj++BcffC82PCjsaEREJ2/evwMJv4ex/a7YgERHxnIORN0HuRug5GJJTw45IRGSXVGQKw9IZMLoftD4eutwadjQiIhK2nOUw9j5ocQwcfFHY0YiISKz47mWY8yGcMhDqtQk7GhGR3dIgQBVty3o/DlPV2hqHSUREvI/uhq0b4PR/glnY0YiISCxYNQ/G/BVadoPDrws7GhGRElFLporkHLx/m08Yl78HGfXCjkhERMI2bwJM/x8cewfU2z/saEREJBYU5MPbN0BSCpz9jC5Mi0jcUJGpIn3/Ksx4A3rcAy26hh2NiIiELXezv/hQp5WfMUhERATgyydg4TdwzmCo2STsaERESkxFpoqy7EcY9Rdo1R2OuS3saEREJBZ88S9Y9Qv0fgdS08OORkREYsGS6TD+b9DuLOhwftjRiIiUitpdVoQtOX4cpvSa0PN5SEoOOyIREQnbijnwxT/hoPOgdY+woxERkViQuxnevs7PMvqHxzVOn4jEHbVkqgij+vovE5e9Cxn1w45GRETC5hx8cBukVIWTHg47GhERiRXjH4LfZ8Ilw32hSUQkzqjIVN6+fx2m/Re69YNW3cKORkREYsH0/8H8z/xscpkNwo5GRERiwYIv4KtB0OlK2O/EsKMREdkj6i5Xnn6fBR/cDi2OgW53hB2NiIjEgo2r4MO7oclh/ouEiIjI5nV+NrnaLeCkh8KORkRkj6klU3nZusGPw1QlA3q9qHGYRETE+/g+2LQa/vCupqQWERFvzF2wLhuu+tB/fxARiVMqMpWXUXfA8tnQ+211hRAREe/Xr+G7V+Dom6Bh+7CjERGRWPDTBzD1NTjmdmh6eNjRiIjsFV1CLQ/ThvlEcWxfzRgkIiJe3lZ4/1ao2RS63xV2NCIiEgtylsPIP0PDDn4MVxGROKciU1lbPtt/iWjeRYlCRES2+3oQLJ8Fp/0D0qqHHY2ISFwws1PMbLaZzTWznf65NrPmZjbOzKab2QQzaxK1vYaZLTKzQRUXdQk5B+/dDFvWQ8/BkJIWdkQiIntNRaaytHWjH4cptSr0egGS1RtRRESA1Qvg07/DAX+ANqeGHY2ISFwws2TgaeBUoB1wkZm1i9rtUeAV51wHoD8wIGr7g8Cn5R3rHpn6Osz+AI6/F+q3DTsaEZEyoSJTWRpzJ/w+01+JqNEo7GhERCQWOAcf9PUTQJz697CjERGJJ4cDc51z85xzW4FhwFlR+7QDxgXL4yO3m1knoAHwUQXEWjqrf4XR/fws1Ef+MexoRETKjIpMZWX6m34w1663wb4nhB2NiIjEipnvwNyx0ONuqNk47GhEROJJY2BhxOPsYF2kaUCvYPkcINPM6ppZEvAY8Jdyj7K0CvLhnRvADM5+RjONikilok+0srBiLrx/CzQ7yn+JEBGREqnUY20AbF7rr1Q37ACH9wk7GhGReGNFrHNRj/sC3czse6AbsAjIA/4IjHLOLWQXzKyPmU02s8nLly8vi5h37+un4dcv4dRHoFazinlNEZEKokGD9lbuJj8OU3Ia9HpR4zCJiJRQxFgbJ+KvTk8ys5HOuZkRuxWOtfGymR2HH2ujd8T22B1rA+CThyBnGVz0H+UHEZHSywaaRjxuAiyO3ME5txjoCWBmGUAv59xaMzsKOMbM/ghkAGlmluOc6xd1/GBgMEDnzp2jC1hlb9mP8MmDfoy+gy8q95cTEaloasm0t8bcBctmwDnPqRuEiEjpVN6xNgAWTYGJz8Ph10LjTmFHIyISjyYB+5lZSzNLAy4ERkbuYGZZQdc4gLuAIQDOuUucc82ccy3wrZ1eiS4wVbi8LfDWdZBeE854wneXExGpZFRk2hs/jIApQ6HLzbD/SWFHIyISbyrnWBsA+Xnw3i2Q0QCOuyfsaERE4pJzLg+4EfgQmAW84Zz70cz6m9mZwW7dgdlm9jP+wsPDoQRbEhMG+IvTZz4F1bPCjkZEpFyo7f6eWvkLjLwZmh4Bx/1f2NGIiMSjko61McjMrgA+o4ixNmwXV4LNrA/QB6BZswoc92LiYFg6Hc57yV+xFhGRPeKcGwWMilp3b8TycGD4bp7jJeClcgiv5H77Br58Ag7tDW1ODTUUEZHypCLTnsjd7MdhSkoOxmFKDTsiEZF4VPnG2gBYuwjGPwz7ngjtzq6QlxQRkRi2ZT28fR3UbAqnDAg7GhGRcqUi05746B5/hfqiYVCr6e73FxGRomwbawPfQulC4OLIHcwsC1jlnCsgaqyNiH2uADqHPtZGodF3+OmpT39U422IiAh8eDes/hWuHAVVMsOORkSkXGlMptL68R2Y9DwcdaOauoqI7IVKN9YGwOzR8NP70O0OqN0i7GhERCRss8fAdy9Dlz9D86PDjkZEpNypJVNprJoHI2+Cxp3h+PvCjkZEJO5VmrE2ALZugFF/gXpt4eibwo5GRETCtmGF/+7QoD30uDvsaEREKoSKTCWVtwXevNJ3fTh3CKSkhR2RiIjEkgkDYO1CuOpDjdUnIpLonIP3b4HNa6D325BSJeyIREQqhLrLldTYe2HJVDjrGajdPOxoREQkliz9Ab5+BjpeBs2ODDsaEREJ27RhMOs934KpYfuwoxERqTAqMpXEzJHw7bNwxA3Q9g9hRyMiIrGkoMBfra5aG054IOxoREQkbGsW+kkgmh2t7tMiknDUXW53Vi+Ad2+ERofCif3DjkZERGLNdy9B9iQ45zmoVifsaEREJEwFBfDODeAK4Jx/Q1Jy2BGJiFQoFZl2JW+rH4cJ4NyhGodJRER2lPM7fHw/tDwWOlwQdjQiIhK2b/8NCz6HM5/SLKMikpBUZNqVj++Hxd/B+a9CnZZhRyMiIrHmw79C7iY4/Z9+YggREUlcv8+Cjx+A/U+FQ3uHHY2ISCg0JlNxfvoAvnkaDu8D7c4MOxoREYk1v3wCM96ErrdC1n5hRyMiImHK2wpv9YEqmXDmk7rwICIJSy2ZirLmN9+Xep+D4aSHwo5GRERiTe5m+OB2qNMaut4WdjQiIhK2Tx+BpdPhgtcho37Y0YiIhKbELZnMrKuZXRks1zOzytl/LD8Xhl/lB+07dyikVAk7IhGRuJAweQLg88dg1Tw4/TFITQ87GhGRuFBp88TCifDFP+GQSzQTtYgkvBIVmczsPuBO4K5gVSrwWgmOO8XMZpvZXDPrV8T25mY2zsymm9kEM2sStb2GmS0ys0ElibNMjHvAzxJ05pNQt3WFvayISDzb0zwRl5b/DF/8Cw46H1r3CDsaEZG4UKnzxLfPQo0mcMrAsCMREQldSVsynQOcCWwAcM4tBjJ3dYCZJQNPA6cC7YCLzKxd1G6PAq845zoA/YEBUdsfBD4tYYx77+cP4aunoPPV0L5nhb2siEglUOo8EZecg/dvhbRqcPLDYUcjIhsYi2EAACAASURBVBJPKm+eOOc5uPxdSK8RdiQiIqH7//buPDyu+rr/+PuMRvtqLd4kecM72GAwBkISCFugSSEECJAFSNKSLiQpv9AW2pSkpGnSNmmTNjSFELbQQoxpWtq4AWK2sARss5jFC8ZgS7bBsuVFuzQz5/fHvZJGsizLHo9Gsj6v59Ez937vvTNnrmwd6cx3GWqRqdPdHXAAMyscwjVLgI3uvsndO4EHgIv6nTMfWBFuP5F83MxOAiYAjw4xxtTsrYdffAkmLoCP/u2wvKSIyFHkcPLE6PPq/bD5GTjnrzXnhojIoTl680RWNpTPyHQUIiIjwlCLTEvN7DagzMx+H/g18JODXFMN1CXt14dtyV4FLgm3LwaKzazCzCLA94E/HewFzOxaM1tlZqsaGhqG+FYG0D0PU7wLLrtH82uIiBy6w8kTo0trIzz6dag9BU68OtPRiIiMNkd/nhARkaGtLufu3zOzc4F9wBzgZnd/7CCXDbRup/fbvwH4kZldAzwNbAViwB8By929zgZZ/tPdbwduB1i8eHH/5x66x/8G6l6AS36qeZhERA7DYeaJ0eWxv4L2vfDxf4LIkNfNEBERxkieEBGRgxeZwrmVHnH3c4BDSQT1QG3Sfg2wLfmEcCz2J8PXKQIucfe9ZnYa8CEz+yOgCMgxs2Z332/y8JS99Rg8+wM46RpYcOkRf3oRkaNdCnli9Nj8HLx8H5z+VZhwbKajEREZVcZEnhAREWAIRSZ3j5tZq5mVuvveQ3julcCscGnSrcAVwKeTTzCzSqDR3RMEK03cGb7mZ5LOuQZYnJYCE0BBBcz5mFaDEBE5TCnkidEh1gn/8ydQOgXO+PNMRyMiMuoc9XlCRER6DGm4HNAOvGZmjxGuCAHg7l850AXuHjOz64BHgCzgTnd/w8xuAVa5+8PAmcB3zMwJhsv98eG9jRRUnwhX/sewv6yIyFHmkPPEqPHcP8PO9fDppZBz9MxTKyIyzI7ePCEiIj2GWmT6Zfh1SNx9ObC8X9vNSdvLgGUHeY67gbsP9bVFRGRYHVaeGBVmnQexDpj90UxHIiIymh29eUJERHoMdeLve8wsB5gdNq139670hSUiIqPJUZ0nJi0MvkRE5LAd1XlCRER6DKnIZGZnAvcA7xKsGldrZle7+9PpC01EREYL5QkRERmM8oSIyNgw1OFy3wfOc/f1AGY2G7gfOCldgYmIyKiiPCEiIoNRnhARGQMiQzwvuzshALj7BiA7PSGJiMgopDwhIiKDUZ4QERkDhtqTaZWZ/RT4Wbj/GWB1ekISEZFRSHlCREQGozwhIjIGDLUn0x8CbwBfAb4KvAn8QbqCEhGRUUd5QkREBnNYecLMzjez9Wa20cxuHOD4VDNbYWZrzOxJM6sJ208ws+fN7I3w2OVH+P2IiMgAhtqTKQr80N3/EcDMsoDctEUlIiKjjfKEiIgM5pDzRHjOrcC5QD2w0swedvc3k077HnBvuHrdWcB3gM8BrcBV7v6WmU0GVpvZI+6+54i/MxER6THUnkwrgPyk/Xzg10c+HBERGaWUJ0REZDCHkyeWABvdfZO7dwIPABf1O2d++NwAT3Qfd/cN7v5WuL0N2AFUpfQORETkoIZaZMpz9+bunXC7ID0hiYjIKKQ8ISIigzmcPFEN1CXt14dtyV4FLgm3LwaKzawi+QQzWwLkAG8fRtwiInIIhlpkajGzE7t3zGwx0JaekEREZBRSnhARkcEcTp6wAdq83/4NwBlm9jJwBrAViCW9ziSCycY/7+6J/V7A7FozW2VmqxoaGob2TkRE5ICGOifTnwAPmtk2gh/skwFNniciIt2UJ0REZDCHkyfqgdqk/RpgW/IJ4VC4TwKYWRFwibvvDfdLgF8CX3f33w70Au5+O3A7wOLFi/sXsERE5BAN2pPJzE42s4nuvhKYC/yc4JOBXwHvDEN8IiIygilPiIjIYFLMEyuBWWY23cxygCuAh/s9f6WZdf9NcxNwZ9ieA/yCYFLwB4/YGxIRkUEdbLjcbUBnuH0a8BcEKzzsJqz4i4jImKY8ISIigznsPOHuMeA64BFgLbDU3d8ws1vM7MLwtDOB9Wa2AZgAfDts/xTwYeAaM3sl/DrhyL0tEREZyMGGy2W5e2O4fTlwu7s/BDxkZq+kNzQRERkFlCdERGQwKeUJd18OLO/XdnPS9jJg2QDX3Qfcl0rgIiJy6A7WkynLzLoLUWcDjycdG+p8TiIicvRSnhARkcEoT4iIjCEH+8F+P/CUme0kWP3hNwBmNhPYm+bYRERk5FOeEBGRwShPiIiMIYMWmdz922a2ApgEPOru3SsuRIAvpzs4EREZ2ZQnRERkMMoTIiJjy0G7qA603Ke7b0hPOCIiMtooT4iIyGCUJ0RExo6DzckkIiKSNmZ2vpmtN7ONZnbjAMenmtkKM1tjZk+aWU3YfoKZPW9mb4THLh/+6EVEREREJJmKTCIikhFmlkWwjPUFwHzgSjOb3++07wH3uvtC4BbgO2F7K3CVux8LnA/8wMzKhidyEREREREZiIpMIiKSKUuAje6+yd07gQeAi/qdMx9YEW4/0X3c3Te4+1vh9jZgB1A1LFGLiIiIiMiAVGQSEZFMqQbqkvbrw7ZkrwKXhNsXA8VmVpF8gpktAXKAt9MUp4iIiIiIDIGKTCIikik2QJv3278BOMPMXgbOALYCsZ4nMJsE/Az4vLsn9nsBs2vNbJWZrWpoaDhykYuIiIiIyH5UZBIRkUypB2qT9muAbcknuPs2d/+kuy8C/jJs2wtgZiXAL4GvD7RyUXju7e6+2N0XV1VpNJ2IiIiISDqpyCQiIpmyEphlZtPNLAe4Ang4+QQzqzSz7lx1E3Bn2J4D/IJgUvAHhzFmERERERE5ABWZREQkI9w9BlwHPAKsBZa6+xtmdouZXRiediaw3sw2ABOAb4ftnwI+DFxjZq+EXycM7zsQEREREZFk0UwHICIiY5e7LweW92u7OWl7GbBsgOvuA+5Le4AiIiIiIjJk6skkIiIiIiIiIiIpU5FJRERERERERERSpiKTiIiIiIiIiIikTEUmERERERERERFJmYpMIiIiIiIiIiKSMhWZREREREREREQkZSoyiYiIiIiIiIhIylRkEhERERERERGRlKnIJCIiIiIiIiIiKUtrkcnMzjez9Wa20cxuHOD4VDNbYWZrzOxJM6sJ208ws+fN7I3w2OXpjFNERERERERERFKTtiKTmWUBtwIXAPOBK81sfr/Tvgfc6+4LgVuA74TtrcBV7n4scD7wAzMrS1esIiIiIiIiIiKSmnT2ZFoCbHT3Te7eCTwAXNTvnPnAinD7ie7j7r7B3d8Kt7cBO4CqNMYqIiIiIiIiIiIpSGeRqRqoS9qvD9uSvQpcEm5fDBSbWUXyCWa2BMgB3k5TnCIiIiIiIiIikqJ0FplsgDbvt38DcIaZvQycAWwFYj1PYDYJ+BnweXdP7PcCZtea2SozW9XQ0HDkIhcRERERERERkUOSziJTPVCbtF8DbEs+wd23ufsn3X0R8Jdh214AMysBfgl83d1/O9ALuPvt7r7Y3RdXVWk0nYiIiIjI0eRwFxIKj11tZm+FX1cPb+QiImNTOotMK4FZZjbdzHKAK4CHk08ws0oz647hJuDOsD0H+AXBpOAPpjFGEREREREZgVJZSMjMyoFvAKcQzBX7DTMbN1yxi4iMVWkrMrl7DLgOeARYCyx19zfM7BYzuzA87UxgvZltACYA3w7bPwV8GLjGzF4Jv05IV6wiIiIiIjLiHPZCQsBHgcfcvdHddwOPEaxaLSIiaRRN55O7+3Jgeb+2m5O2lwHLBrjuPuC+dMYmIiIiIiIj2kALCZ3S75zuhYR+SN+FhIayCJGIiBxh6RwuJyIiIiIicrhSWUhoKNdqISERkSNMRSYRERERERmJUllI6KDXhudqISERkSNIRSYRERERERmJDnshIYJ5Yc8zs3HhhN/nhW0iIpJGKjKJiIiIiMiIk8pCQu7eCHyLoFC1ErglbBMRkTRK68TfIiIiIiIih+twFxIKj91Jb88mEREZBurJJCIiIiIiIiIiKVORSUREREREREREUqYik4iIiIiIiIiIpExFJhERERERERERSZmKTCIiIiIiIiIikjIVmUREREREREREJGUqMomIiIiIiIiISMpUZBIRERERERERkZSpyCQiIiIiIiIiIilTkUlERERERERERFKmIpOIiIiIiIiIiKRMRSYREREREREREUmZikwiIiIiIiIiIpIyFZlERCRjzOx8M1tvZhvN7MYBjk81sxVmtsbMnjSzmqRjV5vZW+HX1cMbuYiIiIiI9Kcik4iIZISZZQG3AhcA84ErzWx+v9O+B9zr7guBW4DvhNeWA98ATgGWAN8ws3HDFbuIiIiIiOxPRSYREcmUJcBGd9/k7p3AA8BF/c6ZD6wIt59IOv5R4DF3b3T33cBjwPnDELOIiIiIiByAikwiIpIp1UBd0n592JbsVeCScPtioNjMKoZ4rYiIiIiIDCMVmUREJFNsgDbvt38DcIaZvQycAWwFYkO8FjO71sxWmdmqhoaGVOMVEREREZFBqMgkIiKZUg/UJu3XANuST3D3be7+SXdfBPxl2LZ3KNeG597u7ovdfXFVVdWRjl9ERERERJKoyCQiIpmyEphlZtPNLAe4Ang4+QQzqzSz7lx1E3BnuP0IcJ6ZjQsn/D4vbBMRERERkQxRkUlERDLC3WPAdQTFobXAUnd/w8xuMbMLw9POBNab2QZgAvDt8NpG4FsEhaqVwC1hm4iIiIiIZEg00wGIiMjY5e7LgeX92m5O2l4GLDvAtXfS27NJREREREQyTD2ZREREREREREQkZSoyiYiIiIiIiIhIylRkEhERERERERGRlKnIJCIiIiIiIiIiKdPE3yJyVHN3YgmnM5agK56gM56gK560HwvbYmF7PE5nzOmK9x7vSjjFuVFKC7Ipy8+mrCCHsvxsSvKzyYpYpt+iiIiIiIjIiKAik4jg7rR2xmlqj9HU3sW+9hjNHTFaO2IkHBLuvV+JYN972pOOJ3r39zve51jvdrz73AGOx8PrguJPdyEoKAp1F4n6FIwGaO+MJ9J670ryokHRqSCb0qQC1ED7QVsOpfnZ5ETVkVREpFtXPEFTe4y9bV3sa+tib/i1rz18bIv17Hcf735s70pQmJtFUW6UorwoxbnZ4WOw39vevZ9NUW6U4rzgq/t4bjQr07dBBmBm5wM/BLKAO9z9u/2OTwHuAcrCc2509+Vmlg3cAZxI8DfPve7+nWENXkRkDFKRSQSIJ5xte9p4u6GZd3a20NIRIy87i9xohNzwMa/fY240i7zs4HhetPcxmjW8xYNEwmnqCIpCTe1dPYWipvYY+8Lt5vZYn/bgWFd4TXBtPOHDFrMZRMyIGJgZWeF2xCw4FrE+x3OyIuREI2RnWfgYfBXmRsNtIyeaFTz2nBt85fS7JicaIScrQnbUyMkKrsmORsjNipAd7b4mQk7Ueq6JRozmjhh72rrY29rFnrZO9rR2sac1+ANnT2sne9qC/frdbexp7WRvWxeD3dLCnCzKCnLCQlRvAaqsp7dU0n5BNmXhdl62/ggSGcvcnd2tXdQ1ttLSESMSMaIRIxIJfpZmRfp9HaCt+7r+banE1dYV71sQOoRiUUtnfNDnz8mKUJKfTUl+lNL8bMYV5DCtopCS/Cj52Vm0dMZpDvNZc3uM+t1tNHf05r/YEHJcTlaktyi1X2EqSnFedm9Rqt/xwtxo8H1Iut8RS/re9Nxjer4nZuoJezBmlgXcCpwL1AMrzexhd38z6bSvA0vd/cdmNh9YDkwDLgNy3X2BmRUAb5rZ/e7+7rC+CRGRMUZFJhkzun8x39TQzKadLbyzs4VNYVHp3V2tdMaOTI+XrIj1KTrlHqRYlZedVLCK9h4D9isWdReEkotFzR2xg8YUjVj4iW3vp7c14wooyYv2tidtF+dFKcmLUpATDX9RDoo9kbAg1FsISioM9RSLDIv0FpG6j/ccC59rtBl/iOd3F//6FKXautjb2rsdFKmC/Q3vN4eFq84D/jH0h2cew5+fPzf1NyMiI1prZ4y6xjbqGlup293KlsZW6hrbqN/dGhSXDlKQSUV3USTarxjVv4jV3ZZIeE/RqCs+eCGnKDcoEJXkZ1OSF2VKeQEl+UGvz5K8bErzg2HJwXZ2n2N52ZHDzh3uTkcs0ZMzm9tjNIUFqOaeD2m6c2pve1N7jPf2tdPc0H1N7Ij9rgD05sZ+9zurp1hFn7asfgWsrJ7vCz1tv/+hGZwzf8IRi3EEWAJsdPdNAGb2AHARkFxkcqAk3C4FtiW1F5pZFMgHOoF9wxG0iMhYpiKTHHXau+K8u6uFTQ1BIam7d9Kmhhb2tnX1nJedZUwpL2BGVREfmTOeGVWFTK8sYkZVISV52XTE4rR3Jfo8dsQStHcFjx1dfffbu+J0dCVoj/V97H/OvrYu2rvidPa/NpYYsLiQE42ExaDsnq79VUVFwaepYXtJvwJR9yeu3del8su5HJ5IxCgN/0CaQsGQr+seurgn7CG1N6kgNXdScRojFpHh0hlLsG1PG3W7g+JRXVg8qtvdRn1jK7taOvucn5+dRW15PrXjCjh1RgU14/KpLS+gND+bRCIYdhxP9Pvq15YI56dLJILH/m3xBMQTCeIHaOt9ru62oJhuxn4Foe6ffd29jkrC3DTcPX27mRl52VnkZWdRVZyb0nN1xOJ9ilDdRauWzhixuPcM8x7o/sfD4eZ928Jz97uWnu0DP5/3fB+6n+coVA3UJe3XA6f0O+ebwKNm9mWgEDgnbF9GUJDaDhQA17t7Y1qjFRGR9BaZhjCGeipwJ1AFNAKfdff68NjVBN1fAf7G3e9JZ6wyunQPb9u0s4V3+vRMamHb3jaSf8+aWJLHjKpCPr5wEjOqiphRWcj0ykJqxuUP+gtvTjRCcd4wvJkksXhQlOqIJXB3zRExBpkZhbnB0IvqsvxMhyMihyGRcHY0dYQ9kFr7FJPqG1t5b197n+G00YhRPS4oIp137ARqxhVQW15AbVhMqijM0QcFI0RuNIvcoiwqilIrVsmQDfQPv3817Urgbnf/vpmdBvzMzI4j6AUVByYD44DfmNmvu3tF9byA2bXAtQBTpkw50vGLiIw5aSsyDXEM9fcIJuG7x8zOAr4DfM7MyoFvAIsJEsnq8Nrd6YpXRqbdLZ1s2tnc0yup+/GdXS19uqwX5UaZUVXIydPGMb2yNuyVFHwV5o6eDnvRrGBOp0L97ioiMmK5O3tauw7YE6l+T9t+w6omlORSO66AU2ZUUDsun5ryAmrHFVBbns+k0nytVCkysHqgNmm/ht7hcN2+CJwP4O7Pm1keUAl8GviVu3cBO8zsWYK/LfoUmdz9duB2gMWLFx+V3cFERIZTOv/6HsoY6vnA9eH2E8B/hdsfBR7r7tJqZo8RJI/70xivZJC7s2lnC0+s28Ha7U28szPonbSntXd4WzRiTKkoYEZlEWfMqWJ6ZWHQK6mqkKqiXH3KKyIiR0wsnmD73nY272plc2MLWxpb2bKrlc27goJSU7/58MoKsqkdV8DcScWcO39CWEQKeiJVl+Vr0n6Rw7MSmGVm04GtwBUExaNkW4CzgbvNbB6QBzSE7WeZ2X0Ew+VOBX4wXIGLiIxV6SwyDWUM9avAJQRD6i4Gis2s4gDXVqcvVMmEzliCF99pZMW693l83Q4272oFgk97p1cW8jsLJjGjsrBnrqTagwxvExERORQtHTE27wom1t7S2JK03crW3W195snLyYpQU57PlPICFk8bx5Ty7iFtBdSU51OSl53BdyJydHL3mJldBzxCMP3Gne7+hpndAqxy94eBrwE/MbPrCUZAXOPubma3AncBrxMMu7vL3ddk5p2IiIwd6SwyDWUM9Q3Aj8zsGuBpgk8oYkO8VmOoR6EdTe08ua6Bx9ft4DdvNdDSGScnGuH0Yyr4vQ9O5yNzx1MzbuiTJIuIiByIu9PQ1MHm7l5Ija1s2dXC5nCupJ3NfSfYLs3PZmpFAQuqS/nYgklMrShgSnkhUyoKmFiSpyFtIhng7suB5f3abk7afhM4fYDrmoHL0h6giIj0kc4i00HHULv7NuCTAGZWBFzi7nvNrB44s9+1T/Z/AY2hHvkSCeeNbftYse59nli3g1fr9wIwqTSPixZVc/bc8XzgmEryczSMQEREDl1HLM7W3W09haQtja1hj6RgiFt7V+/cSBGDSaVBb6Rz5k1gSkUBU8oLmFpeyJTyAkoL1BtJREREJBXpLDIddAy1mVUCje6eAG4iWGkOgi6xf2tm48L988LjMgq0dMR4ZuNOHl+7g8fX76ChqQMzWFRbxg3nzeasuROYN6lYcyiJiMiQtHfF2byrtWe+vs07e4e19V9RND87KygcVRTyoVlVYW+k4KtmXAE5UQ27FhEREUmXtBWZhjiG+kzgO2bmBMPl/ji8ttHMvkVQqAK4pXsScBmZtuxq7Zlb6YVNjXTGExTnRfnw7CrOnjueM2ZXablfERE5oHjC2bq7jU07m4NVRHf2rirav5BUWZTLlPJ8lkwv7ykgTa0oYEpFgRaCEBEREcmgtK7tPoQx1MuAZQe49k56ezbJCNMVT7B6824eX7eDx9ftYOOOZgCOqSrk6g9M5ay5E1g8bRzZmqhbRERC7s7O5s6weBQUkzaFxaQtu1rpjPcObSvOjTKjqpCTp41jemUt06uCFUWnVRZSlJvWX19ERERE5DDptzQZssaWTp5cHxSVntrQQFN7jOws49QZFXzmlCmcNXc8UysKMx2miIhkWFN7F+/ubN2vV9I7DS00dcR6zsvJijCtsoBjqgo5Z94EZlQWMr2qkOmVhVQU5qhHkoiIiMgooyKTHJC7s+69Jh5ft4MVa9/n5bo9uENVcS4XHDeRs+ZO4IOzKvWJsojIGNQRi1PX2MqmhqShbeFjQ1NHz3lmUF2Wz/TKQj55YjXTKwuZXlXEjMpCJpfla8U2ERERkaOIqgPSR1tnnOfe3tkzDG773nYAFtaU8pWzZnH2vPEcN7mUiP4oEBEZU7bsauW+Fzaz4f0mNjW0UL+7lUSfeZJymF5ZyEfmVDG9sojplYXMqApWbcvL1gqiIiJydIrFE2zf205duCBF3e5Wtu5uY0JJHgtryji+tpTqsnz1zpUxQ0WmUS6ecNq74rR1xWnv+Ur07Ld1Bsc6ktvCc9qT9ts64+xr7+LlLXvoiCUozMnig7Mquf6c2Zw5p4rxJXmZfqsiIpIB9btb+dHjG1m2up6IGbMmFLGwppRPLKoOhreF8ySV5mdnOlQREZEjzt3Z1dLZU0Sq393Wp6C0bU878aRPXbIixsSSPBqaOnrmGqwozGFhTSkLa8o4obaMhTWlWhRJjloqMo0ArZ0x/uOFLTS2dPYpALV1xmmPdT8maE/eD89LniT1UOREI+RFI+TnZJGXnUV+dvB45ZIpnD1vPEuml5Mb1SfPIiJj1Xt727n1iY08sHILhvHZU6fyR2ceow8dRETkqNPaGaOusW/xqK6xNWjb3UprZ7zP+ZVFOdSWF7CodhwXHV9AbXk+teMKqC0vYFJpHtGsCB2xOOvfa+LVuj28Wr+XNfV7eHJDQ89qqdVl+T0Fp4U1ZSyoKdU0JHJU0L/iDNvd0snn717JK3V7yM4y8qJZ5OVkkZcd6Sn85GVnUZafTV5Jbp+23uJQb7GoT1ufc/vuaw4MEREZSENTBz9+8m3ue2EziYTzqZNrue4jM5lclp/p0ERERA5L8pC2ut1hIamxLeyZ1MrO5s4+5xfkZDGlPCgafWBmRbA9roApFQXUjMunIOfgf0bnRrNYWFPGwpoyPhe2NXfEeH1rUHDqLjz98rXtQDCH4cyqop4hdgtrypg3qVgf/EtKOmJxVm/ezXMbd/HMxp383SULmTOxOK2vqSJTBm3b08ZVd77IlsZW/u2zJ3H+cRMzHZKIiIxRjS2d3Pb029z73GY64wkuObGaL581i9rygkyHJiIickBd8QS7WzrZ1dJJY0snO5s7eoa0dReUBhrSNrksjynlBZwzbwK1YUGpdlw+U8oLKE/TCqdFuVFOnVHBqTMqetp2NXewZute1tQFRaenNuzgoZfqAcjOMuZNKmFhTSnH15RxfG0Zx1QVqcOAHFAi4by5fR/PbNzJsxt3svLdRtq7EmRFjIU1pTR3dKU9BhWZMuSt95u46s4XaW6Pce8XlvT5QSMiIjJc9rZ28ZPfbOKuZ9+htSvOJ06o5itnz2J6ZWGmQxMRkTGoM5Zgd2snu5qDotGulo6k7U52NXfQ2NK7v7dt4D+aK4tyqBkXDGm78Pj8nt5IyUPaRoKKolw+Mmc8H5kzHgjmgNq2t501dXt4pX4Pa+r28l8vb+O+324BoDAni2OrS3uG2h1fU0bNOE0sPhh3P2rvj7uzeVcrz74dFJWef3sXu1uD/xOzxhdxxclTOH1mJafMKKckb3jmz1SRKQNWb97NF+5eSU40ws+/dBrzJ5dkOiQRERljmtq7uPOZd7njmU00tcf42MJJXH/OLGaOT28XahERGVs6Y4meYlFjS1A8CnoddYQ9jzp7i0bNHexrjw34PBGD8sIcygtzqCjMZd7kEiq694tye7Yri3KYVJpP4Sid38jMqC7Lp7osnwsWTAKC3imbdrYEw+zCOZ7ufu5dOmPB/LzlSROLHx8+VhWPzonFO2MJWjtjtHTGae2I0dwRo7UzTkv4GOzHaOmI95zX0rH/fmtH+BydMbIixuwJxcybWMLcScXMm1TC3InFlBXkZPrtHpaGpg6ee3tnzxC4rXvaAJhYksdZcyfwwVkVfOCYSiZkaB7N0fk/bxR7Yt0O/vDfVzOxJI97v3AKUyo0DEFERIZPS0eMe55/l9uf3sSe1i7Omz+B68+dzbxJ+sBDLga3OAAAGj5JREFUREQOTzzhPL2hgcfWvs+OfR09BaRdLZ00HaBolBUxxhUERaHywhyODYtGFUW5YSGpbwGpND+byBgdJhaJGDPHFzFzfBGfPLEGCIoxG95v4pW6Payp38Oa+r08veEtEkkTi08szSNiEDEjK2JEzIhEjKywLdg2IpHec7LMMDOyIsH3yCxoC7ZJ2g7PCZ+nz2sYPdvxhNPSGQsKQWHhqLv409wR7Ld2xnvO6Yr7IHeir4KcLApyohTlBo+FucFcxtVleRTmRCnMjVKQk0VHLMH695r49dr3+fmqup7rJ5Xm9RSc5k0qYd6kYqZVFI6YXm7dWjpivPhOY88QuHXvNQFQnBflA8dU8KUzZnD6zEpmVBaOiB5bKjINo/98qZ4/XbaGeZOKueuaJaO2uiwiIqNPe1ec+367mR8/+Ta7Wjr5yJwq/t+5c1hQU5rp0EREZJTavKuFpavqeGj1Vt7b105xXpTqsnwqinJYMK4sKBoV5lBelLNfAakkb+wWjY6EnGiE46pLOa66FJgKBKvkvb51X8/E4ntaO4knnHjC6YonSHhQEEx48BVPBL2k4uF+z3aC8Hj3ueF14X73Od3X+RDqQvnZWRTmZoWFnyiFOVmU5mczuTSPwtxgvyB8DPajFORm9SkWFSYdz8/OOuR/P+5OQ3MHa7c3sW77PtZu38e695p4ekMDsbA6lxuNMHtCMXMnFjM3LDzNm1jCuMLh6/XUFU/wSt0eng2LSi9v2UMs4eREIyyeOo4//egcTp9ZyYLq0hE5P5eKTMPkJ09v4tvL1/KBYyq47XMnUTxM4yFFRGRs64jFuf+FLdz65Ns0NHXwoVmV/Mk5szlp6rhMhyYiIqNQe1ec/3t9Oz9fWcdvNzUSMThjdhXfvHA+Z82dQE50ZPUCGUsKcqIsmV7Okunlw/q63lOQ6i1Oxd3xBEQiQVwjoRhiZowvzmN8cR5nzK7qae+MJdi4o5l17/UWnp5Yv4MHV9f3nDOxJI+5k4qZOzEsPE0qYXplIdlHoNeTu7P+/SaeeSsoKr34TiMtnXHMYEF1Kb/3oRl8cGYli6eNIy975K82qCJTmrk73/2/ddz29CY+tmAS/3j58VqGUkRE0q4zluDB1XX86PGNbN/bzinTy/nRlYs4RQtNiIjIIXJ3Xt+6j5+v2sJ/v7KNpvYYU8oLuOG82Vx6Ui0TSzMz94uMDGZGNCvzRaTDlRONMH9yyX5zJTc0dfQWnrY38eb2fTy7cWfPkL6crAizJhT1KTzNnVhMRdHBRyzV727tmVPpubd3srO5E4DplYVcfGI1px9TyWnHVIzKeaNUZEqjrniCGx96jYdequdzp07lmxceOyIquCIicvSKxRP858tb+ecVb1G/u40Tp5TxvcuO5wPHVIyIcfoiIjJ67Gnt5L9e3srPV9Wzdvs+cqMRfmfBJD61uJZTppdruJsc1aqKc6kqruJDs/r2etq0s7lP4enptxp46KXeXk/ji3ODoXbhXE9zJxVTUZjLyneDeZWe27iTd3e1AlBZlMvpMyt7vqrL8of9fR5pKjKlSVtnnD/+j5d4fN0Orj9nNl85e6Z+uRcRkbSJJ5z/eXUbP1zxFu/sbGFhTSnf+sRxnDm7SvlHRESGLJFwnn17Jz9fWcejb7xPZzzBguogp1x4/GRK8zXth4xdOdEIcyeWMHdiCSzqbd/Z3MG67U2se28fb4YFqLve3kVnPNHn+sKcLE6dUcFVp03j9JmVzJ5QdNT9nqYiUxrsae3ki/es4qUtu/mbTxzHZ0+dmumQRETkKJVIOMtf384Pfv0WG3c0M3diMbd/7iTOnT/hqPulRURE0mfrnjYeXFXHg6vq2bqnjdL8bD59yhQ+tbh2v2FEItJXZVEuH5yVywdnVfa0dcUTbGpoYd17+3h/XzsnTR3HwpqyIzKP00imItMRtn1vG1f99EU272rlXz99IhcsmJTpkERE5Cjk7jz65vv802MbWPdeE7PGF/GvnzmR84+dqOELIiIyJB2xOI++8T5LV9XxzMadAHxwZiV/fsFczps/YVRMMiwyUmVnRZgzsZg5E4szHcqwUpHpCNq4o5mrfvoC+9pj3P2Fk/nAMZUHv0hEROQQuDtPrm/gHx/bwGtb9zK9spAfXnECH184WfP+iYjIkKzdvo+fr6zjv17Zyp7WLqrL8vnKWbO49KQaassLMh2eiIxiKjIdIS9v2c0X7l5JViTCA9eeynHVpZkOSURkxDOz84EfAlnAHe7+3X7HpwD3AGXhOTe6+3IzywbuAE4kyGX3uvt3hjX4YebuPLtxF99/bD0vb9lDbXk+/3DpQi5eVE30KO92LSIiqdvX3sXDr2xj6ao61tTvJScrwrnHTuDyxbWcPrNSH1SIyBGhItMR8OT6HfzhfS9RVZzLz764hKkVhZkOSURkxDOzLOBW4FygHlhpZg+7+5tJp30dWOruPzaz+cByYBpwGZDr7gvMrAB408zud/d3h/VNDAN355mNO/mXxzfy4juNTC7N428vXsBli2uO+jH9IiKSGnfnhXcaWbqyjuWvb6e9K8HcicXc/PH5XLyomnGFo295dBEZ2VRkStF/vbyVGx58ldkTirn7Cyczvjgv0yGJiIwWS4CN7r4JwMweAC4CkotMDnTPNloKbEtqLzSzKJAPdAL7hiPo4RKLJ/jla9u57alNvLl9HxNKcvnrC4/liiW15EY1R4aIiBzYe3vbeeilepauqmPzrlaKc6NccmINl59cy4LqUi0MISJpoyJTCn76zDt863/f5NQZ5dx+1WJK8rScp4jIIagG6pL264FT+p3zTeBRM/syUAicE7YvIyhIbQcKgOvdvTGt0Q6T1s4YS1fWcccz71C/u41jqgr5+0sXctEJk1VcEhGRA+qKJ1ixdgdLV9Xx5PodJBxOmV7OV8+exQXHTSI/RzlERNJPRabD4O78/SPr+fGTb3P+sRP5wRUnaOUFEZFDN9DHqN5v/0rgbnf/vpmdBvzMzI4j6AUVByYD44DfmNmvu3tF9byA2bXAtQBTpkw50vEfUbuaO7j3+c3c+/y77G7tYvHUcXzjd4/l7LnjtVqciIgc0Pr3mli2uo5fvLyVnc2djC/O5Q/OOIZPLa5lWqWm8RCR4aUi0yGKxRP8xS9eY+mqej59yhS+ddFxmiRPROTw1AO1Sfs19A6H6/ZF4HwAd3/ezPKASuDTwK/cvQvYYWbPAouBPkUmd78duB1g8eLF/QtYI8KWXa3c8cwmlq6qo70rwbnzJ/ClD89g8bTyTIcmIiIj1J7WTh5+dRsPrqrnta17iUaMs+aO5/KTazljdpUWhBCRjFGR6RC0d8W57j9e5tdr3+crZ8/i+nNmaTyziMjhWwnMMrPpwFbgCoLiUbItwNnA3WY2D8gDGsL2s8zsPoLhcqcCPxiuwI+E17fu5d+eepvlr20nK2JcvKiaaz88g5njizMdmojIiHG4q5CGxxYCtxHM7ZcATnb39mEM/4iKxRP85q2dLFtdz2Nvvk9nPMG8SSXc/PH5XHTCZCqKcjMdooiIikxDtbe1i9+7dyWrNu/mlouO5arTpmU6JBGRUc3dY2Z2HfAIwR8Gd7r7G2Z2C7DK3R8Gvgb8xMyuJxhKd427u5ndCtwFvE4w7O4ud1+TmXcydN0rxd321Cae2biT4twov//hGXzh9OlMKNHCESIiyVJZhTRcGOI+4HPu/qqZVQBdw/wWjoiNO5p4cHU9v3hpKzuaOigvzOEzp07h0pNqOHZyaabDExHpQ0WmIXh/XztX/fRFNu1s5l+uXMTHF07OdEgiIkeF8NPm5f3abk7afhM4fYDrmoHL0h7gEdJ/pbjxxbncdMFcrjxlihaNEBE5sFRWIT0PWOPurwK4+65hifgI2dvWxf+8uo1lq+t5pW4PWRHjI3PGc+lJNZw1dzw5UQ2HE5GRSUWmg9jU0Mznfvoie1o7ufvzSzh9ZmWmQxIRkVGitTPGg6vq+clvNvWuFHfJQi5apJXiRESGIJVVSGcDbmaPAFXAA+7+9/1fYCQtEBFPBL1dl62u55E33qMzlmDOhGK+/rF5XHRCNVXFGg4nIiOfikyDeLVuD5+/eyUGPHDtaSyoUXdUERE5uMaWTu557t2eleJO0kpxIiKHI5VVSKPAB4GTgVZghZmtdvcVfZ5sBCwQsamhmWWr6/nPl7by3r52ygqyufLkWi49qZbjqks0B6yIjCoqMh3Ab95q4Es/W015YQ4/++IpTNfynyIichD9V4o7Z94E/uAMrRQnInKYUlmFtB54yt13ApjZcuBEYAUjQFN7F/+7ZjvLVtezevNuIgZnzhnPzb87n7PnjVdvVxEZtVRkGsDDr27ja0tf4ZiqIu75whJNxioiIoN6fetebnt6E79cs00rxYmIHDmprEL6CPBnZlYAdAJnAP80XIEPJJFwnt+0iwdX1fGrN96jvSvBzPFF3HTBXC5eVM14/c0hIkcBFZn6uevZd/jr/3mTJdPL+clViynN14SsIiKyv/4rxRXlRvn9D83g86dPZ2Kp/lAQEUlVKquQArvN7B8JClUOLHf3X2bifWze1dIzHG7rnjZK8qJcelINl55Uy/E1pRoOJyJHFRWZQu7O9x/dwI+e2Mh58yfwz1cuIi9b3VRFRKSvWDzB8tff47an3uaNbcFKcTdeMJdPa6U4EZEj7nBXIQ2P3Qfcl9YAD6C5I8by17azbFU9L77bSMTgQ7OquPGCuZw7f4L+zhCRo5aKTAR/MPzVf7/O/S/WcfniWr598XFEs7QsqIiI9Oq/UtyMqkL+7pIFfGJRtebOEBEREgnnhXcaeXB1Hf/32nu0dcWZUVnIn50/h08uqlEvVxEZE8Z8kam9K85X7n+ZR998n+s+MpOvnTdbXVZFRKRH/5XiTpxSxs0fn8858yZopTgREaGusZWHXqrnoZfqqWtsozg3yicWVXPpSTWcOKVMf1uIyJgy5otM9/12M4+++T7f/N35XHP69EyHIyIiI8y/PfU2tz+9iXPmjedLZxzDyVopTkREkvzDI+v5nzXb+ODMSm44bw7nzZ9Ifo56uIrI2DTmi0yfP306x1WXcuqMikyHIiIiI9DvfWg6l51Uw6wJWilORET29//Onc2fXzCX6rL8TIciIpJxaZ14yMzON7P1ZrbRzG4c4PgUM3vCzF42szVm9jthe7aZ3WNmr5nZWjO7KV0xZkVMBSYRETmg8cV5KjCJiMgBTassVIFJRCSUtiKTmWUBtwIXAPOBK81sfr/Tvg4sdfdFwBXAv4btlwG57r4AOAn4kplNS1esIiIiIiIiIiKSmnT2ZFoCbHT3Te7eCTwAXNTvHAdKwu1SYFtSe6GZRYF8oBPYl8ZYRUREREREREQkBeksMlUDdUn79WFbsm8CnzWzemA58OWwfRnQAmwHtgDfc/fG/i9gZtea2SozW9XQ0HCEwxcRERERERERkaFKZ5FpoLU6vd/+lcDd7l4D/A7wMzOLEPSCigOTgenA18xsxn5P5n67uy9298VVVVVHNnoRERERERERERmydBaZ6oHapP0aeofDdfsisBTA3Z8H8oBK4NPAr9y9y913AM8Ci9MYq4iIiIiIiIiIpCCdRaaVwCwzm25mOQQTez/c75wtwNkAZjaPoMjUELafZYFC4FRgXRpjFRERERERERGRFKStyOTuMeA64BFgLcEqcm+Y2S1mdmF42teA3zezV4H7gWvc3QlWpSsCXicoVt3l7mvSFauIiIiIiIiIiKQmms4nd/flBBN6J7fdnLT9JnD6ANc1A5elMzYRERERERERETly0jlcTkRERERERERExggVmUREREREREREJGUWTIE0+plZA7A503GkqBLYmekgRgDdh4DuQ0D3oVcq92Kqu1cdyWBGm6MgT+j/Qi/di4DuQ0D3oZfyRAqUJ44quhcB3YeA7kOvw70XQ84RR02R6WhgZqvcfXGm48g03YeA7kNA96GX7sXYpu9/L92LgO5DQPehl+7F2Kbvfy/di4DuQ0D3oddw3AsNlxMRERERERERkZSpyCQiIiIiIiIiIilTkWlkuT3TAYwQug8B3YeA7kMv3YuxTd//XroXAd2HgO5DL92LsU3f/166FwHdh4DuQ6+03wvNySQiIiIiIiIiIilTTyYREREREREREUmZikwZZma1ZvaEma01szfM7KuZjimTzCzLzF42s//NdCyZZGZlZrbMzNaF/zZOy3RMmWBm14f/L143s/vNLC/TMQ0XM7vTzHaY2etJbeVm9piZvRU+jstkjDI8lCf6Up4IKE8ExmqeUI6QZMoTfSlPKEckU54Y/jyhIlPmxYCvufs84FTgj81sfoZjyqSvAmszHcQI8EPgV+4+FzieMXhPzKwa+Aqw2N2PA7KAKzIb1bC6Gzi/X9uNwAp3nwWsCPfl6Kc80ZfyREB5YmznibtRjpBeyhN9KU8oRwDKE2QoT6jIlGHuvt3dXwq3mwh+AFRnNqrMMLMa4GPAHZmOJZPMrAT4MPBTAHfvdPc9mY0qY6JAvplFgQJgW4bjGTbu/jTQ2K/5IuCecPse4BPDGpRkhPJEL+WJgPJEH2MyTyhHSDLliV7KE8oRA1Ce6DUseUJFphHEzKYBi4AXMhtJxvwA+DMgkelAMmwG0ADcFXb1vcPMCjMd1HBz963A94AtwHZgr7s/mtmoMm6Cu2+H4BdKYHyG45FhpjyhPBFSnkB5YgDKEaI8oTwByhE9lCf2Myx5QkWmEcLMioCHgD9x932Zjme4mdnHgR3uvjrTsYwAUeBE4MfuvghoYQx2eQ/HCF8ETAcmA4Vm9tnMRiWSOcoTyhNJlCdQnhDpT3lCeSKkHBFSnsgMFZlGADPLJkgI/+7u/5npeDLkdOBCM3sXeAA4y8zuy2xIGVMP1Lt79ydQywgSxVhzDvCOuze4exfwn8AHMhxTpr1vZpMAwscdGY5HhonyBKA8kUx5IqA80ZdyxBimPAEoT3RTjuilPNHXsOQJFZkyzMyMYLzsWnf/x0zHkynufpO717j7NILJ2B539zFZZXb394A6M5sTNp0NvJnBkDJlC3CqmRWE/0/OZoxOWpjkYeDqcPtq4L8zGIsME+WJgPJEL+WJHsoTfSlHjFHKEwHliYByRB/KE30NS56IpuNJ5ZCcDnwOeM3MXgnb/sLdl2cwJsm8LwP/bmY5wCbg8xmOZ9i5+wtmtgx4iWDVlJeB2zMb1fAxs/uBM4FKM6sHvgF8F1hqZl8kSJqXZS5CGUbKEzIQ5YkxnCeUI6Qf5Qnpb8znCFCeIEN5wtw9Hc8rIiIiIiIiIiJjiIbLiYiIiIiIiIhIylRkEhERERERERGRlKnIJCIiIiIiIiIiKVORSUREREREREREUqYik4iIiIiIiIiIpExFJhkVzMzN7PtJ+zeY2TeP0HPfbWaXHonnOsjrXGZma83siXS/Vr/XvcbMfjScrykiMtyUJ1J6XeUJETnqKU+k9LrKEzJkKjLJaNEBfNLMKjMdSDIzyzqE078I/JG7fyRd8YiIjGHKEyIiMhjlCZFhoCKTjBYx4Hbg+v4H+n9yYGbN4eOZZvaUmS01sw1m9l0z+4yZvWhmr5nZMUlPc46Z/SY87+Ph9Vlm9g9mttLM1pjZl5Ke9wkz+w/gtQHiuTJ8/tfN7O/CtpuBDwL/Zmb/MMA1f5r0On8dtk0zs3Vmdk/YvszMCsJjZ5vZy+Hr3GlmuWH7yWb2nJm9Gr7P4vAlJpvZr8zsLTP7+6T3d3cY52tmtt+9FREZRZQnlCdERAajPKE8IcMgmukARA7BrcCa7h9qQ3Q8MA9oBDYBd7j7EjP7KvBl4E/C86YBZwDHAE+Y2UzgKmCvu58c/tB91sweDc9fAhzn7u8kv5iZTQb+DjgJ2A08amafcPdbzOws4AZ3X9XvmvOAWeFzGvCwmX0Y2ALMAb7o7s+a2Z3AH1nQVfVu4Gx332Bm9wJ/aGb/CvwcuNzdV5pZCdAWvswJwCKCT3DWm9m/AOOBanc/Loyj7BDuq4jISKQ8oTwhIjIY5QnlCUkz9WSSUcPd9wH3Al85hMtWuvt2d+8A3ga6f6i/RpAIui1194S7v0WQPOYC5wFXmdkrwAtABcEPb4AX+yeE0MnAk+7e4O4x4N+BDx8kxvPCr5eBl8LX7n6dOnd/Nty+j+DTiznAO+6+IWy/J3yNOcB2d18Jwf0KYwBY4e573b0deBOYGr7PGWb2L2Z2PrDvIHGKiIxoyhPKEyIig1GeUJ6Q9FNPJhltfkDwg/OupLYYYcHUzAzISTrWkbSdSNpP0Pffv/d7HSf4FODL7v5I8gEzOxNoOUB8dtB3MPA133H32/q9zrRB4jrQ8/Q/v1vyfYgDUXffbWbHAx8F/hj4FPCFQ4pcRGTkUZ5QnhARGYzyhPKEpJF6Msmo4u6NwFKCSe+6vUvQnRTgIiD7MJ76MjOLhOOqZwDrgUcIuo1mA5jZbDMrPMjzvACcYWaVFkzidyXw1EGueQT4gpkVha9TbWbjw2NTzOy0cPtK4BlgHTAt7IIL8LnwNdYRjJU+OXyeYjM7YCHZgkkPI+7+EPBXwIkHiVNEZMRTnlCeEBEZjPKE8oSkl3oyyWj0feC6pP2fAP9tZi8CKzjwpwKDWU/wg3UC8Afu3m5mdxB0gX0p/ESjAfjEYE/i7tvN7CbgCYJPApa7+38f5JpHzWwe8HzwMjQDnyX4hGAtcLWZ3Qa8Bfw4jO3zwIPhD/2VwL+5e6eZXQ78i5nlE4yfPmeQl64G7jKz7mLzTYPFKSIyiihPKE+IiAxGeUJ5QtLE3A/UG05EMins3vq/3RPpiYiIJFOeEBGRwShPSCZouJyIiIiIiIiIiKRMPZlERERERERERCRl6skkIiIiIiIiIiIpU5FJRERERERERERSpiKTiIiIiIiIiIikTEUmERERERERERFJmYpMIiIiIiIiIiKSMhWZREREREREREQkZf8fgsvfR1uMylkAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 1440x360 with 3 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(20, 5))\n",
    "\n",
    "for i in range(2,5):\n",
    "    currentLabel = eval('y'+str(i))\n",
    "    \n",
    "    T = range(1,11,1)\n",
    "    train_score, val_score = validation_curve(Perceptron(), X, currentLabel, 'T', T, cv=10, verbose=10, n_jobs=-1)\n",
    "    \n",
    "    param_grid = [{'T': T}]\n",
    "\n",
    "    svc = Perceptron()\n",
    "    clf = GridSearchCV(svc, param_grid, cv = 10, verbose=10, n_jobs=-1)\n",
    "\n",
    "    clf.fit(X, currentLabel)\n",
    "\n",
    "    params_perceptron.append([clf.best_params_, clf.best_score_])\n",
    "    params_perceptron[i-1]\n",
    "    \n",
    "    plt.subplot(1, 3, i-1)\n",
    "    plt.title('Perceptron vs. number of epochs (Label ' + lbl_idx.get(str(i)) + ')')\n",
    "    plt.plot(T, np.mean(val_score, 1), label='Testing')\n",
    "    plt.plot(T, np.mean(train_score, 1), label='Training')\n",
    "    plt.legend()\n",
    "    plt.xlabel('Number of epochs')\n",
    "    plt.ylabel('Score')\n",
    "\n",
    "plt.savefig('perceptron_validation_curve_other.png')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Results: Perceptron"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now analyze the results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### No resampling"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These are the results for the random split where the test set is a third of the dataset, without resampling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "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>Label</th>\n",
       "      <th>True Positive</th>\n",
       "      <th>False Positive</th>\n",
       "      <th>True Negative</th>\n",
       "      <th>False Negative</th>\n",
       "      <th>Precision</th>\n",
       "      <th>Recall</th>\n",
       "      <th>Accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>33</td>\n",
       "      <td>2521</td>\n",
       "      <td>401</td>\n",
       "      <td>3216</td>\n",
       "      <td>416</td>\n",
       "      <td>0.862765</td>\n",
       "      <td>0.858359</td>\n",
       "      <td>0.875343</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>70</td>\n",
       "      <td>1675</td>\n",
       "      <td>230</td>\n",
       "      <td>4465</td>\n",
       "      <td>184</td>\n",
       "      <td>0.879265</td>\n",
       "      <td>0.901022</td>\n",
       "      <td>0.936832</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>101</td>\n",
       "      <td>1355</td>\n",
       "      <td>167</td>\n",
       "      <td>4779</td>\n",
       "      <td>253</td>\n",
       "      <td>0.890276</td>\n",
       "      <td>0.842662</td>\n",
       "      <td>0.935917</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>966</td>\n",
       "      <td>181</td>\n",
       "      <td>5230</td>\n",
       "      <td>177</td>\n",
       "      <td>0.842197</td>\n",
       "      <td>0.845144</td>\n",
       "      <td>0.945377</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label  True Positive  False Positive  True Negative  False Negative  \\\n",
       "0     33           2521             401           3216             416   \n",
       "1     70           1675             230           4465             184   \n",
       "2    101           1355             167           4779             253   \n",
       "3      4            966             181           5230             177   \n",
       "\n",
       "   Precision    Recall  Accuracy  \n",
       "0   0.862765  0.858359  0.875343  \n",
       "1   0.879265  0.901022  0.936832  \n",
       "2   0.890276  0.842662  0.935917  \n",
       "3   0.842197  0.845144  0.945377  "
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_metrics_perceptron = pd.DataFrame(np.column_stack([highest, TPs_perceptron, FPs_perceptron,\n",
    "                                                      TNs_perceptron, FNs_perceptron, precisions_perceptron,\n",
    "                                                      recalls_perceptron, accs_perceptron]),\n",
    "                                     columns=['Label', 'True Positive', 'False Positive', 'True Negative',\n",
    "                                              'False Negative', 'Precision', 'Recall', 'Accuracy'])\n",
    "\n",
    "df_metrics_perceptron = df_metrics_perceptron.astype({'Label': int, 'True Positive': int,\n",
    "                                                      'False Positive': int, 'True Negative': int,\n",
    "                                                      'False Negative': int})\n",
    "\n",
    "df_metrics_perceptron"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x324 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = df_metrics_perceptron.plot(x='Label', y=['Precision', 'Recall', 'Accuracy'], kind=\"bar\",\n",
    "                                 title='Perceptron metrics', figsize=(6, 4.5)).get_figure()\n",
    "\n",
    "fig.savefig(\"perceptron_all_metrics.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean Precision:  0.8686258273994417\n",
      "Mean Recall:  0.8617967432401756\n",
      "Mean Accuracy:  0.9233674092157461\n"
     ]
    }
   ],
   "source": [
    "print(\"Mean Precision: \", df_metrics_perceptron[\"Precision\"].mean())\n",
    "print(\"Mean Recall: \", df_metrics_perceptron[\"Recall\"].mean())\n",
    "print(\"Mean Accuracy: \", df_metrics_perceptron[\"Accuracy\"].mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Upsampling"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These are the results for the random split where the test set is a third of the dataset, with upsampling."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "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>Label</th>\n",
       "      <th>True Positive</th>\n",
       "      <th>False Positive</th>\n",
       "      <th>True Negative</th>\n",
       "      <th>False Negative</th>\n",
       "      <th>Precision</th>\n",
       "      <th>Recall</th>\n",
       "      <th>Accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>70</td>\n",
       "      <td>1825</td>\n",
       "      <td>1295</td>\n",
       "      <td>3400</td>\n",
       "      <td>34</td>\n",
       "      <td>0.584936</td>\n",
       "      <td>0.981711</td>\n",
       "      <td>0.797223</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>101</td>\n",
       "      <td>1580</td>\n",
       "      <td>1060</td>\n",
       "      <td>3886</td>\n",
       "      <td>28</td>\n",
       "      <td>0.598485</td>\n",
       "      <td>0.982587</td>\n",
       "      <td>0.833995</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4</td>\n",
       "      <td>1117</td>\n",
       "      <td>1099</td>\n",
       "      <td>4312</td>\n",
       "      <td>26</td>\n",
       "      <td>0.504061</td>\n",
       "      <td>0.977253</td>\n",
       "      <td>0.828349</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label  True Positive  False Positive  True Negative  False Negative  \\\n",
       "0     70           1825            1295           3400              34   \n",
       "1    101           1580            1060           3886              28   \n",
       "2      4           1117            1099           4312              26   \n",
       "\n",
       "   Precision    Recall  Accuracy  \n",
       "0   0.584936  0.981711  0.797223  \n",
       "1   0.598485  0.982587  0.833995  \n",
       "2   0.504061  0.977253  0.828349  "
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_metrics_perceptron_resampl = pd.DataFrame(np.column_stack([highest[1:], TPs_resampled_perceptr, FPs_resampled_perceptr,\n",
    "                                                      TNs_resampled_perceptr, FNs_resampled_perceptr, precisions_resampled_perceptr,\n",
    "                                                      recalls_resampled_perceptr, accs_resampled_perceptr]),\n",
    "                                     columns=['Label', 'True Positive', 'False Positive', 'True Negative',\n",
    "                                              'False Negative', 'Precision', 'Recall', 'Accuracy'])\n",
    "\n",
    "df_metrics_perceptron_resampl = df_metrics_perceptron_resampl.astype({'Label': int, 'True Positive': int,\n",
    "                                                      'False Positive': int, 'True Negative': int,\n",
    "                                                      'False Negative': int})\n",
    "\n",
    "df_metrics_perceptron_resampl"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x324 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = df_metrics_perceptron_resampl.plot(x='Label', y=['Precision', 'Recall', 'Accuracy'], kind=\"bar\",\n",
    "                                 title='Perceptron metrics', figsize=(6, 4.5)).get_figure()\n",
    "\n",
    "fig.savefig(\"perceptron_all_metrics_upsample.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean Precision:  0.5624940392539671\n",
      "Mean Recall:  0.9805168350554684\n",
      "Mean Accuracy:  0.8198555589461907\n"
     ]
    }
   ],
   "source": [
    "print(\"Mean Precision: \", df_metrics_perceptron_resampl[\"Precision\"].mean())\n",
    "print(\"Mean Recall: \", df_metrics_perceptron_resampl[\"Recall\"].mean())\n",
    "print(\"Mean Accuracy: \", df_metrics_perceptron_resampl[\"Accuracy\"].mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Undersampling"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These are the results for the random split where the test set is a third of the dataset, with upsampling."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "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>Label</th>\n",
       "      <th>True Positive</th>\n",
       "      <th>False Positive</th>\n",
       "      <th>True Negative</th>\n",
       "      <th>False Negative</th>\n",
       "      <th>Precision</th>\n",
       "      <th>Recall</th>\n",
       "      <th>Accuracy</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>70</td>\n",
       "      <td>1825</td>\n",
       "      <td>1295</td>\n",
       "      <td>3400</td>\n",
       "      <td>34</td>\n",
       "      <td>0.584936</td>\n",
       "      <td>0.981711</td>\n",
       "      <td>0.797223</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>101</td>\n",
       "      <td>1580</td>\n",
       "      <td>1060</td>\n",
       "      <td>3886</td>\n",
       "      <td>28</td>\n",
       "      <td>0.598485</td>\n",
       "      <td>0.982587</td>\n",
       "      <td>0.833995</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>4</td>\n",
       "      <td>1117</td>\n",
       "      <td>1099</td>\n",
       "      <td>4312</td>\n",
       "      <td>26</td>\n",
       "      <td>0.504061</td>\n",
       "      <td>0.977253</td>\n",
       "      <td>0.828349</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label  True Positive  False Positive  True Negative  False Negative  \\\n",
       "0     70           1825            1295           3400              34   \n",
       "1    101           1580            1060           3886              28   \n",
       "2      4           1117            1099           4312              26   \n",
       "\n",
       "   Precision    Recall  Accuracy  \n",
       "0   0.584936  0.981711  0.797223  \n",
       "1   0.598485  0.982587  0.833995  \n",
       "2   0.504061  0.977253  0.828349  "
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_metrics_perceptron_unders = pd.DataFrame(np.column_stack([highest[1:], TPs_unders_perceptr, FPs_unders_perceptr,\n",
    "                                                      TNs_unders_perceptr, FNs_unders_perceptr, precisions_unders_perceptr,\n",
    "                                                      recalls_unders_perceptr, accs_unders_perceptr]),\n",
    "                                     columns=['Label', 'True Positive', 'False Positive', 'True Negative',\n",
    "                                              'False Negative', 'Precision', 'Recall', 'Accuracy'])\n",
    "\n",
    "df_metrics_perceptron_unders = df_metrics_perceptron_unders.astype({'Label': int, 'True Positive': int,\n",
    "                                                      'False Positive': int, 'True Negative': int,\n",
    "                                                      'False Negative': int})\n",
    "\n",
    "df_metrics_perceptron_unders"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x324 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = df_metrics_perceptron_resampl.plot(x='Label', y=['Precision', 'Recall', 'Accuracy'], kind=\"bar\",\n",
    "                                 title='Perceptron metrics', figsize=(6, 4.5)).get_figure()\n",
    "\n",
    "fig.savefig(\"perceptron_all_metrics_resampled.png\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean Precision:  0.5624940392539671\n",
      "Mean Recall:  0.9805168350554684\n",
      "Mean Accuracy:  0.8198555589461907\n"
     ]
    }
   ],
   "source": [
    "print(\"Mean Precision: \", df_metrics_perceptron_resampl[\"Precision\"].mean())\n",
    "print(\"Mean Recall: \", df_metrics_perceptron_resampl[\"Recall\"].mean())\n",
    "print(\"Mean Accuracy: \", df_metrics_perceptron_resampl[\"Accuracy\"].mean())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Best score"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These are the results of the `GridSearchCV`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "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>Label</th>\n",
       "      <th>Epochs</th>\n",
       "      <th>F1 Test score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>33</td>\n",
       "      <td>6</td>\n",
       "      <td>0.901259</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>70</td>\n",
       "      <td>9</td>\n",
       "      <td>0.903416</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>101</td>\n",
       "      <td>8</td>\n",
       "      <td>0.895035</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>5</td>\n",
       "      <td>0.873443</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label  Epochs  F1 Test score\n",
       "0     33       6       0.901259\n",
       "1     70       9       0.903416\n",
       "2    101       8       0.895035\n",
       "3      4       5       0.873443"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Ts_perceptron, scores_perceptron = [], []\n",
    "\n",
    "for i in params_perceptron:\n",
    "    Ts_perceptron.append(i[0]['T'])\n",
    "    scores_perceptron.append(i[1])\n",
    "\n",
    "df_results_perceptron = pd.DataFrame(np.column_stack([highest, Ts_perceptron, scores_perceptron]),\n",
    "                                     columns=['Label', 'Epochs', 'F1 Test score'])\n",
    "\n",
    "df_results_perceptron = df_results_perceptron.astype({\"Label\": int, \"Epochs\": int})\n",
    "df_results_perceptron"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x324 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x324 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = df_results_perceptron.plot(x='Label', y=['F1 Test score'], title='Perceptron F1 score',\n",
    "                                 kind=\"bar\", figsize=(6, 4.5)).get_figure()\n",
    "\n",
    "fig.savefig(\"perceptron_test_scores.png\")\n",
    "\n",
    "if len(df_results_perceptron['Epochs'].unique()) != 1:\n",
    "    fig = df_results_perceptron.plot(x='Label', y=['Epochs'], title='Perceptron epochs',\n",
    "                                     kind=\"bar\", color='g', figsize=(6, 4.5)).get_figure()\n",
    "    \n",
    "    fig.savefig(\"perceptron_epochs.png\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Our final average F1 test score is:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8932882749610566"
      ]
     },
     "execution_count": 68,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_results_perceptron[\"F1 Test score\"].mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Comparison: Pegasos vs. Perceptron"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can now run both algorithms with the best value for their parameters, on the same split, and compare their performances and execution times"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For Pegasos, we now set: `lmbd=0.0001` and `T=19860`. For the Perceptron, we average the number of epochs, so we set `T=7`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training on label 33\n",
      "Training on label 70\n",
      "Training on label 101\n",
      "Training on label 4\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>Label</th>\n",
       "      <th>Training score</th>\n",
       "      <th>Test score</th>\n",
       "      <th>Training time</th>\n",
       "      <th>Test time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>33</td>\n",
       "      <td>0.954629</td>\n",
       "      <td>0.914847</td>\n",
       "      <td>4.209161</td>\n",
       "      <td>0.038344</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>70</td>\n",
       "      <td>0.960900</td>\n",
       "      <td>0.921089</td>\n",
       "      <td>3.822495</td>\n",
       "      <td>0.038160</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>101</td>\n",
       "      <td>0.952308</td>\n",
       "      <td>0.911535</td>\n",
       "      <td>4.120745</td>\n",
       "      <td>0.037800</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>0.940500</td>\n",
       "      <td>0.887369</td>\n",
       "      <td>4.030551</td>\n",
       "      <td>0.037777</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label  Training score  Test score  Training time  Test time\n",
       "0     33        0.954629    0.914847       4.209161   0.038344\n",
       "1     70        0.960900    0.921089       3.822495   0.038160\n",
       "2    101        0.952308    0.911535       4.120745   0.037800\n",
       "3      4        0.940500    0.887369       4.030551   0.037777"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lblidx = { '1':'33', '2': '70', '3': '101', '4': '4' }\n",
    "avg_pegasos,avg_train_pegasos,train_time_pegasos,pred_time_pegasos=[],[],[],[]\n",
    "for i in range(1,5):\n",
    "    print(\"Training on label\",lblidx.get(str(i)))\n",
    "    currentLabel = eval('y'+str(i))\n",
    "    \n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, currentLabel, test_size=0.33,\n",
    "                                                        random_state=42, stratify=currentLabel)\n",
    "    \n",
    "    classifier = Pegasos(lmbd=0.0001,T=19860)\n",
    "    \n",
    "    start_time_train = timeit.default_timer()\n",
    "    classifier.fit(X_train, y_train)\n",
    "    elapsed_train = timeit.default_timer() - start_time_train\n",
    "    avg_train_pegasos.append(classifier.score(X_train, y_train))\n",
    "    \n",
    "    start_time_test = timeit.default_timer()\n",
    "    test_score = classifier.score(X_test, y_test)\n",
    "    elapsed_test = timeit.default_timer() - start_time_test\n",
    "    \n",
    "    avg_pegasos.append(test_score)\n",
    "    train_time_pegasos.append(elapsed_train)\n",
    "    pred_time_pegasos.append(elapsed_test)\n",
    "    \n",
    "df_pegasos=pd.DataFrame(np.column_stack([highest, avg_train_pegasos, avg_pegasos, train_time_pegasos, pred_time_pegasos]),\n",
    "                    columns=[\"Label\", \"Training score\", \"Test score\", \"Training time\", \"Test time\"])\n",
    "df_pegasos = df_pegasos.astype({\"Label\": int})\n",
    "df_pegasos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training on label 33\n",
      "Training on label 70\n",
      "Training on label 101\n",
      "Training on label 4\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>Label</th>\n",
       "      <th>Training score</th>\n",
       "      <th>Test score</th>\n",
       "      <th>Training time</th>\n",
       "      <th>Test time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>33</td>\n",
       "      <td>0.990695</td>\n",
       "      <td>0.897238</td>\n",
       "      <td>9.792032</td>\n",
       "      <td>0.037281</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>70</td>\n",
       "      <td>0.991919</td>\n",
       "      <td>0.910322</td>\n",
       "      <td>9.934725</td>\n",
       "      <td>0.038611</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>101</td>\n",
       "      <td>0.981542</td>\n",
       "      <td>0.886895</td>\n",
       "      <td>9.991619</td>\n",
       "      <td>0.035273</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>0.994836</td>\n",
       "      <td>0.873177</td>\n",
       "      <td>9.132115</td>\n",
       "      <td>0.033235</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label  Training score  Test score  Training time  Test time\n",
       "0     33        0.990695    0.897238       9.792032   0.037281\n",
       "1     70        0.991919    0.910322       9.934725   0.038611\n",
       "2    101        0.981542    0.886895       9.991619   0.035273\n",
       "3      4        0.994836    0.873177       9.132115   0.033235"
      ]
     },
     "execution_count": 70,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lblidx = { '1':'33', '2': '70', '3': '101', '4': '4' }\n",
    "avg_perc,avg_train_perc,train_time_perc,pred_time_perc=[],[],[],[]\n",
    "for i in range(1,5):\n",
    "    print(\"Training on label\",lblidx.get(str(i)))\n",
    "    currentLabel = eval('y'+str(i))\n",
    "    \n",
    "    X_train, X_test, y_train, y_test = train_test_split(X, currentLabel, test_size=0.33,\n",
    "                                                        random_state=42, stratify=currentLabel)\n",
    "    \n",
    "    classifier = Perceptron(T=7)\n",
    "    \n",
    "    start_time_train = timeit.default_timer()\n",
    "    classifier.fit(X_train, y_train)\n",
    "    elapsed_train = timeit.default_timer() - start_time_train\n",
    "    avg_train_perc.append(classifier.score(X_train, y_train))\n",
    "    \n",
    "    start_time_test = timeit.default_timer()\n",
    "    test_score = classifier.score(X_test, y_test)\n",
    "    elapsed_test = timeit.default_timer() - start_time_test\n",
    "    \n",
    "    avg_perc.append(test_score)\n",
    "    train_time_perc.append(elapsed_train)\n",
    "    pred_time_perc.append(elapsed_test)\n",
    "    \n",
    "df_perc=pd.DataFrame(np.column_stack([highest, avg_train_perc, avg_perc, train_time_perc, pred_time_perc]),\n",
    "                    columns=[\"Label\", \"Training score\", \"Test score\", \"Training time\", \"Test time\"])\n",
    "df_perc = df_perc.astype({\"Label\": int})\n",
    "df_perc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "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>Algorithm</th>\n",
       "      <th>Training score</th>\n",
       "      <th>Test score</th>\n",
       "      <th>Training time</th>\n",
       "      <th>Test time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Pegasos</td>\n",
       "      <td>95.21%</td>\n",
       "      <td>90.87%</td>\n",
       "      <td>4.05s</td>\n",
       "      <td>0.04s</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Perceptron</td>\n",
       "      <td>98.97%</td>\n",
       "      <td>89.19%</td>\n",
       "      <td>9.71s</td>\n",
       "      <td>0.04s</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    Algorithm Training score Test score Training time Test time\n",
       "0     Pegasos         95.21%     90.87%         4.05s     0.04s\n",
       "1  Perceptron         98.97%     89.19%         9.71s     0.04s"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_mean=[str(np.round(df_pegasos['Training score'].mean()*100,2))+\"%\", str(np.round(df_perc['Training score'].mean()*100,2))+\"%\"]\n",
    "test_mean=[str(np.round(df_pegasos['Test score'].mean()*100,2))+\"%\", str(np.round(df_perc['Test score'].mean()*100,2))+\"%\"]\n",
    "train_time_mean=[str(np.round(df_pegasos['Training time'].mean(),2))+\"s\", str(np.round(df_perc['Training time'].mean(),2))+\"s\"]\n",
    "test_time_mean=[str(np.round(df_pegasos['Test time'].mean(),2))+\"s\", str(np.round(df_perc['Test time'].mean(),2))+\"s\"]\n",
    "\n",
    "df_comp = pd.DataFrame(np.column_stack([[\"Pegasos\",\"Perceptron\"],train_mean,test_mean,train_time_mean,test_time_mean]),\n",
    "                      columns=[\"Algorithm\", \"Training score\", \"Test score\", \"Training time\", \"Test time\"])\n",
    "df_comp"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Finally, we can use CV to estimate the risk in order to to evaluate performance of our algorithms."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training on label 33\n",
      "Training on label 70\n",
      "Training on label 101\n",
      "Training on label 4\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>Label</th>\n",
       "      <th>Score</th>\n",
       "      <th>CV time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>33</td>\n",
       "      <td>0.913814</td>\n",
       "      <td>32.621747</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>70</td>\n",
       "      <td>0.917984</td>\n",
       "      <td>22.754393</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>101</td>\n",
       "      <td>0.907101</td>\n",
       "      <td>23.060167</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>0.883928</td>\n",
       "      <td>23.336162</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label     Score    CV time\n",
       "0     33  0.913814  32.621747\n",
       "1     70  0.917984  22.754393\n",
       "2    101  0.907101  23.060167\n",
       "3      4  0.883928  23.336162"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lblidx = { '1':'33', '2': '70', '3': '101', '4': '4' }\n",
    "avg_pegasos,train_time_pegasos=[],[]\n",
    "for i in range(1,5):\n",
    "    print(\"Training on label\",lblidx.get(str(i)))\n",
    "    currentLabel = eval('y'+str(i))\n",
    "    \n",
    "    classifier = Pegasos(lmbd=0.0001,T=19860)   \n",
    "    \n",
    "    start_time_train = timeit.default_timer()\n",
    "    scores = cross_val_score(classifier, X, currentLabel, cv=10, n_jobs=-1)\n",
    "    elapsed_time = timeit.default_timer() - start_time_train\n",
    "    avg_pegasos.append(scores.mean())\n",
    "    train_time_pegasos.append(elapsed_time)\n",
    "    \n",
    "df_pegasos=pd.DataFrame(np.column_stack([highest, avg_pegasos, train_time_pegasos]),\n",
    "                    columns=[\"Label\", \"Score\", \"CV time\"])\n",
    "df_pegasos = df_pegasos.astype({\"Label\": int})\n",
    "df_pegasos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training on label 33\n",
      "Training on label 70\n",
      "Training on label 101\n",
      "Training on label 4\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>Label</th>\n",
       "      <th>Score</th>\n",
       "      <th>CV time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>33</td>\n",
       "      <td>0.898497</td>\n",
       "      <td>124.642592</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>70</td>\n",
       "      <td>0.902933</td>\n",
       "      <td>120.251307</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>101</td>\n",
       "      <td>0.894081</td>\n",
       "      <td>119.448373</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>0.869438</td>\n",
       "      <td>117.703586</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Label     Score     CV time\n",
       "0     33  0.898497  124.642592\n",
       "1     70  0.902933  120.251307\n",
       "2    101  0.894081  119.448373\n",
       "3      4  0.869438  117.703586"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "lblidx = { '1':'33', '2': '70', '3': '101', '4': '4' }\n",
    "avg_perc,avg_train_perc,train_time_perc,pred_time_perc=[],[],[],[]\n",
    "for i in range(1,5):\n",
    "    print(\"Training on label\",lblidx.get(str(i)))\n",
    "    currentLabel = eval('y'+str(i))\n",
    "    \n",
    "    classifier = Perceptron(T=7)\n",
    "    \n",
    "    start_time_train = timeit.default_timer()\n",
    "    scores = cross_val_score(classifier, X, currentLabel, cv=10, n_jobs=-1)\n",
    "    elapsed_time = timeit.default_timer() - start_time_train\n",
    "    avg_perc.append(scores.mean())\n",
    "    train_time_perc.append(elapsed_time)\n",
    "    \n",
    "df_perc=pd.DataFrame(np.column_stack([highest, avg_perc, train_time_perc]),\n",
    "                    columns=[\"Label\", \"Score\", \"CV time\"])\n",
    "df_perc = df_perc.astype({\"Label\": int})\n",
    "df_perc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "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>Algorithm</th>\n",
       "      <th>Score</th>\n",
       "      <th>CV time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Pegasos</td>\n",
       "      <td>90.57%</td>\n",
       "      <td>25.44s</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Perceptron</td>\n",
       "      <td>89.12%</td>\n",
       "      <td>120.51s</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    Algorithm   Score  CV time\n",
       "0     Pegasos  90.57%   25.44s\n",
       "1  Perceptron  89.12%  120.51s"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "score_mean=[str(np.round(df_pegasos['Score'].mean()*100,2))+\"%\", str(np.round(df_perc['Score'].mean()*100,2))+\"%\"]\n",
    "time_mean=[str(np.round(df_pegasos['CV time'].mean(),2))+\"s\", str(np.round(df_perc['CV time'].mean(),2))+\"s\"]\n",
    "\n",
    "df_comp = pd.DataFrame(np.column_stack([[\"Pegasos\",\"Perceptron\"],score_mean,time_mean]),\n",
    "                      columns=[\"Algorithm\", \"Score\", \"CV time\"])\n",
    "df_comp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}