yeast-multilabel-classifier.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Multi-label Classification"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Import Packages"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from IPython.display import display, HTML, Image\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import copy\n",
    "import random\n",
    "\n",
    "from sklearn.base import BaseEstimator, ClassifierMixin, clone\n",
    "from sklearn.utils.validation import check_X_y, check_array, check_is_fitted\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "\n",
    "from sklearn.tree import DecisionTreeClassifier\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.naive_bayes import GaussianNB\n",
    "from sklearn.neighbors import KNeighborsClassifier\n",
    "from sklearn.svm import SVC\n",
    "\n",
    "from sklearn.metrics import make_scorer\n",
    "from sklearn import metrics\n",
    "\n",
    "# to avoid future warnings for sklearn\n",
    "import warnings\n",
    "warnings.filterwarnings(\"ignore\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Task 0: Load the Yeast Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset.shape: (2417, 117)\n",
      "X.shape: (2417, 103)\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>Att1</th>\n",
       "      <th>Att2</th>\n",
       "      <th>Att3</th>\n",
       "      <th>Att4</th>\n",
       "      <th>Att5</th>\n",
       "      <th>Att6</th>\n",
       "      <th>Att7</th>\n",
       "      <th>Att8</th>\n",
       "      <th>Att9</th>\n",
       "      <th>Att10</th>\n",
       "      <th>...</th>\n",
       "      <th>Att94</th>\n",
       "      <th>Att95</th>\n",
       "      <th>Att96</th>\n",
       "      <th>Att97</th>\n",
       "      <th>Att98</th>\n",
       "      <th>Att99</th>\n",
       "      <th>Att100</th>\n",
       "      <th>Att101</th>\n",
       "      <th>Att102</th>\n",
       "      <th>Att103</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.004168</td>\n",
       "      <td>-0.170975</td>\n",
       "      <td>-0.156748</td>\n",
       "      <td>-0.142151</td>\n",
       "      <td>0.058781</td>\n",
       "      <td>0.026851</td>\n",
       "      <td>0.197719</td>\n",
       "      <td>0.041850</td>\n",
       "      <td>0.066938</td>\n",
       "      <td>-0.056617</td>\n",
       "      <td>...</td>\n",
       "      <td>0.006166</td>\n",
       "      <td>-0.012976</td>\n",
       "      <td>-0.014259</td>\n",
       "      <td>-0.015024</td>\n",
       "      <td>-0.010747</td>\n",
       "      <td>0.000411</td>\n",
       "      <td>-0.032056</td>\n",
       "      <td>-0.018312</td>\n",
       "      <td>0.030126</td>\n",
       "      <td>0.124722</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>-0.103956</td>\n",
       "      <td>0.011879</td>\n",
       "      <td>-0.098986</td>\n",
       "      <td>-0.054501</td>\n",
       "      <td>-0.007970</td>\n",
       "      <td>0.049113</td>\n",
       "      <td>-0.030580</td>\n",
       "      <td>-0.077933</td>\n",
       "      <td>-0.080529</td>\n",
       "      <td>-0.016267</td>\n",
       "      <td>...</td>\n",
       "      <td>0.007680</td>\n",
       "      <td>0.027719</td>\n",
       "      <td>-0.085811</td>\n",
       "      <td>0.111123</td>\n",
       "      <td>0.050541</td>\n",
       "      <td>0.027565</td>\n",
       "      <td>-0.063569</td>\n",
       "      <td>-0.041471</td>\n",
       "      <td>-0.079758</td>\n",
       "      <td>0.017161</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.509949</td>\n",
       "      <td>0.401709</td>\n",
       "      <td>0.293799</td>\n",
       "      <td>0.087714</td>\n",
       "      <td>0.011686</td>\n",
       "      <td>-0.006411</td>\n",
       "      <td>-0.006255</td>\n",
       "      <td>0.013646</td>\n",
       "      <td>-0.040666</td>\n",
       "      <td>-0.024447</td>\n",
       "      <td>...</td>\n",
       "      <td>0.096277</td>\n",
       "      <td>-0.044932</td>\n",
       "      <td>-0.089470</td>\n",
       "      <td>-0.009162</td>\n",
       "      <td>-0.012010</td>\n",
       "      <td>0.308378</td>\n",
       "      <td>-0.028053</td>\n",
       "      <td>0.026710</td>\n",
       "      <td>-0.066565</td>\n",
       "      <td>-0.122352</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.119092</td>\n",
       "      <td>0.004412</td>\n",
       "      <td>-0.002262</td>\n",
       "      <td>0.072254</td>\n",
       "      <td>0.044512</td>\n",
       "      <td>-0.051467</td>\n",
       "      <td>0.074686</td>\n",
       "      <td>-0.007670</td>\n",
       "      <td>0.079438</td>\n",
       "      <td>0.062184</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.083809</td>\n",
       "      <td>0.200354</td>\n",
       "      <td>-0.075716</td>\n",
       "      <td>0.196605</td>\n",
       "      <td>0.152758</td>\n",
       "      <td>-0.028484</td>\n",
       "      <td>-0.074207</td>\n",
       "      <td>-0.089227</td>\n",
       "      <td>-0.049913</td>\n",
       "      <td>-0.043893</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.042037</td>\n",
       "      <td>0.007054</td>\n",
       "      <td>-0.069483</td>\n",
       "      <td>0.081015</td>\n",
       "      <td>-0.048207</td>\n",
       "      <td>0.089446</td>\n",
       "      <td>-0.004947</td>\n",
       "      <td>0.064456</td>\n",
       "      <td>-0.133387</td>\n",
       "      <td>0.068878</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.060467</td>\n",
       "      <td>0.044351</td>\n",
       "      <td>-0.057209</td>\n",
       "      <td>0.028047</td>\n",
       "      <td>0.029661</td>\n",
       "      <td>-0.050026</td>\n",
       "      <td>0.023248</td>\n",
       "      <td>-0.061539</td>\n",
       "      <td>-0.035160</td>\n",
       "      <td>0.067834</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 103 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       Att1      Att2      Att3      Att4      Att5      Att6      Att7  \\\n",
       "0  0.004168 -0.170975 -0.156748 -0.142151  0.058781  0.026851  0.197719   \n",
       "1 -0.103956  0.011879 -0.098986 -0.054501 -0.007970  0.049113 -0.030580   \n",
       "2  0.509949  0.401709  0.293799  0.087714  0.011686 -0.006411 -0.006255   \n",
       "3  0.119092  0.004412 -0.002262  0.072254  0.044512 -0.051467  0.074686   \n",
       "4  0.042037  0.007054 -0.069483  0.081015 -0.048207  0.089446 -0.004947   \n",
       "\n",
       "       Att8      Att9     Att10  ...     Att94     Att95     Att96     Att97  \\\n",
       "0  0.041850  0.066938 -0.056617  ...  0.006166 -0.012976 -0.014259 -0.015024   \n",
       "1 -0.077933 -0.080529 -0.016267  ...  0.007680  0.027719 -0.085811  0.111123   \n",
       "2  0.013646 -0.040666 -0.024447  ...  0.096277 -0.044932 -0.089470 -0.009162   \n",
       "3 -0.007670  0.079438  0.062184  ... -0.083809  0.200354 -0.075716  0.196605   \n",
       "4  0.064456 -0.133387  0.068878  ... -0.060467  0.044351 -0.057209  0.028047   \n",
       "\n",
       "      Att98     Att99    Att100    Att101    Att102    Att103  \n",
       "0 -0.010747  0.000411 -0.032056 -0.018312  0.030126  0.124722  \n",
       "1  0.050541  0.027565 -0.063569 -0.041471 -0.079758  0.017161  \n",
       "2 -0.012010  0.308378 -0.028053  0.026710 -0.066565 -0.122352  \n",
       "3  0.152758 -0.028484 -0.074207 -0.089227 -0.049913 -0.043893  \n",
       "4  0.029661 -0.050026  0.023248 -0.061539 -0.035160  0.067834  \n",
       "\n",
       "[5 rows x 103 columns]"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y.shape: (2417, 14)\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>Class1</th>\n",
       "      <th>Class2</th>\n",
       "      <th>Class3</th>\n",
       "      <th>Class4</th>\n",
       "      <th>Class5</th>\n",
       "      <th>Class6</th>\n",
       "      <th>Class7</th>\n",
       "      <th>Class8</th>\n",
       "      <th>Class9</th>\n",
       "      <th>Class10</th>\n",
       "      <th>Class11</th>\n",
       "      <th>Class12</th>\n",
       "      <th>Class13</th>\n",
       "      <th>Class14</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   Class1  Class2  Class3  Class4  Class5  Class6  Class7  Class8  Class9  \\\n",
       "0       0       0       0       0       0       0       1       1       0   \n",
       "1       0       0       1       1       0       0       0       0       0   \n",
       "2       0       1       1       0       0       0       0       0       0   \n",
       "3       0       0       1       1       0       0       0       0       0   \n",
       "4       0       0       1       1       1       1       0       0       0   \n",
       "\n",
       "   Class10  Class11  Class12  Class13  Class14  \n",
       "0        0        0        1        1        0  \n",
       "1        0        0        0        0        0  \n",
       "2        0        0        1        1        0  \n",
       "3        0        0        0        0        0  \n",
       "4        0        0        0        0        0  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Descriptive stats:\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>Att1</th>\n",
       "      <th>Att2</th>\n",
       "      <th>Att3</th>\n",
       "      <th>Att4</th>\n",
       "      <th>Att5</th>\n",
       "      <th>Att6</th>\n",
       "      <th>Att7</th>\n",
       "      <th>Att8</th>\n",
       "      <th>Att9</th>\n",
       "      <th>Att10</th>\n",
       "      <th>...</th>\n",
       "      <th>Att94</th>\n",
       "      <th>Att95</th>\n",
       "      <th>Att96</th>\n",
       "      <th>Att97</th>\n",
       "      <th>Att98</th>\n",
       "      <th>Att99</th>\n",
       "      <th>Att100</th>\n",
       "      <th>Att101</th>\n",
       "      <th>Att102</th>\n",
       "      <th>Att103</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>count</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "      <td>2417.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>mean</td>\n",
       "      <td>0.001173</td>\n",
       "      <td>-0.000436</td>\n",
       "      <td>-0.000257</td>\n",
       "      <td>0.000265</td>\n",
       "      <td>0.001228</td>\n",
       "      <td>0.000475</td>\n",
       "      <td>0.001107</td>\n",
       "      <td>0.000420</td>\n",
       "      <td>0.001076</td>\n",
       "      <td>-0.000009</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.000773</td>\n",
       "      <td>0.000464</td>\n",
       "      <td>-0.000515</td>\n",
       "      <td>0.000667</td>\n",
       "      <td>0.000324</td>\n",
       "      <td>-0.001483</td>\n",
       "      <td>-0.001047</td>\n",
       "      <td>-0.001539</td>\n",
       "      <td>0.000284</td>\n",
       "      <td>0.007605</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>std</td>\n",
       "      <td>0.097411</td>\n",
       "      <td>0.097885</td>\n",
       "      <td>0.097746</td>\n",
       "      <td>0.096969</td>\n",
       "      <td>0.096909</td>\n",
       "      <td>0.097306</td>\n",
       "      <td>0.097170</td>\n",
       "      <td>0.096803</td>\n",
       "      <td>0.096326</td>\n",
       "      <td>0.096805</td>\n",
       "      <td>...</td>\n",
       "      <td>0.093316</td>\n",
       "      <td>0.096684</td>\n",
       "      <td>0.096209</td>\n",
       "      <td>0.096635</td>\n",
       "      <td>0.096280</td>\n",
       "      <td>0.094369</td>\n",
       "      <td>0.096900</td>\n",
       "      <td>0.094211</td>\n",
       "      <td>0.093154</td>\n",
       "      <td>0.099368</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>min</td>\n",
       "      <td>-0.371146</td>\n",
       "      <td>-0.472632</td>\n",
       "      <td>-0.339195</td>\n",
       "      <td>-0.467945</td>\n",
       "      <td>-0.367044</td>\n",
       "      <td>-0.509447</td>\n",
       "      <td>-0.319928</td>\n",
       "      <td>-0.594498</td>\n",
       "      <td>-0.369712</td>\n",
       "      <td>-0.767128</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.455191</td>\n",
       "      <td>-0.283594</td>\n",
       "      <td>-0.279408</td>\n",
       "      <td>-0.226420</td>\n",
       "      <td>-0.225374</td>\n",
       "      <td>-0.501572</td>\n",
       "      <td>-0.236589</td>\n",
       "      <td>-0.267052</td>\n",
       "      <td>-0.194079</td>\n",
       "      <td>-0.237752</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25%</td>\n",
       "      <td>-0.053655</td>\n",
       "      <td>-0.058734</td>\n",
       "      <td>-0.057526</td>\n",
       "      <td>-0.057149</td>\n",
       "      <td>-0.058461</td>\n",
       "      <td>-0.060212</td>\n",
       "      <td>-0.058445</td>\n",
       "      <td>-0.062849</td>\n",
       "      <td>-0.063472</td>\n",
       "      <td>-0.065010</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.054133</td>\n",
       "      <td>-0.056415</td>\n",
       "      <td>-0.056414</td>\n",
       "      <td>-0.059382</td>\n",
       "      <td>-0.058025</td>\n",
       "      <td>-0.053591</td>\n",
       "      <td>-0.063318</td>\n",
       "      <td>-0.059542</td>\n",
       "      <td>-0.054078</td>\n",
       "      <td>-0.077191</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>50%</td>\n",
       "      <td>0.003649</td>\n",
       "      <td>-0.003513</td>\n",
       "      <td>0.002892</td>\n",
       "      <td>-0.000153</td>\n",
       "      <td>0.005565</td>\n",
       "      <td>0.000321</td>\n",
       "      <td>0.006179</td>\n",
       "      <td>0.001436</td>\n",
       "      <td>0.003515</td>\n",
       "      <td>0.002432</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.012893</td>\n",
       "      <td>-0.023595</td>\n",
       "      <td>-0.024313</td>\n",
       "      <td>-0.023059</td>\n",
       "      <td>-0.021942</td>\n",
       "      <td>-0.018216</td>\n",
       "      <td>-0.033623</td>\n",
       "      <td>-0.023519</td>\n",
       "      <td>-0.012007</td>\n",
       "      <td>0.022126</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>75%</td>\n",
       "      <td>0.057299</td>\n",
       "      <td>0.048047</td>\n",
       "      <td>0.061007</td>\n",
       "      <td>0.054522</td>\n",
       "      <td>0.066286</td>\n",
       "      <td>0.059908</td>\n",
       "      <td>0.068892</td>\n",
       "      <td>0.061418</td>\n",
       "      <td>0.064958</td>\n",
       "      <td>0.063096</td>\n",
       "      <td>...</td>\n",
       "      <td>0.027977</td>\n",
       "      <td>0.034937</td>\n",
       "      <td>0.036057</td>\n",
       "      <td>0.041430</td>\n",
       "      <td>0.035730</td>\n",
       "      <td>0.019583</td>\n",
       "      <td>0.038901</td>\n",
       "      <td>0.025408</td>\n",
       "      <td>0.028087</td>\n",
       "      <td>0.103185</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>max</td>\n",
       "      <td>0.520272</td>\n",
       "      <td>0.614114</td>\n",
       "      <td>0.353241</td>\n",
       "      <td>0.568960</td>\n",
       "      <td>0.307649</td>\n",
       "      <td>0.336971</td>\n",
       "      <td>0.351401</td>\n",
       "      <td>0.454591</td>\n",
       "      <td>0.419852</td>\n",
       "      <td>0.420876</td>\n",
       "      <td>...</td>\n",
       "      <td>0.609175</td>\n",
       "      <td>0.542867</td>\n",
       "      <td>0.547134</td>\n",
       "      <td>0.385928</td>\n",
       "      <td>0.540493</td>\n",
       "      <td>0.569250</td>\n",
       "      <td>0.509963</td>\n",
       "      <td>0.587358</td>\n",
       "      <td>0.700340</td>\n",
       "      <td>0.163431</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>8 rows × 103 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              Att1         Att2         Att3         Att4         Att5  \\\n",
       "count  2417.000000  2417.000000  2417.000000  2417.000000  2417.000000   \n",
       "mean      0.001173    -0.000436    -0.000257     0.000265     0.001228   \n",
       "std       0.097411     0.097885     0.097746     0.096969     0.096909   \n",
       "min      -0.371146    -0.472632    -0.339195    -0.467945    -0.367044   \n",
       "25%      -0.053655    -0.058734    -0.057526    -0.057149    -0.058461   \n",
       "50%       0.003649    -0.003513     0.002892    -0.000153     0.005565   \n",
       "75%       0.057299     0.048047     0.061007     0.054522     0.066286   \n",
       "max       0.520272     0.614114     0.353241     0.568960     0.307649   \n",
       "\n",
       "              Att6         Att7         Att8         Att9        Att10  ...  \\\n",
       "count  2417.000000  2417.000000  2417.000000  2417.000000  2417.000000  ...   \n",
       "mean      0.000475     0.001107     0.000420     0.001076    -0.000009  ...   \n",
       "std       0.097306     0.097170     0.096803     0.096326     0.096805  ...   \n",
       "min      -0.509447    -0.319928    -0.594498    -0.369712    -0.767128  ...   \n",
       "25%      -0.060212    -0.058445    -0.062849    -0.063472    -0.065010  ...   \n",
       "50%       0.000321     0.006179     0.001436     0.003515     0.002432  ...   \n",
       "75%       0.059908     0.068892     0.061418     0.064958     0.063096  ...   \n",
       "max       0.336971     0.351401     0.454591     0.419852     0.420876  ...   \n",
       "\n",
       "             Att94        Att95        Att96        Att97        Att98  \\\n",
       "count  2417.000000  2417.000000  2417.000000  2417.000000  2417.000000   \n",
       "mean     -0.000773     0.000464    -0.000515     0.000667     0.000324   \n",
       "std       0.093316     0.096684     0.096209     0.096635     0.096280   \n",
       "min      -0.455191    -0.283594    -0.279408    -0.226420    -0.225374   \n",
       "25%      -0.054133    -0.056415    -0.056414    -0.059382    -0.058025   \n",
       "50%      -0.012893    -0.023595    -0.024313    -0.023059    -0.021942   \n",
       "75%       0.027977     0.034937     0.036057     0.041430     0.035730   \n",
       "max       0.609175     0.542867     0.547134     0.385928     0.540493   \n",
       "\n",
       "             Att99       Att100       Att101       Att102       Att103  \n",
       "count  2417.000000  2417.000000  2417.000000  2417.000000  2417.000000  \n",
       "mean     -0.001483    -0.001047    -0.001539     0.000284     0.007605  \n",
       "std       0.094369     0.096900     0.094211     0.093154     0.099368  \n",
       "min      -0.501572    -0.236589    -0.267052    -0.194079    -0.237752  \n",
       "25%      -0.053591    -0.063318    -0.059542    -0.054078    -0.077191  \n",
       "50%      -0.018216    -0.033623    -0.023519    -0.012007     0.022126  \n",
       "75%       0.019583     0.038901     0.025408     0.028087     0.103185  \n",
       "max       0.569250     0.509963     0.587358     0.700340     0.163431  \n",
       "\n",
       "[8 rows x 103 columns]"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Read the CSV file\n",
    "dataset = pd.read_csv('data/yeast.csv')\n",
    "print(\"Dataset.shape: \" + str(dataset.shape))\n",
    "\n",
    "# split the features-X and class labels-y\n",
    "X = dataset.iloc[:, :103]\n",
    "y = dataset.iloc[:, 103:]\n",
    "\n",
    "print(\"X.shape: \" + str(X.shape))\n",
    "display(X.head())\n",
    "print(\"y.shape: \" + str(y.shape))\n",
    "display(y.head())\n",
    "print(\"Descriptive stats:\")\n",
    "X.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Split into Train and Test Set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 163,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train.shape: (1691, 103)\n",
      "X_test.shape: (726, 103)\n",
      "y_train.shape: (1691, 14)\n",
      "y_test.shape: (726, 14)\n"
     ]
    }
   ],
   "source": [
    "# Normalise the data\n",
    "X = (X-X.min())/(X.max()-X.min())\n",
    "\n",
    "# split into train and test\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=0, train_size=0.7)\n",
    "\n",
    "print(\"X_train.shape: \" + str(X_train.shape))\n",
    "print(\"X_test.shape: \" + str(X_test.shape))\n",
    "print(\"y_train.shape: \" + str(y_train.shape))\n",
    "print(\"y_test.shape: \" + str(y_test.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Task 1: Implement the Binary Relevance Algorithm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 164,
   "metadata": {},
   "outputs": [],
   "source": [
    "class BinaryRelevanceClassifier(BaseEstimator, ClassifierMixin):\n",
    "    def __init__(self, base_classifier=LogisticRegression()):\n",
    "        self.base_classifier=base_classifier\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        \"\"\"Build a Binary Relevance classifier from the training set (X, y).\n",
    "        Parameters\n",
    "        ----------\n",
    "        X : array-like or sparse matrix, shape = [n_samples, n_features]\n",
    "            The training input samples. Internally, it will be converted to\n",
    "            ``dtype=np.float32`` and if a sparse matrix is provided\n",
    "            to a sparse ``csc_matrix``.\n",
    "        y : array-like, shape = [n_samples, n_labels] \n",
    "            The target values (class labels) as integers or strings.\n",
    "        \"\"\"\n",
    "\n",
    "        # list of individual classifiers\n",
    "        self.models = []\n",
    "       \n",
    "        # for every class label\n",
    "        for label in list(y.columns):\n",
    "            # Check that X and y have correct shape\n",
    "            x_checked, y_checked = check_X_y(X, y[label])\n",
    "            # every classifier is independent of the others\n",
    "            # hence we create a copy of the base classifier instance\n",
    "            base_model = clone(self.base_classifier)\n",
    "            # fit the base model - one model each for Y1, Y2....Y14\n",
    "            basel_model = base_model.fit(x_checked, y_checked)\n",
    "            # add the fitted model list of individual classifiers\n",
    "            self.models.append(base_model)\n",
    "\n",
    "    # The predict function to make a set of predictions for a set of query instances\n",
    "    def predict(self, X):\n",
    "        # check if the models list has been set up\n",
    "        check_is_fitted(self, ['models'])\n",
    "        X = check_array(X)\n",
    "        \n",
    "        all_preds = pd.DataFrame()\n",
    "        i=0\n",
    "        # list of individual classifier predictions\n",
    "        preds = []\n",
    "        \n",
    "        # predict against each fitted model - one model per label\n",
    "        for model in self.models:\n",
    "            pred = model.predict(X)\n",
    "            # add the prediction to the dataframe\n",
    "            preds.append(pd.DataFrame({'Class'+ str(i+1): pred}))\n",
    "            i+=1\n",
    "        \n",
    "        # dataframe with predictions for all class labels\n",
    "        all_preds = pd.concat(preds, axis=1)\n",
    "        # standard sklearn classifiers return predictions as numpy arrays\n",
    "        # hence convert the dataframe to a numpy array\n",
    "        return all_preds.to_numpy()\n",
    "    \n",
    "\n",
    "\n",
    "    def predict_proba(self,X):\n",
    "        # check if the models list has been set up\n",
    "        check_is_fitted(self, ['models'])\n",
    "        X = check_array(X)\n",
    "        \n",
    "        all_preds = pd.DataFrame()\n",
    "        i = 0\n",
    "        \n",
    "        for model in self.models:\n",
    "            # Call predict_proba of the each base model\n",
    "            pred = model.predict_proba(X)\n",
    "            # Add the probabilities of 1 to the dataframe\n",
    "            all_preds['Class'+str(i+1)] = [one_prob[1] for one_prob in pred]\n",
    "            i+=1\n",
    "        \n",
    "        #return probabilities\n",
    "        return all_preds.to_numpy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Predictions against the test data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 165,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_pred.shape: (726, 14)\n"
     ]
    }
   ],
   "source": [
    "# instantiate the classifier\n",
    "br_clf = BinaryRelevanceClassifier(LogisticRegression())\n",
    "# fit\n",
    "br_clf.fit(X_train, y_train)\n",
    "# predict\n",
    "y_pred = br_clf.predict(X_test)\n",
    "print(\"y_pred.shape: \" + str(y_pred.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Custom Accuracy Measure for Classifiers"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The Accuracy measure used here has been described in - \"A Literature Survey on Algorithms for Multi-label Learning\" by \"Mohammad S Sorower\"."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 166,
   "metadata": {},
   "outputs": [],
   "source": [
    "def accuracy_score(y_test, y_pred):\n",
    "    # y_pred is a numpy array, y_test is a dataframe\n",
    "    # to compare the two, convert to a single type\n",
    "    y_test = y_test.to_numpy()\n",
    "    \n",
    "    # shape of test and preds must be equal\n",
    "    assert y_test.shape == y_pred.shape\n",
    "    i=0\n",
    "    # list of scores for each training sample\n",
    "    scores = []\n",
    "    \n",
    "    # for each test sample\n",
    "    while i < len(y_test):\n",
    "        count=0\n",
    "        # count the number of matches in the sample\n",
    "        # y_test[i] -> row values in test set (true values)\n",
    "        # y_pred[i] -> row values in predictions set (predicted values)\n",
    "        for p, q in zip(y_test[i], y_pred[i]):\n",
    "            if p == q:\n",
    "                count += 1\n",
    "\n",
    "        # accuracy score for the sample = no. of correctly predicted labels/total no. of labels\n",
    "        scores.append(count / y_pred.shape[1])\n",
    "        i+=1 \n",
    "\n",
    "    # final accuracy = avg. accuracy over all test samples =\n",
    "    # sum of the accuracy of all training samples/no. of training samples\n",
    "    return round((sum(scores)/len(y_test)), 5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 167,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Binary Relevance Classifier: 0.79073\n"
     ]
    }
   ],
   "source": [
    "print(\"Accuracy of Binary Relevance Classifier: \" + str(accuracy_score(y_test, y_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Experiment with different Base Classifiers - GridSearch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 168,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 5 folds for each of 6 candidates, totalling 30 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done  30 out of  30 | elapsed:   31.5s finished\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Parameters Found: \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'base_classifier': LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,\n",
       "                    intercept_scaling=1, l1_ratio=None, max_iter=20000,\n",
       "                    multi_class='warn', n_jobs=None, penalty='l2',\n",
       "                    random_state=None, solver='warn', tol=0.0001, verbose=0,\n",
       "                    warm_start=False)}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "0.799951691306919"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cv_folds=5\n",
    "\n",
    "# Set up the parameter grid to search\n",
    "param_grid ={'base_classifier': [DecisionTreeClassifier(criterion='entropy', max_depth=15, min_samples_leaf=2),\n",
    "                                 RandomForestClassifier(criterion='entropy'),\n",
    "                                 LogisticRegression(max_iter=20000), GaussianNB(), KNeighborsClassifier(), SVC()] }\n",
    "\n",
    "# Perform the search\n",
    "# Using the custom accuracy function defined earlier\n",
    "tuned_model = GridSearchCV(BinaryRelevanceClassifier(), \\\n",
    "                            param_grid, scoring=make_scorer(accuracy_score), verbose = 2, n_jobs = -1, cv=cv_folds)\n",
    "tuned_model.fit(X_train, y_train)\n",
    "\n",
    "# Print details of the best model\n",
    "print(\"Best Parameters Found: \")\n",
    "display(tuned_model.best_params_)\n",
    "display(tuned_model.best_score_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluate the best model against the Test Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 169,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_pred.shape: (726, 14)\n"
     ]
    }
   ],
   "source": [
    "# instantiate the classifier\n",
    "br_clf = BinaryRelevanceClassifier(LogisticRegression(max_iter=20000))\n",
    "# fit\n",
    "br_clf.fit(X_train, y_train)\n",
    "# predict\n",
    "y_pred = br_clf.predict(X_test)\n",
    "print(\"y_pred.shape: \" + str(y_pred.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 170,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Binary Relevance Classifier - Best Model: 0.79073\n"
     ]
    }
   ],
   "source": [
    "print(\"Accuracy of Binary Relevance Classifier - Best Model: \" + str(accuracy_score(y_test, y_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Task 2: Implement the Binary Relevance Algorithm with Under-Sampling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 171,
   "metadata": {},
   "outputs": [],
   "source": [
    "class BinaryRelevanceClassifierUS(BaseEstimator, ClassifierMixin):\n",
    "    def __init__(self, base_classifier=LogisticRegression(max_iter=20000)):\n",
    "        self.base_classifier=base_classifier\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        \"\"\"Build a Binary Relevance classifier with Under sampling from the training set (X, y).\n",
    "        Parameters\n",
    "        ----------\n",
    "        X : array-like or sparse matrix, shape = [n_samples, n_features]\n",
    "            The training input samples. Internally, it will be converted to\n",
    "            ``dtype=np.float32`` and if a sparse matrix is provided\n",
    "            to a sparse ``csc_matrix``.\n",
    "        y : array-like, shape = [n_samples] \n",
    "            The target values (class labels) as integers or strings.\n",
    "        \"\"\"\n",
    "\n",
    "        # list of individual classifiers\n",
    "        self.models = []\n",
    "        \n",
    "        # for each class label\n",
    "        for label in list(y.columns):\n",
    "            \n",
    "            X_cp = X.copy()\n",
    "            # pick the column values for the label\n",
    "            y_cp = y[label]\n",
    "            \n",
    "            # sampling is done on both X and y, hence join the two dataframes\n",
    "            X_y_data = pd.concat([X_cp, y_cp], axis=1)\n",
    "            \n",
    "            # counters for 0 values and 1 values\n",
    "            n_val0, n_val1 = 0,0\n",
    "            \n",
    "            j=0\n",
    "            # for each sample\n",
    "            while j<len(X_y_data):\n",
    "                # if value for the label is 0\n",
    "                if(X_y_data.iloc[j][label] == 0):\n",
    "                    n_val0+=1\n",
    "                else:\n",
    "                    # value 1\n",
    "                    n_val1+=1\n",
    "                j+=1\n",
    "            \n",
    "            # under sample the majority class\n",
    "            # randomly pick samples from majority class equal to the number of samples in the minority class\n",
    "            # both the classes will have the same number of samples\n",
    "            if n_val0 > n_val1:\n",
    "                # majority 0 values\n",
    "                val1 = X_y_data[X_y_data[label]==1]\n",
    "                val0 = X_y_data[X_y_data[label]==0].sample(n_val1)\n",
    "                \n",
    "                X_y_data = pd.concat([val0, val1], axis=0)\n",
    "            \n",
    "            elif n_val1 > n_val0:\n",
    "                # majority 1 values\n",
    "                val1 = X_y_data[X_y_data[label]==1].sample(n_val0)\n",
    "                val0 = X_y_data[X_y_data[label]==0]\n",
    "                \n",
    "                X_y_data = pd.concat([val0, val1], axis=0)\n",
    "            \n",
    "            # split back into X and y\n",
    "            X_cp = X_y_data.iloc[:, :-1]\n",
    "            y_cp = X_y_data.iloc[:, -1]\n",
    "            \n",
    "            base_model = clone(self.base_classifier)\n",
    "            # fit the base model - one model each for Y1, Y2....Y14\n",
    "            a, b = check_X_y(X_cp, y_cp)\n",
    "            base_model.fit(a, b)\n",
    "            # list of individual classifiers classifiers\n",
    "            self.models.append(base_model)\n",
    "\n",
    "\n",
    "    # The predict function to make a set of predictions for a set of query instances\n",
    "    def predict(self, X):\n",
    "        # check if the models list has been set up\n",
    "        check_is_fitted(self, ['models'])\n",
    "        X = check_array(X)\n",
    "        \n",
    "        all_preds = pd.DataFrame()\n",
    "        i=0\n",
    "        # list of individual classifier predictions\n",
    "        preds = []\n",
    "        \n",
    "        # for every fitted model\n",
    "        for model in self.models:\n",
    "            # predict for X\n",
    "            pred = model.predict(X)\n",
    "            # add to the list of predictions\n",
    "            preds.append(pd.DataFrame({'Class'+ str(i+1): pred}))\n",
    "            i+=1\n",
    "        \n",
    "        # store predictions for each label in a single dataframe\n",
    "        all_preds = pd.concat(preds, axis=1)\n",
    "        # standard sklearn classifiers return predictions as numpy arrays\n",
    "        # hence convert the dataframe to a numpy array\n",
    "        return all_preds.to_numpy()\n",
    "    \n",
    "    \n",
    "    def predict_proba(self,X):\n",
    "        # check if the models list has been set up\n",
    "        check_is_fitted(self, ['models'])\n",
    "        X = check_array(X)\n",
    "        \n",
    "        all_preds = pd.DataFrame()\n",
    "        i = 0\n",
    "        \n",
    "        for model in self.models:\n",
    "            # Call predict_proba of the each base model\n",
    "            pred = model.predict_proba(X)\n",
    "            # Add the probabilities of 1 to the dataframe\n",
    "            all_preds['Class'+str(i+1)] = [one_prob[1] for one_prob in pred]\n",
    "            i+=1\n",
    "        \n",
    "        #return probabilities\n",
    "        return all_preds.to_numpy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Prediction against the Test Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 172,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_pred.shape=(726, 14)\n"
     ]
    }
   ],
   "source": [
    "brus_clf = BinaryRelevanceClassifierUS()\n",
    "# fit\n",
    "brus_clf.fit(X_train, y_train)\n",
    "# predict\n",
    "y_pred = brus_clf.predict(X_test)\n",
    "print(\"y_pred.shape=\" + str(y_pred.shape))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Calculate the Accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 173,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Binary Relevance Classifier with Under-sampling: 0.62584\n"
     ]
    }
   ],
   "source": [
    "accuracy = accuracy_score(y_test, y_pred)\n",
    "print(\"Accuracy of Binary Relevance Classifier with Under-sampling: \" + str(accuracy))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Experiment with different Base Classifiers - GridSearch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 174,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 5 folds for each of 6 candidates, totalling 30 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done  30 out of  30 | elapsed:  1.5min finished\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Parameters Found: \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'base_classifier': RandomForestClassifier(bootstrap=True, class_weight=None, criterion='entropy',\n",
       "                        max_depth=None, max_features='auto', max_leaf_nodes=None,\n",
       "                        min_impurity_decrease=0.0, min_impurity_split=None,\n",
       "                        min_samples_leaf=1, min_samples_split=2,\n",
       "                        min_weight_fraction_leaf=0.0, n_estimators='warn',\n",
       "                        n_jobs=None, oob_score=False, random_state=None,\n",
       "                        verbose=0, warm_start=False)}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "0.6374084979302187"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cv_folds=5\n",
    "\n",
    "# Set up the parameter grid to search\n",
    "param_grid ={'base_classifier': [DecisionTreeClassifier(criterion='entropy', max_depth=15, min_samples_leaf=2),\n",
    "                                 RandomForestClassifier(criterion='entropy'),\n",
    "                                 LogisticRegression(max_iter=20000), GaussianNB(), KNeighborsClassifier(), SVC()] }\n",
    "\n",
    "# Perform the search\n",
    "tuned_model = GridSearchCV(BinaryRelevanceClassifierUS(), \\\n",
    "                            param_grid, scoring=make_scorer(accuracy_score), verbose = 2, n_jobs = -1, cv=cv_folds)\n",
    "tuned_model.fit(X_train, y_train)\n",
    "\n",
    "# Print details\n",
    "print(\"Best Parameters Found: \")\n",
    "display(tuned_model.best_params_)\n",
    "display(tuned_model.best_score_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluate the best model against the Test Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 175,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_pred.shape=(726, 14)\n"
     ]
    }
   ],
   "source": [
    "brus_clf = BinaryRelevanceClassifierUS(RandomForestClassifier(criterion='entropy'))\n",
    "# fit\n",
    "brus_clf.fit(X_train, y_train)\n",
    "# predict\n",
    "y_pred = brus_clf.predict(X_test)\n",
    "print(\"y_pred.shape=\" + str(y_pred.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 176,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Binary Relevance Classifier with Under-sampling - Best Model: 0.64856\n"
     ]
    }
   ],
   "source": [
    "accuracy = accuracy_score(y_test, y_pred)\n",
    "print(\"Accuracy of Binary Relevance Classifier with Under-sampling - Best Model: \" + str(accuracy))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Task 3: Compare the Performance of Different Binary Relevance Approaches"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Accuracy and F-1 Scores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 177,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===================Accuracy Scores=====================\n",
      "Binary Relevance\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'Decision Tree': 0.72845,\n",
       " 'Random Forest': 0.78817,\n",
       " 'Logistic Regression': 0.79073,\n",
       " 'GaussianNB': 0.70268,\n",
       " 'kNN': 0.78975,\n",
       " 'SVM': 0.77893}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Binary Relevance with Under-Sampling\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'Decision Tree': 0.58392,\n",
       " 'Random Forest': 0.64482,\n",
       " 'Logistic Regression': 0.62997,\n",
       " 'GaussianNB': 0.61196,\n",
       " 'kNN': 0.63243,\n",
       " 'SVM': 0.5671}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "======================F1 Scores========================\n",
      "Binary Relevance\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'Decision Tree': 0.39125767994484223,\n",
       " 'Random Forest': 0.34573811112703584,\n",
       " 'Logistic Regression': 0.343975003895816,\n",
       " 'GaussianNB': 0.4441966958169542,\n",
       " 'kNN': 0.42152433224517327,\n",
       " 'SVM': 0.19093646991285598}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Binary Relevance with Under-Sampling\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'Decision Tree': 0.38810897756820945,\n",
       " 'Random Forest': 0.4031524704019302,\n",
       " 'Logistic Regression': 0.4409926522752094,\n",
       " 'GaussianNB': 0.4374084595299819,\n",
       " 'kNN': 0.4584384830288687,\n",
       " 'SVM': 0.40455807802540894}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# list of base models\n",
    "base_models = [DecisionTreeClassifier(criterion='entropy', max_depth=15, min_samples_leaf=2),\n",
    "               RandomForestClassifier(criterion='entropy'),\n",
    "               LogisticRegression(max_iter=20000), GaussianNB(), KNeighborsClassifier(), SVC()]\n",
    "base_model_names = [\"Decision Tree\", \"Random Forest\", \"Logistic Regression\", \"GaussianNB\", \"kNN\", \"SVM\"]\n",
    "\n",
    "# store accuracy scores\n",
    "br_clf_accuracies = dict()\n",
    "br_clfus_accuracies = dict()\n",
    "\n",
    "# store F1 scores\n",
    "br_clf_f1 = dict()\n",
    "br_clfus_f1 = dict()\n",
    "\n",
    "i=0\n",
    "for clf in base_models:\n",
    "    # without undersampling\n",
    "    br_clf = BinaryRelevanceClassifier(clf)\n",
    "    br_clf.fit(X_train, y_train)\n",
    "    br_y_pred = br_clf.predict(X_test)\n",
    "\n",
    "    # find accuracy using custom accuracy function defined\n",
    "    accuracy = accuracy_score(y_test, br_y_pred)\n",
    "    br_clf_accuracies[base_model_names[i]] = accuracy\n",
    "\n",
    "    # find f1 score using sklearn\n",
    "    y_pred_df = pd.DataFrame(br_y_pred)\n",
    "    f1_score_br = metrics.f1_score(y_test, y_pred_df, average='macro')\n",
    "    br_clf_f1[base_model_names[i]] = f1_score_br\n",
    "\n",
    "    \n",
    "    # with undersampling\n",
    "    brus_clf = BinaryRelevanceClassifierUS(clf)\n",
    "    brus_clf.fit(X_train, y_train)\n",
    "    brus_y_pred = brus_clf.predict(X_test)\n",
    "    \n",
    "    # find accuracy using custom accuracy function defined\n",
    "    accuracy_us = accuracy_score(y_test, brus_y_pred)\n",
    "    br_clfus_accuracies[base_model_names[i]] = accuracy_us\n",
    "    \n",
    "    # find f1 score using sklearn\n",
    "    y_pred_df = pd.DataFrame(brus_y_pred)\n",
    "    f1_score_us = metrics.f1_score(y_test, y_pred_df, average='macro')\n",
    "    br_clfus_f1[base_model_names[i]] = f1_score_us\n",
    "    \n",
    "    i+=1\n",
    "\n",
    "print(\"===================Accuracy Scores=====================\")\n",
    "print(\"Binary Relevance\")\n",
    "display(br_clf_accuracies)\n",
    "print(\"Binary Relevance with Under-Sampling\")\n",
    "display(br_clfus_accuracies)\n",
    "\n",
    "print(\"======================F1 Scores========================\")\n",
    "print(\"Binary Relevance\")\n",
    "display(br_clf_f1)\n",
    "print(\"Binary Relevance with Under-Sampling\")\n",
    "display(br_clfus_f1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Plot the Accuracy  Scores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 178,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x21d806b2708>"
      ]
     },
     "execution_count": 178,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(list(br_clf_accuracies.keys()), list(br_clf_accuracies.values()))\n",
    "plt.plot(list(br_clfus_accuracies.keys()), list(br_clfus_accuracies.values()))\n",
    "plt.xticks(rotation=90)\n",
    "plt.legend(['Binary Relevance','Binary Relevance Under-Sampling'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Plot the F-1 Scores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 179,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x21d801525c8>"
      ]
     },
     "execution_count": 179,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(list(br_clf_f1.keys()), list(br_clf_f1.values()))\n",
    "plt.plot(list(br_clfus_f1.keys()), list(br_clfus_f1.values()))\n",
    "plt.xticks(rotation=90)\n",
    "plt.legend(['Binary Relevance','Binary Relevance Under-Sampling'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Task 4: Implement the Classifier Chains Algorithm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 180,
   "metadata": {},
   "outputs": [],
   "source": [
    "class ClassifierChains(BaseEstimator, ClassifierMixin):\n",
    "    def __init__(self, base_classifier=LogisticRegression(max_iter=20000), order=None):\n",
    "        self.base_classifier=base_classifier\n",
    "        self.order = order\n",
    "\n",
    "    def fit(self, X, y):\n",
    "        \"\"\"\n",
    "        Build a Classifier Chain from the training set (X, y).\n",
    "        Parameters\n",
    "        ----------\n",
    "        X : array-like or sparse matrix, shape = [n_samples, n_features]\n",
    "            The training input samples. Internally, it will be converted to\n",
    "            ``dtype=np.float32`` and if a sparse matrix is provided\n",
    "            to a sparse ``csc_matrix``.\n",
    "        y : array-like, shape = [n_samples, n_labels] \n",
    "            The target values (class labels) as integers or strings.\n",
    "\n",
    "        \"\"\"\n",
    "        \n",
    "        # check the order parameter\n",
    "        if self.order is None:\n",
    "            # default value - natural order for number of labels\n",
    "            self.order = list(range(y.shape[1]))\n",
    "        elif self.order == 'random':\n",
    "            # random order\n",
    "            self.order = list(range(y.shape[1]))\n",
    "            random.shuffle(self.order)\n",
    "        else:\n",
    "            # order specified\n",
    "            if(len(self.order)==y.shape[1]):\n",
    "                # expect order from 1, hence reduce 1 to consider zero indexing\n",
    "                self.order = [o - 1 for o in self.order]\n",
    "    \n",
    "        \n",
    "        # list of base models for each class\n",
    "        self.models = [clone(self.base_classifier) for clf in range(y.shape[1])]\n",
    "\n",
    "        # create a copy of X\n",
    "        X_joined = X.copy()\n",
    "       # X_joined.reset_index(drop=True, inplace=True)\n",
    "        \n",
    "        # create a new dataframe with X and y-in the order specified\n",
    "        # if order = [2,4,5,6...] -> X_joined= X, y2, y4, y5...\n",
    "        for val in self.order:\n",
    "            X_joined = pd.concat([X_joined, y['Class'+str(val+1)]], axis=1)\n",
    "\n",
    "        \n",
    "        # for each ith model, fit the model on X + y0 to yi-1 (in the order specified)\n",
    "        # if order = [2,4,6,....] fit 1st model on X for y2, fit second model on X+y2 for y4...\n",
    "        for chain_index, model in enumerate(self.models):\n",
    "            # select values of the class in order\n",
    "            y_vals = y.loc[:, 'Class'+str(self.order[chain_index]+1)]\n",
    "            # pick values for training - X+y upto the current label\n",
    "            t_X = X_joined.iloc[:, :(X.shape[1]+chain_index)]\n",
    "            check_X_y(t_X, y_vals)\n",
    "            # fit the model\n",
    "            model.fit(t_X, y_vals)\n",
    "\n",
    "\n",
    "            \n",
    "    # The predict function to make a set of predictions for a set of query instances\n",
    "    def predict(self, X):\n",
    "        \n",
    "        # check if the models list has been set up\n",
    "        check_is_fitted(self, ['models'])\n",
    "        \n",
    "        # dataframe to maintain previous predictions\n",
    "        pred_chain = pd.DataFrame(columns=['Class'+str(o+1) for o in self.order])\n",
    "        \n",
    "        X_copy = X.copy()\n",
    "        X_joined = X.copy()\n",
    "        \n",
    "        # use default indexing\n",
    "        X_joined.reset_index(drop=True, inplace=True)\n",
    "        X_copy.reset_index(drop=True, inplace=True)\n",
    "\n",
    "        i=0\n",
    "        \n",
    "        # for each ith model, predict based on X + predictions of all models upto i-1\n",
    "        # happens in the specified order since models are already fitted according to the order\n",
    "        for chain_index, model in enumerate(self.models):\n",
    "            # select previous predictions - all columns upto the current index\n",
    "            prev_preds = pred_chain.iloc[:, :chain_index]\n",
    "            # join the previous predictions with X\n",
    "            X_joined = pd.concat([X_copy, prev_preds], axis=1)\n",
    "            # predict on the base model\n",
    "            pred = model.predict(X_joined)\n",
    "            # add the new prediction to the pred chain\n",
    "            pred_chain['Class'+str(self.order[i]+1)] = pred\n",
    "            i+=1\n",
    "\n",
    "        # re-arrange the columns in natural order to return the predictions\n",
    "        pred_chain = pred_chain.loc[:, ['Class'+str(j+1) for j in range(0, len(self.order))]]\n",
    "        # all sklearn implementations return numpy array\n",
    "        # hence convert the dataframe to numpy array\n",
    "        return pred_chain.to_numpy()\n",
    "    \n",
    "    \n",
    "    \n",
    "    # Function to predict probabilities of 1s\n",
    "    def predict_proba(self, X):\n",
    "        # check if the models list has been set up\n",
    "        check_is_fitted(self, ['models'])\n",
    "        \n",
    "        # dataframe to maintain previous predictions\n",
    "        pred_chain = pd.DataFrame(columns=['Class'+str(o+1) for o in self.order])\n",
    "        # dataframe to maintain probabilities of class labels\n",
    "        pred_probs = pd.DataFrame(columns=['Class'+str(o+1) for o in self.order])\n",
    "        X_copy = X.copy()\n",
    "        X_joined = X.copy()\n",
    "        \n",
    "        # use default indexing\n",
    "        X_joined.reset_index(drop=True, inplace=True)\n",
    "        X_copy.reset_index(drop=True, inplace=True)\n",
    "\n",
    "        i=0\n",
    "        \n",
    "        # for each ith model, predict based on X + predictions of all models upto i-1\n",
    "        # happens in the specified order since models are already fitted according to the order\n",
    "        for chain_index, model in enumerate(self.models):\n",
    "            \n",
    "            # select previous predictions - all columns upto the current index\n",
    "            prev_preds = pred_chain.iloc[:, :chain_index]\n",
    "            # join the previous predictions with X\n",
    "            X_joined = pd.concat([X_copy, prev_preds], axis=1)\n",
    "            # predict on the base model\n",
    "            pred = model.predict(X_joined)\n",
    "            # predict probabilities\n",
    "            pred_proba = model.predict_proba(X_joined)\n",
    "            # add the new prediction to the pred chain\n",
    "            pred_chain['Class'+str(self.order[i]+1)] = pred\n",
    "            # save the probabilities of 1 according to label order\n",
    "            pred_probs['Class'+str(self.order[i]+1)] = [one_prob[1] for one_prob in pred_proba]\n",
    "            i+=1\n",
    "\n",
    "        # re-arrange the columns in natural order to return the probabilities\n",
    "        pred_probs = pred_probs.loc[:, ['Class'+str(j+1) for j in range(0, len(self.order))]]\n",
    "        # all sklearn implementations return numpy array\n",
    "        # hence convert the dataframe to numpy array\n",
    "        return pred_probs.to_numpy()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Predictions against the Test data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 181,
   "metadata": {},
   "outputs": [],
   "source": [
    "cc = ClassifierChains(order=[1,2,3,4,5,6,7,8,9,10,11,12,13,14])\n",
    "# fit\n",
    "cc.fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 182,
   "metadata": {},
   "outputs": [],
   "source": [
    "# predict\n",
    "cc_pred = cc.predict(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Calculate the Accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 183,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Classifier Chains: 0.78424\n"
     ]
    }
   ],
   "source": [
    "print(\"Accuracy of Classifier Chains: \" + str(accuracy_score(y_test, cc_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Experiment with different Base Classifiers - GridSearch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 184,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fitting 5 folds for each of 120 candidates, totalling 600 fits\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done  33 tasks      | elapsed:   49.1s\n",
      "[Parallel(n_jobs=-1)]: Done 154 tasks      | elapsed:  3.4min\n",
      "[Parallel(n_jobs=-1)]: Done 357 tasks      | elapsed:  5.2min\n",
      "[Parallel(n_jobs=-1)]: Done 600 out of 600 | elapsed:  9.7min finished\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best Parameters Found: \n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'base_classifier': LogisticRegression(C=1.0, class_weight=None, dual=False, fit_intercept=True,\n",
       "                    intercept_scaling=1, l1_ratio=None, max_iter=20000,\n",
       "                    multi_class='warn', n_jobs=None, penalty='l2',\n",
       "                    random_state=None, solver='warn', tol=0.0001, verbose=0,\n",
       "                    warm_start=False),\n",
       " 'order': [11, 4, 10, 1, 9, 5, 3, 12, 6, 8, 2, 7, 13, 14]}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "0.7951338024837374"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cv_folds=5\n",
    "\n",
    "# generate 20 random orders for class labels \n",
    "rand_orders = [list(range(1, y_test.shape[1]+1)) for i in list(range(1, 20))]\n",
    "\n",
    "for lst in rand_orders:\n",
    "    random.shuffle(lst)\n",
    "\n",
    "# make sure natural order is present\n",
    "rand_orders.append([1,2,3,4,5,6,7,8,9,10,11,12,13,14])\n",
    "\n",
    "# Set up the parameter grid to search\n",
    "param_grid ={'base_classifier': [DecisionTreeClassifier(criterion='entropy', max_depth=15, min_samples_leaf=2),\n",
    "                                 RandomForestClassifier(criterion='entropy'),\n",
    "                                 LogisticRegression(max_iter=20000), GaussianNB(), KNeighborsClassifier(), SVC()],\n",
    "            'order': rand_orders}\n",
    "\n",
    "# Perform the search\n",
    "tuned_model = GridSearchCV(ClassifierChains(), \\\n",
    "                            param_grid, scoring=make_scorer(accuracy_score), verbose = 2, n_jobs = -1, cv=cv_folds)\n",
    "tuned_model.fit(X_train, y_train)\n",
    "\n",
    "# Print details\n",
    "print(\"Best Parameters Found: \")\n",
    "display(tuned_model.best_params_)\n",
    "display(tuned_model.best_score_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 197,
   "metadata": {},
   "outputs": [],
   "source": [
    "# store the best parameters\n",
    "best = tuned_model.best_params_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluate the best model against the Test Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 198,
   "metadata": {},
   "outputs": [],
   "source": [
    "cc = ClassifierChains(base_classifier=best['base_classifier'], order=best['order'])\n",
    "cc.fit(X_train, y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 199,
   "metadata": {},
   "outputs": [],
   "source": [
    "cc_pred = cc.predict(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 200,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy of Classifier Chains - Best Model: 0.78847\n"
     ]
    }
   ],
   "source": [
    "print(\"Accuracy of Classifier Chains - Best Model: \" + str(accuracy_score(y_test, cc_pred)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Task 5: Evaluate the Performance of the Classifier Chains Algorithm"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Accuracy and F1 Scores of Classifier Chains"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 204,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "====================Classifier Chains Accuracy====================\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'Decision Tree': 0.72137,\n",
       " 'Random Forest': 0.79113,\n",
       " 'Logistic Regression': 0.78847,\n",
       " 'GaussianNB': 0.68939,\n",
       " 'kNN': 0.78394,\n",
       " 'SVM': 0.77588}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "===================Classifier Chains F1 Scores====================\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'Decision Tree': 0.38906778855908924,\n",
       " 'Random Forest': 0.3609104334374137,\n",
       " 'Logistic Regression': 0.3846810487809151,\n",
       " 'GaussianNB': 0.43195691283452253,\n",
       " 'kNN': 0.4175588720744822,\n",
       " 'SVM': 0.23331639887751227}"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# list of base models\n",
    "base_models = [DecisionTreeClassifier(criterion='entropy', max_depth=15, min_samples_leaf=2),\n",
    "               RandomForestClassifier(criterion='entropy'),\n",
    "               LogisticRegression(max_iter=20000), GaussianNB(), KNeighborsClassifier(), SVC()]\n",
    "base_model_names = [\"Decision Tree\", \"Random Forest\", \"Logistic Regression\", \"GaussianNB\", \"kNN\", \"SVM\"]\n",
    "\n",
    "cc_accuracies = dict()\n",
    "cc_f1 = dict()\n",
    "\n",
    "i=0\n",
    "for clf in base_models:\n",
    "    cc = ClassifierChains(clf, order=best['order'])\n",
    "    cc.fit(X_train, y_train)\n",
    "    cc_pred = cc.predict(X_test)\n",
    "    # accuracy score\n",
    "    accuracy = accuracy_score(y_test, cc_pred)\n",
    "    cc_accuracies[base_model_names[i]] = accuracy\n",
    "    # F1 score\n",
    "    cc_f1_score = metrics.f1_score(y_test, pd.DataFrame(cc_pred), average='macro')\n",
    "    cc_f1[base_model_names[i]] = cc_f1_score\n",
    "    i+=1\n",
    "\n",
    "print(\"====================Classifier Chains Accuracy====================\")\n",
    "display(cc_accuracies)\n",
    "print(\"===================Classifier Chains F1 Scores====================\")\n",
    "display(cc_f1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Comparison with Binary Relevance Approaches"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Accuracy Scores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 205,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x21d89857f08>"
      ]
     },
     "execution_count": 205,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(list(br_clf_accuracies.keys()), list(br_clf_accuracies.values()))\n",
    "plt.plot(list(br_clfus_accuracies.keys()), list(br_clfus_accuracies.values()))\n",
    "plt.plot(list(cc_accuracies.keys()), list(cc_accuracies.values()))\n",
    "plt.xticks(rotation=90)\n",
    "plt.legend(['Binary Relevance','BR-Under-Sampling', 'Classifier Chains'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### F1 Scores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 206,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x21d898c55c8>"
      ]
     },
     "execution_count": 206,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(list(br_clf_f1.keys()), list(br_clf_f1.values()))\n",
    "plt.plot(list(br_clfus_f1.keys()), list(br_clfus_f1.values()))\n",
    "plt.plot(list(cc_f1.keys()), list(cc_f1.values()))\n",
    "plt.xticks(rotation=90)\n",
    "plt.legend(['Binary Relevance','BR-Under-Sampling', 'Classifier Chains'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Task 6: Reflect on the Performance of the Different Models Evaluated"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Consider **BR=Binary Relevance Classifier, BRUS=Binary Relevance Classifier with Under Sampling, CC=Classifier Chains.**\n",
    "\n",
    "**From the above experiment we can conclude that:**\n",
    "\n",
    "* BR works much better than BRUS. This is because with undersampling, relevant training samples may be lost which affect the accuracy.\n",
    "* BRUS gives better F1 scores for all the base models as compared to BR. Lower F1 score is an indication that the data is highly biased which means high precision and low recall. So when the data gets more balanced after undersampling, we achieve better F1 scores.\n",
    "* CC works better than BR and BRUS because it takes label dependency into consideration while making predictions. BR on the other hand fits and predicts independently of other labels. To my surprise, in this experiment we can obseve that CC outperforms BR only in case of RandomForest as the base model. \n",
    "* The performance of CC depends on the label order as while making current predictions, we consider the results and labels of previous predictions. To experiment with label ordering, random 20 label orders were used in grid search. To my observation, the results accross different runs are inconsistent pertaining to label orders.(refer the above plots)\n",
    "* In terms of complexity, BR is simple and faster as compared to CC. In terms of performance, CC has better performance than BR due to reasons mentioned above.\n",
    "* While dealing with Yeast Dataset, CC works slightly better than BR which is a surprise. So, while dealing with Yeast Dataset, we can use BR as an adequate model if complexity and speed is a concern."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# References:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "[1] Read, J., Pfahringer, B., Holmes, G. and Frank, E. (2011). Classifier chains for multi-label classification. Machine Learning, [online] 85(3), pp.333-359. Available at: https://link.springer.com/content/pdf/10.1007/s10994-011-5256-5.pdf."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "[2] A Literature Survey on Algorithms for Multi-label Learning by Mohammad S Sorower. Available at: https://pdfs.semanticscholar.org/6b56/91db1e3a79af5e3c136d2dd322016a687a0b.pdf"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "\n"
   ]
  }
 ],
 "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": "3.7.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}