online_purchase_propensity.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "# Online Purchase Propensity"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "## Project Overview"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "This [Kaggle](https://www.kaggle.com/benpowis/customer-propensity-to-purchase-data) dataset \n",
    "tracks a day's worth of user visits for a ficticious online retail website. 607055 rows, 25 \n",
    "columns, and (excluding \"UserID\") entirely composed of binary data. Features #2 through #24 track \n",
    "various user states and actions across the website, while #25 denotes whether the user session \n",
    "resulted in an order. The end goal is to prototype a couple machine learning models to predict \n",
    "user propensity to purchase for our hypothetical client."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "## Data Exploration"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:47:41.095721Z",
     "iopub.status.busy": "2021-07-12T19:47:41.094537Z",
     "iopub.status.idle": "2021-07-12T19:47:43.314231Z",
     "shell.execute_reply": "2021-07-12T19:47:43.315200Z"
    },
    "lines_to_next_cell": 1
   },
   "outputs": [],
   "source": [
    "# module imports\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import seaborn as sb\n",
    "from hyperopt import fmin, hp, STATUS_OK, tpe, Trials\n",
    "from imblearn.ensemble import BalancedRandomForestClassifier\n",
    "from IPython.display import Markdown, display\n",
    "from sklearn.metrics import classification_report, f1_score\n",
    "from sklearn.model_selection import StratifiedKFold\n",
    "from sklearn.naive_bayes import BernoulliNB\n",
    "from statsmodels.stats.outliers_influence import variance_inflation_factor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:47:43.320801Z",
     "iopub.status.busy": "2021-07-12T19:47:43.319654Z",
     "iopub.status.idle": "2021-07-12T19:47:43.322027Z",
     "shell.execute_reply": "2021-07-12T19:47:43.322975Z"
    },
    "lines_to_next_cell": 1
   },
   "outputs": [],
   "source": [
    "# markdown printer\n",
    "def printmd(text):\n",
    "    display(Markdown(text))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:47:43.328481Z",
     "iopub.status.busy": "2021-07-12T19:47:43.327680Z",
     "iopub.status.idle": "2021-07-12T19:47:44.507366Z",
     "shell.execute_reply": "2021-07-12T19:47:44.508380Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "**Train info:**"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 455401 entries, 0 to 455400\n",
      "Data columns (total 25 columns):\n",
      " #   Column                   Non-Null Count   Dtype \n",
      "---  ------                   --------------   ----- \n",
      " 0   UserID                   455401 non-null  object\n",
      " 1   basket_icon_click        455401 non-null  int64 \n",
      " 2   basket_add_list          455401 non-null  int64 \n",
      " 3   basket_add_detail        455401 non-null  int64 \n",
      " 4   sort_by                  455401 non-null  int64 \n",
      " 5   image_picker             455401 non-null  int64 \n",
      " 6   account_page_click       455401 non-null  int64 \n",
      " 7   promo_banner_click       455401 non-null  int64 \n",
      " 8   detail_wishlist_add      455401 non-null  int64 \n",
      " 9   list_size_dropdown       455401 non-null  int64 \n",
      " 10  closed_minibasket_click  455401 non-null  int64 \n",
      " 11  checked_delivery_detail  455401 non-null  int64 \n",
      " 12  checked_returns_detail   455401 non-null  int64 \n",
      " 13  sign_in                  455401 non-null  int64 \n",
      " 14  saw_checkout             455401 non-null  int64 \n",
      " 15  saw_sizecharts           455401 non-null  int64 \n",
      " 16  saw_delivery             455401 non-null  int64 \n",
      " 17  saw_account_upgrade      455401 non-null  int64 \n",
      " 18  saw_homepage             455401 non-null  int64 \n",
      " 19  device_mobile            455401 non-null  int64 \n",
      " 20  device_computer          455401 non-null  int64 \n",
      " 21  device_tablet            455401 non-null  int64 \n",
      " 22  returning_user           455401 non-null  int64 \n",
      " 23  loc_uk                   455401 non-null  int64 \n",
      " 24  ordered                  455401 non-null  int64 \n",
      "dtypes: int64(24), object(1)\n",
      "memory usage: 86.9+ MB\n",
      "None \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/markdown": [
       "**Test info:**"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 151655 entries, 0 to 151654\n",
      "Data columns (total 25 columns):\n",
      " #   Column                   Non-Null Count   Dtype \n",
      "---  ------                   --------------   ----- \n",
      " 0   UserID                   151655 non-null  object\n",
      " 1   basket_icon_click        151655 non-null  int64 \n",
      " 2   basket_add_list          151655 non-null  int64 \n",
      " 3   basket_add_detail        151655 non-null  int64 \n",
      " 4   sort_by                  151655 non-null  int64 \n",
      " 5   image_picker             151655 non-null  int64 \n",
      " 6   account_page_click       151655 non-null  int64 \n",
      " 7   promo_banner_click       151655 non-null  int64 \n",
      " 8   detail_wishlist_add      151655 non-null  int64 \n",
      " 9   list_size_dropdown       151655 non-null  int64 \n",
      " 10  closed_minibasket_click  151655 non-null  int64 \n",
      " 11  checked_delivery_detail  151655 non-null  int64 \n",
      " 12  checked_returns_detail   151655 non-null  int64 \n",
      " 13  sign_in                  151655 non-null  int64 \n",
      " 14  saw_checkout             151655 non-null  int64 \n",
      " 15  saw_sizecharts           151655 non-null  int64 \n",
      " 16  saw_delivery             151655 non-null  int64 \n",
      " 17  saw_account_upgrade      151655 non-null  int64 \n",
      " 18  saw_homepage             151655 non-null  int64 \n",
      " 19  device_mobile            151655 non-null  int64 \n",
      " 20  device_computer          151655 non-null  int64 \n",
      " 21  device_tablet            151655 non-null  int64 \n",
      " 22  returning_user           151655 non-null  int64 \n",
      " 23  loc_uk                   151655 non-null  int64 \n",
      " 24  ordered                  151655 non-null  int64 \n",
      "dtypes: int64(24), object(1)\n",
      "memory usage: 28.9+ MB\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "# importing data\n",
    "train = pd.read_csv('train.csv')\n",
    "test = pd.read_csv('test.csv')\n",
    "printmd('**Train info:**')\n",
    "print(train.info(), '\\n')\n",
    "printmd('**Test info:**')\n",
    "print(test.info())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "Both dataframes are quite clean, no missing or null values. Dropping the \"UserID\" column as it \n",
    "won't be needed in the remaining analysis."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:47:44.571527Z",
     "iopub.status.busy": "2021-07-12T19:47:44.570330Z",
     "iopub.status.idle": "2021-07-12T19:47:44.572743Z",
     "shell.execute_reply": "2021-07-12T19:47:44.573683Z"
    }
   },
   "outputs": [],
   "source": [
    "# dropping userid\n",
    "train.drop(\"UserID\", axis = 1, inplace = True)\n",
    "test.drop(\"UserID\", axis = 1, inplace = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:47:44.579763Z",
     "iopub.status.busy": "2021-07-12T19:47:44.578616Z",
     "iopub.status.idle": "2021-07-12T19:47:44.648391Z",
     "shell.execute_reply": "2021-07-12T19:47:44.649176Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "**Train balance:**"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0    0.958074\n",
      "1    0.041926\n",
      "Name: ordered, dtype: float64 \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/markdown": [
       "**Test balance:**"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0    1.0\n",
      "Name: ordered, dtype: float64\n"
     ]
    }
   ],
   "source": [
    "# class balances\n",
    "printmd('**Train balance:**')\n",
    "print(train.ordered.value_counts()/train.shape[0], '\\n')\n",
    "printmd('**Test balance:**')\n",
    "print(test.ordered.value_counts()/test.shape[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "This dataset suffers from heavy class imbalance, something we will need to keep in mind \n",
    "when building predictive models sensitive to said imbalance. Note that the test dataset is \n",
    "composed entirely of \"0\" \"ordered\" instances. This is expected to have a significant impact on \n",
    "certain metrics for assessing model performance."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:47:44.667782Z",
     "iopub.status.busy": "2021-07-12T19:47:44.666718Z",
     "iopub.status.idle": "2021-07-12T19:47:46.514548Z",
     "shell.execute_reply": "2021-07-12T19:47:46.515526Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# correlation plot\n",
    "fig, ax = plt.subplots()\n",
    "fig.set_size_inches(10, 8)\n",
    "sb.heatmap(train.corr(), cmap = 'winter', ax = ax)\n",
    "ax.set_title('Column Correlation Matrix')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "The above heatmap helps visualize predictor-predictor and predictor-response correlations at a \n",
    "glance. \"checked_delivery_detail\", \"sign_in\", and \"saw_checkout\" all show promising amounts of \n",
    "correlation with the response variable, albeit with similarly high levels of multicollinearity \n",
    "between themselves.\n",
    "\n",
    "A more quantitative method of assessing multicollinearity amongst the predictors would be to \n",
    "be to calculate variance inflation factors (VIFs)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:47:46.523714Z",
     "iopub.status.busy": "2021-07-12T19:47:46.522832Z",
     "iopub.status.idle": "2021-07-12T19:47:46.524551Z",
     "shell.execute_reply": "2021-07-12T19:47:46.525209Z"
    }
   },
   "outputs": [],
   "source": [
    "# x-y splits\n",
    "X_train = train.iloc[:, :-1]\n",
    "y_train = train['ordered']\n",
    "X_test = test.iloc[:, :-1]\n",
    "y_test = test['ordered']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:47:46.531535Z",
     "iopub.status.busy": "2021-07-12T19:47:46.530718Z",
     "iopub.status.idle": "2021-07-12T19:48:02.305105Z",
     "shell.execute_reply": "2021-07-12T19:48:02.305784Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "                               vif\n",
      "col                               \n",
      "loc_uk                   14.688576\n",
      "device_mobile            11.780301\n",
      "sign_in                   9.786845\n",
      "saw_checkout              9.661590\n",
      "device_computer           3.687012\n",
      "device_tablet             3.110726\n",
      "returning_user            2.209838\n",
      "checked_delivery_detail   2.129950\n",
      "basket_icon_click         2.047719\n",
      "basket_add_detail         1.780923\n",
      "list_size_dropdown        1.765275\n",
      "basket_add_list           1.674745\n",
      "saw_homepage              1.579528\n",
      "sort_by                   1.185449\n",
      "closed_minibasket_click   1.151321\n",
      "image_picker              1.138775\n",
      "account_page_click        1.130253\n",
      "checked_returns_detail    1.075586\n",
      "saw_delivery              1.066949\n",
      "detail_wishlist_add       1.066770\n",
      "promo_banner_click        1.042587\n",
      "saw_account_upgrade       1.005718\n",
      "saw_sizecharts            1.001376\n"
     ]
    }
   ],
   "source": [
    "# calculating vifs\n",
    "vifs = pd.DataFrame(X_train.columns, columns = ['col'])\n",
    "vifs.loc[:, 'vif'] = [variance_inflation_factor(X_train.values, i) for i in range(X_train.shape[1])]\n",
    "vifs = vifs.sort_values('vif', ascending = False).set_index('col', drop = True)\n",
    "print(vifs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "The top four variables in this list make up those first to be addressed: \"loc_uk\", \n",
    "\"device_mobile\", \"sign_in\", and \"saw_checkout\". In a much wider dataset, feature engineering and \n",
    "reduction might be necessary to wrangle in our model; our current set of 23 features will be kept \n",
    "as is."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "## Building Models"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "### 1. BernoulliNB (bnb)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "The first model will serve as the baseline to beat, a derivative of the naive bayes theorem \n",
    "specifically for boolean features. For reasons covered later on, this model should perform well \n",
    "despite some hidden gotchas in this particular dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:48:02.309059Z",
     "iopub.status.busy": "2021-07-12T19:48:02.308379Z",
     "iopub.status.idle": "2021-07-12T19:48:02.924336Z",
     "shell.execute_reply": "2021-07-12T19:48:02.925041Z"
    },
    "lines_to_next_cell": 1
   },
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "**Classification Report**"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0     1.0000    0.9800    0.9899    151655\n",
      "           1     0.0000    0.0000    0.0000         0\n",
      "\n",
      "    accuracy                         0.9800    151655\n",
      "   macro avg     0.5000    0.4900    0.4950    151655\n",
      "weighted avg     1.0000    0.9800    0.9899    151655\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# model construction\n",
    "bnb = BernoulliNB()\n",
    "bnb.fit(X_train, y_train)\n",
    "y_pred = bnb.predict(X_test)\n",
    "printmd('**Classification Report**')\n",
    "print(classification_report(y_test, y_pred, digits = 4, zero_division = 0))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "The impact of only having one class in the test dataset is evident in the class \"1\" f1 score (0) \n",
    "and macro average f1 score (.495). Both metrics suffer with no true positives for class \"1\"; \n",
    "false positives make up the only \"1\" class predictions and therefore tank precision and recall. \n",
    "For a better representation of predictive performance on a two-class test dataset, we turn to \n",
    "stratified cross validation of the training dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:48:02.932739Z",
     "iopub.status.busy": "2021-07-12T19:48:02.931898Z",
     "iopub.status.idle": "2021-07-12T19:48:02.935114Z",
     "shell.execute_reply": "2021-07-12T19:48:02.934268Z"
    },
    "lines_to_next_cell": 1
   },
   "outputs": [],
   "source": [
    "# validation function\n",
    "def model_validation(clf, X, y, format = 'report'):\n",
    "    '''\n",
    "    Stratified cross validation function, outputs either classification_report or f1_score.\n",
    "\n",
    "        Arguments:\n",
    "            clf (object): classifier, supports .fit / .predict\n",
    "            X (array): feature array, n x d\n",
    "            y (array): label array, n x 1\n",
    "            format (string): output format, 'report' or 'score'\n",
    "        \n",
    "        Returns:\n",
    "            result (string / float): validation metric, depends on format variable\n",
    "    '''\n",
    "    preds = []\n",
    "    tests = []\n",
    "    skf = StratifiedKFold(shuffle = True)\n",
    "    for train_ind, test_ind in skf.split(X, y):\n",
    "        train_X, test_X = X.values[train_ind], X.values[test_ind]\n",
    "        train_y, test_y = y[train_ind], y[test_ind]\n",
    "        clf.fit(train_X, train_y)\n",
    "        pred_y = clf.predict(test_X)\n",
    "        preds.append(pred_y)\n",
    "        tests.append(test_y)\n",
    "    if format == 'report':\n",
    "        result = classification_report(np.hstack(tests), np.hstack(preds), digits = 4)\n",
    "    else:\n",
    "        result = f1_score(np.hstack(tests), np.hstack(preds), average = 'macro')\n",
    "    return result"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:48:02.939985Z",
     "iopub.status.busy": "2021-07-12T19:48:02.938993Z",
     "iopub.status.idle": "2021-07-12T19:48:05.673166Z",
     "shell.execute_reply": "2021-07-12T19:48:05.673862Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "**Classification Report**"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0     0.9997    0.9810    0.9903    436308\n",
      "           1     0.6962    0.9927    0.8184     19093\n",
      "\n",
      "    accuracy                         0.9815    455401\n",
      "   macro avg     0.8479    0.9869    0.9043    455401\n",
      "weighted avg     0.9870    0.9815    0.9831    455401\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# bernoulli validation\n",
    "printmd('**Classification Report**')\n",
    "bnb = BernoulliNB()\n",
    "print(model_validation(bnb, X_train, y_train))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "Now that the naive bayes model is predicting on instances from both classes, we get a more \n",
    "realistic look at predictive performance. Our cross validated model's precision for class \"1\" \n",
    "does leave something to be desired, something we hope to improve with another model. \n",
    "\n",
    "Above it was mentioned that the heavy class imbalance would impact models sensitive to said \n",
    "imbalance. BernoulliNB is a probabilistic, binary classification model (in this specific \n",
    "instance), and therefore is mostly unaffected. Any deterministic models we attempt to implement \n",
    "will need to take steps to address this imbalance."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "### 2. BalancedRandomForestClassifier (brf)\n",
    "\n",
    "Enter \"imbalanced-learn\" (also \"imblearn\"), a Python library built for the explicit purpose \n",
    "of dealing with class imbalanced datasets. Touted for it's high degree of compatibility with \n",
    "scikit-learn, \"imbalanced-learn\" contains a hodge podge of resampling and \n",
    "sklearn-wrapped methods tailor-built for handling unbalanced datasets.\n",
    "\n",
    "In contrast to sklearn's RandomForestClassifier, brf includes parameters to control random \n",
    "undersampling. Each weak learner in the bagged model is trained on an undersampled (with \n",
    "replacement) subset of the original dataset, albeit with a more favorable / equal balance of \n",
    "classes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:48:05.680681Z",
     "iopub.status.busy": "2021-07-12T19:48:05.679805Z",
     "iopub.status.idle": "2021-07-12T19:48:05.681789Z",
     "shell.execute_reply": "2021-07-12T19:48:05.682514Z"
    },
    "lines_to_next_cell": 1
   },
   "outputs": [],
   "source": [
    "# parameter space\n",
    "params = {\n",
    "    'n_estimators': hp.quniform('n_estimators', 20, 80, 1),\n",
    "    'max_depth': hp.quniform('max_depth', 1, 15, 1),\n",
    "    'min_samples_split': hp.quniform('min_samples_split', 2, 50, 1),\n",
    "    'max_features': hp.quniform('max_features', 1, 23, 1),\n",
    "    'sampling_strategy': hp.choice(\n",
    "        'sampling_strategy',\n",
    "        ['majority', 'not minority', 'not majority', 'all']\n",
    "    ),\n",
    "    'replacement': hp.choice('replacement', ['True', 'False']),\n",
    "    'class_weight': hp.choice('class_weight', [None, 'balanced', 'balanced_subsample'])\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:48:05.688208Z",
     "iopub.status.busy": "2021-07-12T19:48:05.687422Z",
     "iopub.status.idle": "2021-07-12T19:48:05.689702Z",
     "shell.execute_reply": "2021-07-12T19:48:05.690325Z"
    },
    "lines_to_next_cell": 1
   },
   "outputs": [],
   "source": [
    "# optimization function\n",
    "def objective(params):\n",
    "    keys = list(params.keys())\n",
    "    floats = ['n_estimators', 'max_depth', 'min_samples_split', 'max_features']\n",
    "    params = {k:(int(params[k]) if k in floats else params[k]) for k in keys}\n",
    "    brf = BalancedRandomForestClassifier(**params, n_jobs = -1, random_state = 42)\n",
    "    score = model_validation(brf, X_train, y_train, 'score')\n",
    "    return {'loss': -score, 'status': STATUS_OK}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "To eke out as much predictive performance as possible, we will be relying on HyperOpt for some\n",
    "hyperparameter tuning. The fmin function below will test various combinations of \n",
    "hyperparameters from a pre-defined parameter space, using outcomes to guide it to the optimal \n",
    "set. Compared to GridSearch and RandomSearch, this translates to less time spent testing \n",
    "conditions with low chances of improving performance.\n",
    "\n",
    "One quick warning, this notebook was run on a 24 core system and the below cell took 12 minutes \n",
    "and 57 seconds to run. Your execution time will scale based on how your system compares with mine."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T19:48:05.702819Z",
     "iopub.status.busy": "2021-07-12T19:48:05.695824Z",
     "iopub.status.idle": "2021-07-12T20:03:29.746456Z",
     "shell.execute_reply": "2021-07-12T20:03:29.747166Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      "  0%|          | 0/50 [00:00<?, ?trial/s, best loss=?]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      "  2%|▏         | 1/50 [00:15<12:22, 15.14s/trial, best loss: -0.8458885202288859]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      "  4%|▍         | 2/50 [00:29<11:37, 14.53s/trial, best loss: -0.9499473556917339]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      "  6%|▌         | 3/50 [00:56<16:06, 20.57s/trial, best loss: -0.9520158543837349]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      "  8%|▊         | 4/50 [01:27<18:42, 24.40s/trial, best loss: -0.9520158543837349]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 10%|█         | 5/50 [01:56<19:34, 26.10s/trial, best loss: -0.9520158543837349]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 12%|█▏        | 6/50 [02:26<20:01, 27.30s/trial, best loss: -0.9520158543837349]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 14%|█▍        | 7/50 [02:35<15:26, 21.54s/trial, best loss: -0.9596473668450722]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 16%|█▌        | 8/50 [03:06<17:09, 24.52s/trial, best loss: -0.9596473668450722]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 18%|█▊        | 9/50 [03:23<15:07, 22.14s/trial, best loss: -0.9596473668450722]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 20%|██        | 10/50 [03:51<15:59, 23.98s/trial, best loss: -0.9596473668450722]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 22%|██▏       | 11/50 [04:21<16:46, 25.80s/trial, best loss: -0.9596473668450722]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 24%|██▍       | 12/50 [04:38<14:38, 23.12s/trial, best loss: -0.9596473668450722]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 26%|██▌       | 13/50 [04:56<13:21, 21.66s/trial, best loss: -0.9596473668450722]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 28%|██▊       | 14/50 [05:29<14:59, 24.99s/trial, best loss: -0.9596473668450722]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 30%|███       | 15/50 [05:41<12:15, 21.01s/trial, best loss: -0.9599469736524371]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 32%|███▏      | 16/50 [05:51<10:07, 17.87s/trial, best loss: -0.9599469736524371]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 34%|███▍      | 17/50 [06:09<09:43, 17.68s/trial, best loss: -0.9599469736524371]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 36%|███▌      | 18/50 [06:14<07:27, 14.00s/trial, best loss: -0.9599469736524371]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 38%|███▊      | 19/50 [06:28<07:12, 13.95s/trial, best loss: -0.9599469736524371]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 40%|████      | 20/50 [06:49<08:05, 16.17s/trial, best loss: -0.9599469736524371]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 42%|████▏     | 21/50 [06:54<06:13, 12.87s/trial, best loss: -0.9599469736524371]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 44%|████▍     | 22/50 [07:06<05:47, 12.42s/trial, best loss: -0.9599469736524371]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 46%|████▌     | 23/50 [07:17<05:27, 12.14s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 48%|████▊     | 24/50 [07:29<05:10, 11.92s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 50%|█████     | 25/50 [07:40<04:53, 11.73s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 52%|█████▏    | 26/50 [07:52<04:41, 11.73s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 54%|█████▍    | 27/50 [08:02<04:19, 11.27s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 56%|█████▌    | 28/50 [08:15<04:23, 11.96s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 58%|█████▊    | 29/50 [08:26<03:59, 11.42s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 60%|██████    | 30/50 [08:34<03:32, 10.61s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 62%|██████▏   | 31/50 [08:45<03:19, 10.53s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 64%|██████▍   | 32/50 [08:53<02:55,  9.76s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 66%|██████▌   | 33/50 [09:14<03:44, 13.19s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 68%|██████▊   | 34/50 [09:40<04:35, 17.19s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 70%|███████   | 35/50 [10:06<04:54, 19.62s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 72%|███████▏  | 36/50 [10:16<03:54, 16.76s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 74%|███████▍  | 37/50 [10:27<03:18, 15.27s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 76%|███████▌  | 38/50 [10:38<02:45, 13.78s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 78%|███████▊  | 39/50 [11:14<03:45, 20.52s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 80%|████████  | 40/50 [11:30<03:11, 19.14s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 82%|████████▏ | 41/50 [11:49<02:53, 19.24s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 84%|████████▍ | 42/50 [12:21<03:02, 22.87s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 86%|████████▌ | 43/50 [12:33<02:17, 19.60s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 88%|████████▊ | 44/50 [13:00<02:12, 22.03s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 90%|█████████ | 45/50 [13:29<02:00, 24.12s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 92%|█████████▏| 46/50 [13:58<01:41, 25.48s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 94%|█████████▍| 47/50 [14:24<01:16, 25.65s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 96%|█████████▌| 48/50 [14:49<00:50, 25.46s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      " 98%|█████████▊| 49/50 [15:09<00:23, 23.68s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      "100%|██████████| 50/50 [15:24<00:00, 21.02s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\r",
      "100%|██████████| 50/50 [15:24<00:00, 18.48s/trial, best loss: -0.9600838705554923]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    },
    {
     "data": {
      "text/markdown": [
       "**Best Parameters**"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'class_weight': None, 'max_depth': 13, 'max_features': 14, 'min_samples_split': 49, 'n_estimators': 66, 'replacement': 'True', 'sampling_strategy': 'not majority'} \n",
      "\n"
     ]
    },
    {
     "data": {
      "text/markdown": [
       "**Best Macro F1 Score**"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9600838705554923\n"
     ]
    }
   ],
   "source": [
    "# tuning model\n",
    "trials = Trials()\n",
    "best = fmin(fn = objective, space = params, algo = tpe.suggest, max_evals = 50, trials = trials)\n",
    "sampling_strategy = ['majority', 'not minority', 'not majority', 'all']\n",
    "replacement = ['True', 'False']\n",
    "class_weight = [None, 'balanced', 'balanced_subsample']\n",
    "for name in ['sampling_strategy', 'replacement', 'class_weight']:\n",
    "    best[name] = eval(f'{name}[best[name]]')\n",
    "for name in ['n_estimators', 'max_depth', 'min_samples_split', 'max_features']:\n",
    "    best[name] = int(best[name])\n",
    "printmd('**Best Parameters**')\n",
    "print(best, '\\n')\n",
    "printmd('**Best Macro F1 Score**')\n",
    "print(-trials.best_trial['result']['loss'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "This second model and macro f1 score mark an improvement of ~.055 over the previous. Some notes \n",
    "about the best parameter set from our hyperparameter tuning:\n",
    "\n",
    "  1. the chosen combination of parameters varies quite wildly\n",
    "  2. most stable parameters are \"n_estimators\", \"max_depth\", and \"min_samples_split\"\n",
    "          a. \"n_estimators\" is usually between 40 and 60\n",
    "          b. \"max_depth\" is usually below 10\n",
    "          c. \"min_samples_split\" is usually between 15 and 25\n",
    "  3. macro f1 score achieved is usually in the .959 to .960 range\n",
    "\n",
    "Let's generate a classification report so we can compare per class f1 scores."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2021-07-12T20:03:29.754805Z",
     "iopub.status.busy": "2021-07-12T20:03:29.754015Z",
     "iopub.status.idle": "2021-07-12T20:03:41.333484Z",
     "shell.execute_reply": "2021-07-12T20:03:41.334633Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "**Classification Report**"
      ],
      "text/plain": [
       "<IPython.core.display.Markdown object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0     0.9992    0.9936    0.9964    436308\n",
      "           1     0.8704    0.9826    0.9231     19093\n",
      "\n",
      "    accuracy                         0.9931    455401\n",
      "   macro avg     0.9348    0.9881    0.9597    455401\n",
      "weighted avg     0.9938    0.9931    0.9933    455401\n",
      "\n"
     ]
    }
   ],
   "source": [
    "# model report\n",
    "brf = BalancedRandomForestClassifier(**best, n_jobs = -1, random_state = 42)\n",
    "printmd('**Classification Report**')\n",
    "print(model_validation(brf, X_train, y_train))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "## Closing Thoughts"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "region_name": "md"
   },
   "source": [
    "Swapping models resulted in f1 score net-positive changes for both classes, though only minimally \n",
    "so for class \"0\". The gains for class \"1\" were in precision, meaning more correct \"1\" predictions \n",
    "per overall \"1\" predictions; recall did suffer, but only minimally so. In a business context this \n",
    "would means fewer false positives and more false negatives, respectively, but in general better \n",
    "predictive abilities for class \"1\". Depending on client requirements these specific models could \n",
    "be re-weighted, though the end result would be a trade off between each class' predictive power.\n",
    "\n",
    "Further exploration, post-project, would be to investigate more computationally complex models; \n",
    "XGBoost, LightGBM, and Catboost come to mind."
   ]
  }
 ],
 "metadata": {
  "interpreter": {
   "hash": "a011a4f2480411cb5b783cb926da2fb11dad82886afea6ea3593c3d7d2d65e29"
  },
  "jupytext": {
   "formats": "ipynb,py:percent"
  },
  "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": "3.8.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}