time_decay.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from sklearn.preprocessing import MultiLabelBinarizer\n",
    "from sklearn.linear_model import RidgeCV\n",
    "from scipy.sparse import hstack\n",
    "from scipy import stats\n",
    "import optuna\n",
    "\n",
    "stint_16 = pd.read_csv('luck_adjusted_one_way_possessions/luck_adjusted_one_way_possessions_2016-17.csv')\n",
    "stint_17 = pd.read_csv('luck_adjusted_one_way_possessions/luck_adjusted_one_way_possessions_2017-18.csv')\n",
    "\n",
    "stint_all = pd.concat([stint_16,stint_17]).reset_index()\n",
    "stint_all = stint_all[stint_all.possessions > 0]\n",
    "\n",
    "stint_all = stint_all.groupby(['offensePlayer1Id','offensePlayer2Id','offensePlayer3Id','offensePlayer4Id','offensePlayer5Id','defensePlayer1Id','defensePlayer2Id','defensePlayer3Id','defensePlayer4Id','defensePlayer5Id']).agg(expectedPoints=('expectedPoints','sum'),possessions=('possessions','sum'),gameDate=('gameDate','first'),season=('season','first')).reset_index()\n",
    "stint_all['datetime'] = pd.to_datetime(stint_all['gameDate'],unit='ms')\n",
    "stint_all['combined_off_player'] = stint_all[['offensePlayer1Id','offensePlayer2Id','offensePlayer3Id','offensePlayer4Id','offensePlayer5Id']].values.tolist()\n",
    "stint_all['combined_def_player'] = stint_all[['defensePlayer1Id','defensePlayer2Id','defensePlayer3Id','defensePlayer4Id','defensePlayer5Id']].values.tolist()\n",
    "\n",
    "stint_post = stint_all[stint_all.datetime >= stint_all.datetime.mean()]\n",
    "\n",
    "off_mlb = MultiLabelBinarizer(sparse_output=True)\n",
    "off_array = off_mlb.fit_transform(stint_post['combined_off_player'] )\n",
    "def_mlb = MultiLabelBinarizer(sparse_output=True)\n",
    "def_array = def_mlb.fit_transform(stint_post['combined_def_player']) \n",
    "\n",
    "stint_array = hstack([off_array,-def_array])\n",
    "\n",
    "clf = RidgeCV(alphas=[3000])\n",
    "clf.fit(stint_array,100*stint_post['expectedPoints']/stint_post['possessions'],sample_weight=stint_post['possessions'])\n",
    "\n",
    "class_list = ['{}_{}'.format(a, 'offense')for a in off_mlb.classes_] + ['{}_{}'.format(a, 'defense')for a in def_mlb.classes_]\n",
    "\n",
    "out_post_df = pd.DataFrame({'playerId':class_list,'coef':clf.coef_})\n",
    "out_post_df.loc[out_post_df.playerId.str.contains('defense'),'coef'] *= -1\n",
    "\n",
    "stint_pre = stint_all[stint_all.datetime <= stint_all.datetime.mean()]\n",
    "\n",
    "stint_pre['day'] =   (stint_pre['datetime'].max()-stint_pre['datetime'])/ np.timedelta64(1, 'D')\n",
    "\n",
    "off_mlb = MultiLabelBinarizer(sparse_output=True)\n",
    "off_array = off_mlb.fit_transform(stint_pre['combined_off_player'] )\n",
    "def_mlb = MultiLabelBinarizer(sparse_output=True)\n",
    "def_array = def_mlb.fit_transform(stint_pre['combined_def_player']) \n",
    "\n",
    "stint_array = hstack([off_array,-def_array])\n",
    "\n",
    "def objective(trial):\n",
    "    decay = trial.suggest_loguniform('decay', 10, 1000)\n",
    "    clf.fit(stint_array,100*stint_pre['expectedPoints']/stint_pre['possessions'],sample_weight=stint_pre['possessions']*2.71828**(-stint_pre['day']/decay))\n",
    "\n",
    "    class_list = ['{}_{}'.format(a, 'offense')for a in off_mlb.classes_] + ['{}_{}'.format(a, 'defense')for a in def_mlb.classes_]\n",
    "\n",
    "    out_pre_df = pd.DataFrame({'playerId':class_list,'coef':clf.coef_})\n",
    "    out_pre_df.loc[out_pre_df.playerId.str.contains('defense'),'coef'] *= -1\n",
    "\n",
    "    merged_df = pd.merge(out_pre_df, out_post_df,on='playerId')\n",
    "    slope, intercept, r_value, p_value, std_err = stats.linregress(merged_df['coef_x'], merged_df['coef_y'])\n",
    "    \n",
    "    return r_value**2\n",
    "\n",
    "study = optuna.create_study(direction='maximize')\n",
    "study.optimize(objective, n_trials=100)"
   ]
  }
 ],
 "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}