XGB-features.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Preamble"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from scipy.stats import describe\n",
    "pd.options.display.max_columns = 12\n",
    "pd.options.display.max_rows = 24\n",
    "\n",
    "# disable warnings in Anaconda\n",
    "import warnings\n",
    "warnings.simplefilter('ignore')\n",
    "\n",
    "# plots inisde jupyter notebook\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "import seaborn as sns\n",
    "sns.set(style='darkgrid', palette='muted')\n",
    "color_scheme = {\n",
    "    'red': '#F1637A',\n",
    "    'green': '#6ABB3E',\n",
    "    'blue': '#3D8DEA',\n",
    "    'black': '#000000'\n",
    "}\n",
    "\n",
    "import datetime as dt\n",
    "import xgboost as xgb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def smape(y_pred, y_true):\n",
    "    # calculate error\n",
    "    denom = (abs(y_pred) + abs(y_true)) / 2\n",
    "    errors = abs(y_pred - y_true) / denom\n",
    "    return 100 * np.sum(errors) / len(y_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "def serie_split(s, fcast_len = 90):\n",
    "    \"\"\"\n",
    "    We split our datasets: year 2017 is used for validation and the rest is for training.\n",
    "    As our goal is to predict the first 90 days of 2018, we use only the first 90 days for validation.\n",
    "    \"\"\"\n",
    "    train = s.iloc[s.index < '2017-01-01']\n",
    "    test = s.iloc[s.index >= '2017-01-01'].iloc[:fcast_len]\n",
    "    \n",
    "    return train, test"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_csv('../data/train.csv')\n",
    "df['date'] = pd.to_datetime(df['date'])\n",
    "df.index = pd.DatetimeIndex(df['date'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feature extraction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from rolling import Rolling"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def compute_over_interval(functions, interval):\n",
    "    return np.array([f(interval) for f in functions])\n",
    "\n",
    "def features(prev_values, current_date, metadata=None):\n",
    "    last_2months = prev_values[-60:]\n",
    "    last_month = prev_values[-30:]\n",
    "    last_week = prev_values[-7:]\n",
    "    \n",
    "    featureset = [\n",
    "        lambda x: np.median(x),\n",
    "        #lambda x: np.mean(x),\n",
    "        #lambda x: np.min(x),\n",
    "        #lambda x: np.max(x),\n",
    "        lambda x: np.var(x),\n",
    "    ]\n",
    "    \n",
    "    weekday_v = np.zeros(7)\n",
    "    weekday_v[current_date.dayofweek] = 1\n",
    "    \n",
    "    return np.concatenate([\n",
    "        #weekday_v,\n",
    "        last_2months,\n",
    "        compute_over_interval(featureset, last_month),\n",
    "        prev_values[:30], # one month almost a year before\n",
    "    ])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def train_and_validate(serie):\n",
    "    r = Rolling(window=365, extract_features=features)\n",
    "\n",
    "    model = xgb.XGBRegressor(n_jobs=-1)\n",
    "    \n",
    "    train, test = serie_split(serie)\n",
    "    train_X, train_y = r.make_training_data(train)\n",
    "\n",
    "    model.fit(train_X, train_y)\n",
    "\n",
    "    y = r.predict(model.predict, train)\n",
    "\n",
    "    return smape(y, test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Run all models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "23.87113765793728\n",
      "16.99160090105278\n",
      "16.141614314794275\n",
      "28.64544802575536\n",
      "26.453506047715074\n",
      "15.990990711570038\n",
      "14.017887875891574\n",
      "12.456157106776926\n",
      "13.245837265091568\n",
      "12.687403477013433\n",
      "11.997499667995571\n",
      "12.336846490887913\n",
      "12.313948453848758\n",
      "14.51198997929609\n",
      "11.671782363569319\n",
      "21.336957734547575\n",
      "17.081446784094602\n",
      "13.579809834786705\n",
      "14.899975363219918\n",
      "14.680308257559899\n",
      "16.899036250421325\n",
      "15.791734558427525\n",
      "17.211455848666738\n",
      "13.718790631921538\n",
      "10.703860694996468\n",
      "14.54913591003008\n",
      "23.49658293051857\n",
      "12.131304941515413\n",
      "14.092751893101362\n",
      "15.70123948218163\n",
      "14.023008009442437\n",
      "17.90507033840503\n",
      "13.133528274251614\n",
      "20.855897924540958\n",
      "10.950528703107484\n",
      "10.494224777943742\n",
      "20.463013334106375\n",
      "12.476130841230422\n",
      "15.923918245805567\n",
      "22.576516722950696\n",
      "23.69048516137382\n",
      "16.344950834713902\n",
      "14.451519875800939\n",
      "19.25797359346662\n",
      "11.45298399368433\n",
      "12.369521804596776\n",
      "22.321977942648832\n",
      "11.557411972591478\n",
      "19.476030293772315\n",
      "13.083247076272398\n",
      "17.13121087936198\n",
      "12.13667540317369\n",
      "13.368218702790463\n",
      "17.047586016195034\n",
      "22.517937297218662\n",
      "10.435155917708512\n",
      "10.776582936759267\n",
      "11.855884116557345\n",
      "11.114365555668504\n",
      "12.199764407963691\n",
      "11.440809790523481\n",
      "11.100593301668535\n",
      "11.198155128271555\n",
      "12.109589756082748\n",
      "10.356547944904461\n",
      "14.80646295819545\n",
      "14.257632648772793\n",
      "8.6513308444192\n",
      "13.17292066855217\n",
      "12.552270733576133\n",
      "12.115588389348984\n",
      "10.842307449344938\n",
      "17.46884946627776\n",
      "13.092736462106958\n",
      "12.290984559541975\n",
      "11.250041140169133\n",
      "16.203003032787763\n",
      "13.180231563942339\n",
      "10.912923949065842\n",
      "13.501286956875104\n",
      "14.388617587968888\n",
      "14.00458471518346\n",
      "11.38261202347833\n",
      "20.40628863834678\n",
      "10.007711217842958\n",
      "9.152971948645867\n",
      "15.452625187815244\n",
      "11.19123525259093\n",
      "13.004616884169488\n",
      "16.595199926340676\n",
      "19.479194819049557\n",
      "13.913773652013052\n",
      "11.995088394428608\n",
      "17.628433929888114\n",
      "11.012331322639701\n",
      "12.71171144993314\n",
      "17.525241979829307\n",
      "11.254314693193647\n",
      "15.793015484923684\n",
      "13.488072398623398\n",
      "14.992590515424114\n",
      "13.588215675219532\n",
      "15.847018478060601\n",
      "19.96351852563633\n",
      "20.74517072709113\n",
      "13.09241682971775\n",
      "11.323054622371313\n",
      "11.458371004410045\n",
      "13.692410619762967\n",
      "10.778263169196977\n",
      "11.234377728351145\n",
      "10.41214053483852\n",
      "11.298310569736854\n",
      "14.30802335439698\n",
      "10.91522576310903\n",
      "16.587409215268337\n",
      "15.739105177565886\n",
      "11.684859296077265\n",
      "13.347847040708231\n",
      "12.477952115653407\n",
      "13.89082928963118\n",
      "10.200756927088035\n",
      "15.749579114144916\n",
      "12.466784763799106\n",
      "12.246724134887321\n",
      "20.12769034205891\n",
      "21.020605218495213\n",
      "10.13356911563521\n",
      "11.748002424858685\n",
      "15.233176177760136\n",
      "11.956233368453347\n",
      "12.905940800703256\n",
      "11.516637215722433\n",
      "18.14092412741941\n",
      "12.145031614918093\n",
      "8.978314276915642\n",
      "16.571169453482625\n",
      "10.713295708539642\n",
      "17.217896032613\n",
      "14.59886777077799\n",
      "20.25422760075955\n",
      "15.731924226507115\n",
      "11.51183058782278\n",
      "18.40977262195443\n",
      "10.950245508381855\n",
      "13.001020637778943\n",
      "23.75876152341625\n",
      "12.638836421187667\n",
      "16.720164126899917\n",
      "10.285712018775534\n",
      "21.030970289175777\n",
      "12.16803192353621\n",
      "13.724731398984053\n",
      "20.47835077644959\n",
      "24.843564212341008\n",
      "12.867274293556296\n",
      "15.082059963114627\n",
      "11.836780957829383\n",
      "12.74637051305049\n",
      "12.48087577729408\n",
      "12.620647507707437\n",
      "13.101967657235766\n",
      "11.536565734137724\n",
      "12.751307361354172\n",
      "10.766815482767152\n",
      "19.852873895317714\n",
      "14.887034698059523\n",
      "11.81027886015163\n",
      "14.720175282771354\n",
      "13.73274021750489\n",
      "15.573137471297496\n",
      "9.856330187495741\n",
      "16.070601459816643\n",
      "11.59236247160353\n",
      "12.771103702472026\n",
      "13.232840080698397\n",
      "22.482554939186482\n",
      "13.634841384244833\n",
      "12.74172206536606\n",
      "17.81035884952235\n",
      "13.005313472201962\n",
      "16.80400124327286\n",
      "11.523701425037224\n",
      "20.5990406107118\n",
      "13.435155049508598\n",
      "13.456303717480985\n",
      "17.46358455792213\n",
      "10.619775634348962\n",
      "14.96956324346667\n",
      "18.948875308893527\n",
      "21.25915218434177\n",
      "17.05976794683174\n",
      "12.74863370414432\n",
      "19.63598891546681\n",
      "10.176283925487455\n",
      "15.024859467779363\n",
      "18.453588223176315\n",
      "13.54781256918899\n",
      "16.613857689689148\n",
      "11.04322758603172\n",
      "23.554388666033553\n",
      "16.586818473342856\n",
      "19.42672943198623\n",
      "28.404701832366108\n",
      "24.790276416398957\n",
      "14.848371362461762\n",
      "17.719007269134497\n",
      "13.73826315900906\n",
      "15.722191542337518\n",
      "10.936683470492493\n",
      "15.361681286147604\n",
      "13.686354622456829\n",
      "12.995619692720384\n",
      "16.29779809799747\n",
      "13.73105041924479\n",
      "25.12786692870503\n",
      "20.180141464648838\n",
      "12.45853133301449\n",
      "16.99744831619319\n",
      "17.444276694625408\n",
      "17.77908360322428\n",
      "12.747352034712337\n",
      "19.760308564049932\n",
      "14.766251459859783\n",
      "11.17437589910694\n",
      "17.397304000862555\n",
      "25.582265175927457\n",
      "9.540540963948484\n",
      "16.37206106179275\n",
      "16.914826564842215\n",
      "18.439648557909678\n",
      "14.238792779557096\n",
      "13.96367360133653\n",
      "20.825691207903898\n",
      "11.289598326529765\n",
      "13.87248964530023\n",
      "17.125098098392932\n",
      "11.749394893222911\n",
      "17.72055519966441\n",
      "24.58680481930025\n",
      "27.271519279827817\n",
      "21.46091009819667\n",
      "16.412933894435557\n",
      "18.82775894110208\n",
      "12.777844079613033\n",
      "15.657519783526297\n",
      "28.081855706229817\n",
      "15.72274318952631\n",
      "18.95181095176833\n",
      "15.315657323355728\n",
      "22.05758975049735\n",
      "15.356703337482308\n",
      "19.670076476252365\n",
      "20.990333454479188\n",
      "24.194166575197304\n",
      "13.378168519898885\n",
      "15.714416061079838\n",
      "12.878215426345502\n",
      "14.44693662917413\n",
      "14.353770620300518\n",
      "15.428740824232776\n",
      "13.868949197933473\n",
      "10.785831605502361\n",
      "15.179868433137566\n",
      "12.246021479870024\n",
      "20.413883658927443\n",
      "20.955006446244028\n",
      "12.95687969203753\n",
      "18.477550822052923\n",
      "16.15106256569919\n",
      "17.055240767716487\n",
      "13.539764698901942\n",
      "22.35648046977383\n",
      "15.171945996342233\n",
      "13.560789943140739\n",
      "16.88959549167191\n",
      "20.907924924241975\n",
      "9.478902812897086\n",
      "15.155591290450143\n",
      "21.538816431237283\n",
      "12.665825328702114\n",
      "20.625474545980637\n",
      "13.31659990126007\n",
      "22.189121399007547\n",
      "14.696859465658198\n",
      "14.089862398868933\n",
      "20.32955574008201\n",
      "12.274346682274883\n",
      "15.266061925222177\n",
      "26.405088729873412\n",
      "24.80125536847212\n",
      "19.963538756538853\n",
      "17.170617715949426\n",
      "20.625275061417447\n",
      "13.989870212374177\n",
      "15.024333946699098\n",
      "23.12561409464823\n",
      "16.335969363998196\n",
      "20.07283742311259\n",
      "13.19575675863355\n",
      "24.087665265060664\n",
      "17.645842383844002\n",
      "16.304972375463652\n",
      "20.885761555127097\n",
      "31.58593318447265\n",
      "15.863329833511742\n",
      "17.295065894241404\n",
      "11.715222604896432\n",
      "14.734274631775442\n",
      "16.406754922637408\n",
      "16.89821806841404\n",
      "12.575677551870148\n",
      "14.22279477740662\n",
      "16.09941631391621\n",
      "12.17685198225884\n",
      "25.07354328378021\n",
      "21.46769426512005\n",
      "13.020715402026733\n",
      "19.1686190993485\n",
      "16.721810185269828\n",
      "18.959276584283188\n",
      "12.866446872664511\n",
      "21.647153642460626\n",
      "13.708761485748472\n",
      "13.377224026458292\n",
      "17.270328255474706\n",
      "25.837692239062683\n",
      "14.82653219212987\n",
      "14.519964376913258\n",
      "16.975486588095738\n",
      "16.294555179111367\n",
      "16.06872129709943\n",
      "12.902436599168459\n",
      "22.542904263765774\n",
      "14.139343767424053\n",
      "14.723766726131135\n",
      "17.998466213452193\n",
      "14.144058199726503\n",
      "20.246044034304003\n",
      "20.56777630809112\n",
      "24.240864615123247\n",
      "21.482230180703226\n",
      "16.905253876943945\n",
      "20.639411374742245\n",
      "15.46281836457442\n",
      "19.367381503611185\n",
      "28.268661773431152\n",
      "17.775207857380256\n",
      "22.54420450347106\n",
      "14.352218477625003\n",
      "17.680939148516366\n",
      "13.786499131443957\n",
      "16.579607654931298\n",
      "21.146746859415607\n",
      "19.819786593261103\n",
      "10.690048712303192\n",
      "11.686828057151988\n",
      "10.508672982282015\n",
      "13.07284720819801\n",
      "11.495923412170871\n",
      "11.101030881000481\n",
      "13.103455685461446\n",
      "9.285744636011113\n",
      "12.247577645877932\n",
      "12.327731968681903\n",
      "17.497588307790146\n",
      "16.43173589547952\n",
      "10.822581366989363\n",
      "15.306727111562592\n",
      "13.657933390937291\n",
      "14.070070841386453\n",
      "12.190052593716945\n",
      "20.66046875236637\n",
      "11.400653787682225\n",
      "12.964833958934186\n",
      "14.071418252504115\n",
      "17.87423281143325\n",
      "9.120533404092718\n",
      "12.479936378867976\n",
      "15.807171545324529\n",
      "11.384942795282363\n",
      "13.406454346699288\n",
      "12.018460921232508\n",
      "16.06756746700191\n",
      "10.428844525227486\n",
      "12.532323272998065\n",
      "16.193290370761268\n",
      "10.266061276622642\n",
      "14.319386958053203\n",
      "21.10764780824413\n",
      "18.41641731171029\n",
      "15.575998246184481\n",
      "13.861322104630888\n",
      "17.45298578708685\n",
      "8.537481686884203\n",
      "11.298221124967542\n",
      "19.321655808416736\n",
      "13.170198294604239\n",
      "16.529855244269168\n",
      "10.94032727420105\n",
      "19.48154457562275\n",
      "14.045420183041744\n",
      "16.20197764143328\n",
      "23.05207604875756\n",
      "22.62245630823433\n",
      "12.590115510502653\n",
      "12.033681449175834\n",
      "11.01620456829503\n",
      "13.209038966949246\n",
      "10.755645982093096\n",
      "12.778838732933583\n",
      "10.796854331186918\n",
      "10.655524890468417\n",
      "11.89030576733016\n",
      "9.745006378652677\n",
      "18.428707584980707\n",
      "18.110043078419963\n",
      "13.708717445581389\n",
      "15.865918791766621\n",
      "14.626778526034878\n",
      "14.482666293656509\n",
      "11.93536044571356\n",
      "16.313409383514095\n",
      "11.865647107110915\n",
      "13.40698322852571\n",
      "13.86687298776843\n",
      "22.632717860136264\n",
      "10.808146793612627\n",
      "12.207685234131011\n",
      "15.079863075168468\n",
      "13.240540905746009\n",
      "13.522876204524321\n",
      "11.570821066306571\n",
      "17.980588834621308\n",
      "13.082322008270676\n",
      "10.822184887615267\n",
      "16.957564357751536\n",
      "14.012660480171403\n",
      "15.551562035333783\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "16.60247320411955\n",
      "20.35714418948094\n",
      "16.196684747066165\n",
      "13.729617039841477\n",
      "20.31487748441642\n",
      "10.302271527138254\n",
      "13.439512183322044\n",
      "21.005290011416697\n",
      "13.450612797022417\n",
      "16.501114507759482\n",
      "12.179044066284076\n",
      "21.556963139298627\n",
      "12.33327321078022\n",
      "17.197581458803608\n",
      "18.979082334131174\n",
      "20.44431146256982\n",
      "12.537804134490816\n",
      "12.922281573790592\n",
      "12.268027217773907\n",
      "10.330956229030264\n",
      "11.425936846481155\n",
      "10.18338957919352\n",
      "12.308412556396458\n",
      "10.56482732403133\n",
      "13.953919730602312\n",
      "10.404180612311524\n",
      "16.883577362962534\n",
      "19.141086411749345\n",
      "11.029219866619513\n",
      "16.45456449090657\n",
      "11.452054181308702\n",
      "16.337782557352266\n",
      "12.429624179847984\n",
      "15.463232503803608\n",
      "12.618158391445041\n",
      "11.520684638919885\n",
      "14.571931802399169\n",
      "17.349889283112354\n",
      "8.456667533909537\n",
      "10.375180853237094\n",
      "15.613341780585433\n",
      "11.83141098177827\n",
      "14.578257162871653\n",
      "11.630192467104441\n",
      "20.65821984344638\n",
      "12.618388826047353\n",
      "11.030055298113528\n",
      "19.357369880038195\n",
      "11.331278223690086\n",
      "14.214754751934601\n",
      "16.09430299783017\n",
      "23.641589899706215\n",
      "15.392940603911075\n",
      "11.352340188028073\n",
      "15.61531035448032\n",
      "12.66681205225313\n",
      "12.924287724785202\n",
      "20.84914822096515\n",
      "13.323548408932878\n",
      "19.07914665710944\n",
      "12.351793563402447\n"
     ]
    }
   ],
   "source": [
    "errors = []\n",
    "for store in range(1,11):\n",
    "    for item in range(1,51):\n",
    "        sales = df[(df[\"store\"] == store) & (df[\"item\"] == item)][\"sales\"]\n",
    "        error = train_and_validate(sales)\n",
    "        print(error)\n",
    "        errors.append(error)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "SMAPE = 15.37532205938642\n"
     ]
    }
   ],
   "source": [
    "print(f\"SMAPE = {np.mean(errors)}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.distplot(errors)\n",
    "plt.savefig('xgboost_features.svg')\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [default]",
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}