evaluate_treatment_models.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.metrics import roc_auc_score, mean_squared_error, mean_absolute_error"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import sys\n",
    "sys.path.append('/Users/hauptjoh/projects/research/treatment-learn')\n",
    "\n",
    "from treatlearn.policy import bayesian_targeting_policy\n",
    "from treatlearn.evaluation import transformed_outcome_loss, expected_policy_profit"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load Data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Confirm the path to the actual data and the path where the model results are saved."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "DATA_PATH = \"../data/fashionB_clean_nonlinear.csv\"\n",
    "RESULT_PATH = \"20201220-fullRun\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load data\n",
    "X = pd.read_csv(DATA_PATH)\n",
    "\n",
    "c = X.pop('converted').to_numpy()\n",
    "g = X.pop('TREATMENT').to_numpy()\n",
    "y = X.pop('checkoutAmount').to_numpy()\n",
    "tau_conversion = X.pop('TREATMENT_EFFECT_CONVERSION')\n",
    "tau_basket = X.pop('TREATMENT_EFFECT_BASKET')\n",
    "tau_response = X.pop('TREATMENT_EFFECT_RESPONSE')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load model predictions\n",
    "predictions = np.load(f\"../results/{RESULT_PATH}/treatment_model_predictions.npy\", allow_pickle=True)\n",
    "predictions_train = [fold[\"train\"] for fold in predictions]\n",
    "predictions_test = [fold[\"test\"] for fold in predictions]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# for _pred in [predictions_train, predictions_test]:\n",
    "#     for _x in _pred:\n",
    "#         _x['treatment_spending']['xbcf_outcome_gbt'] = _x['treatment_spending']['xbcf']\n",
    "#         del _x['treatment_spending']['xbcf']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Conversion Predictions C(T=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# ##### Comparison on AUC\n",
    "\n",
    "def calc_classification_error(prediction_dict, y_true, g):\n",
    "    \"\"\"\n",
    "    Calculate the prediction error of the model predictions\n",
    "    prediction_dict : dict\n",
    "        Dictionary with the model predictions in the form model_name: array of predictions\n",
    "    y_true : 1D array-like\n",
    "        Observed outcomes\n",
    "    g : 1D array-like\n",
    "        Binary group indicator\n",
    "    prob_treatment : array-like or int\n",
    "        The group propensity for each observation. If None or int, the constant probability\n",
    "        to be in binary treatment group 1.\n",
    "    tau_true : 1D array-like\n",
    "        Array of the true treatment effect. The true treatment effect\n",
    "        is only known in simulations\n",
    "    \"\"\"\n",
    "    output = {}\n",
    "    \n",
    "    for model_name, pred in prediction_dict.items():\n",
    "        output[model_name] = {}\n",
    "        output[model_name][\"ROC-AUC\"] = roc_auc_score(y_true=y_true[g], y_score=pred[g]) \n",
    "        output[model_name][\"brier\"] = mean_squared_error(y_pred=pred[g], y_true=y_true[g])\n",
    "\n",
    "    return output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_conversion = [calc_classification_error(outcome_dict[\"conversion\"], y_true=c[outcome_dict[\"idx\"]], g=np.nonzero(g[outcome_dict[\"idx\"]]))\n",
    "             for outcome_dict in predictions_test]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_conversion = pd.concat([pd.DataFrame(x) for x in eval_conversion], axis=0, keys=range(len(eval_conversion)))\n",
    "eval_conversion.index.rename([\"fold\",\"metric\"], inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_conversion = eval_conversion.groupby(\"metric\").mean().T"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_conversion.index = pd.MultiIndex.from_tuples(eval_conversion.index.str.split(\"_\", expand=True).tolist())\n",
    "eval_conversion = eval_conversion.rename(mapper={\"ROC-AUC\": \"ROC-AUC\", \"brier\": \"Brier Score\"}, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>metric</th>\n",
       "      <th>ROC-AUC</th>\n",
       "      <th>Brier Score</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">single-model</th>\n",
       "      <th rowspan=\"2\" valign=\"top\">outcome</th>\n",
       "      <th>linear</th>\n",
       "      <td>0.654213</td>\n",
       "      <td>0.101727</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>gbt</th>\n",
       "      <td>0.669268</td>\n",
       "      <td>0.100719</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>hurdle</th>\n",
       "      <th>gbt</th>\n",
       "      <td>0.669039</td>\n",
       "      <td>0.100560</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">two-model</th>\n",
       "      <th rowspan=\"2\" valign=\"top\">hurdle</th>\n",
       "      <th>linear</th>\n",
       "      <td>0.654232</td>\n",
       "      <td>0.101720</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>gbt</th>\n",
       "      <td>0.663386</td>\n",
       "      <td>0.101037</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Conversion-Rate</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.105053</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "metric                           ROC-AUC  Brier Score\n",
       "single-model    outcome linear  0.654213     0.101727\n",
       "                        gbt     0.669268     0.100719\n",
       "                hurdle  gbt     0.669039     0.100560\n",
       "two-model       hurdle  linear  0.654232     0.101720\n",
       "                        gbt     0.663386     0.101037\n",
       "Conversion-Rate                 0.500000     0.105053"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "eval_conversion"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_conversion.to_latex(buf=\"../results/conversion_prediction_quality.tex\", float_format=\"%.3f\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Treatment Effect Precision"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# ##### Comparison on transformed outcome loss\n",
    "\n",
    "def calc_prediction_error(prediction_dict, y_true, g, prob_treatment=None, tau_true=None):\n",
    "    \"\"\"\n",
    "    Calculate the prediction error of the model predictions\n",
    "    prediction_dict : dict\n",
    "        Dictionary with the model predictions in the form model_name: array of predictions\n",
    "    y_true : 1D array-like\n",
    "        Observed outcomes\n",
    "    g : 1D array-like\n",
    "        Binary group indicator\n",
    "    prob_treatment : array-like or int\n",
    "        The group propensity for each observation. If None or int, the constant probability\n",
    "        to be in binary treatment group 1.\n",
    "    tau_true : 1D array-like\n",
    "        Array of the true treatment effect. The true treatment effect\n",
    "        is only known in simulations\n",
    "    \"\"\"\n",
    "    output = {}\n",
    "    \n",
    "    if prob_treatment is None:\n",
    "        prob_treatment = g.mean()\n",
    "    \n",
    "    for model_name, pred in prediction_dict.items():\n",
    "        output[model_name] = {}\n",
    "        print(model_name, pred.min().round(0), pred.mean().round(0), pred.max().round(0))\n",
    "        pred = pred.clip(-200,200)\n",
    "        output[model_name][\"transformed_outcome_loss\"] = transformed_outcome_loss(tau_pred=pred, y_true=y_true, g=g, \n",
    "                                                                                  prob_treatment=prob_treatment)\n",
    "        if tau_true is not None:\n",
    "            output[model_name][\"root_mean_squared_error\"] = np.sqrt(mean_squared_error(y_pred=pred, y_true=tau_true))\n",
    "            output[model_name][\"mean_absolute_error\"] = mean_absolute_error(y_pred=pred, y_true=tau_true)\n",
    "\n",
    "    return output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "single-model_outcome_gbt -2.0 4.0 22.0\n",
      "single-model_hurdle_gbt 0.0 4.0 29.0\n",
      "two-model_outcome_linear -1555.0 5.0 49.0\n",
      "two-model_outcome_gbt -10.0 5.0 38.0\n",
      "two-model_hurdle_linear -40.0 5.0 46.0\n",
      "two-model_hurdle_gbt -14.0 5.0 34.0\n",
      "dr_outcome_linear -1528.0 5.0 53.0\n",
      "dr_outcome_gbt -11.0 5.0 41.0\n",
      "xbcf_outcome_xbcf -3.0 5.0 19.0\n",
      "oracle__ -3.0 5.0 50.0\n",
      "ATE__ 5.0 5.0 5.0\n",
      "single-model_outcome_gbt -5.0 4.0 36.0\n",
      "single-model_hurdle_gbt -6.0 4.0 32.0\n",
      "two-model_outcome_linear -414.0 5.0 79.0\n",
      "two-model_outcome_gbt -10.0 5.0 31.0\n",
      "two-model_hurdle_linear -21.0 5.0 191.0\n",
      "two-model_hurdle_gbt -17.0 5.0 56.0\n",
      "dr_outcome_linear -418.0 5.0 88.0\n",
      "dr_outcome_gbt -381.0 5.0 327.0\n",
      "xbcf_outcome_xbcf -3.0 5.0 22.0\n",
      "oracle__ -4.0 5.0 40.0\n",
      "ATE__ 5.0 5.0 5.0\n",
      "single-model_outcome_gbt -5.0 4.0 34.0\n",
      "single-model_hurdle_gbt -4.0 4.0 35.0\n",
      "two-model_outcome_linear -227.0 5.0 97.0\n",
      "two-model_outcome_gbt -8.0 5.0 37.0\n",
      "two-model_hurdle_linear -58.0 5.0 103.0\n",
      "two-model_hurdle_gbt -10.0 5.0 29.0\n",
      "dr_outcome_linear -237.0 5.0 111.0\n",
      "dr_outcome_gbt -7.0 5.0 53.0\n",
      "xbcf_outcome_xbcf -2.0 5.0 22.0\n",
      "oracle__ -4.0 5.0 48.0\n",
      "ATE__ 5.0 5.0 5.0\n",
      "single-model_outcome_gbt 0.0 4.0 17.0\n",
      "single-model_hurdle_gbt -6.0 4.0 29.0\n",
      "two-model_outcome_linear -17.0 5.0 134.0\n",
      "two-model_outcome_gbt -9.0 5.0 36.0\n",
      "two-model_hurdle_linear -311.0 5.0 575.0\n",
      "two-model_hurdle_gbt -26.0 5.0 49.0\n",
      "dr_outcome_linear -22.0 5.0 143.0\n",
      "dr_outcome_gbt -3.0 5.0 46.0\n",
      "xbcf_outcome_xbcf -2.0 5.0 23.0\n",
      "oracle__ -4.0 5.0 47.0\n",
      "ATE__ 5.0 5.0 5.0\n",
      "single-model_outcome_gbt -6.0 4.0 32.0\n",
      "single-model_hurdle_gbt -13.0 4.0 39.0\n",
      "two-model_outcome_linear -12.0 5.0 175.0\n",
      "two-model_outcome_gbt -18.0 5.0 40.0\n",
      "two-model_hurdle_linear -47.0 5.0 161.0\n",
      "two-model_hurdle_gbt -22.0 5.0 44.0\n",
      "dr_outcome_linear -13.0 5.0 180.0\n",
      "dr_outcome_gbt -31.0 5.0 43.0\n",
      "xbcf_outcome_xbcf -2.0 5.0 20.0\n",
      "oracle__ -3.0 5.0 118.0\n",
      "ATE__ 5.0 5.0 5.0\n"
     ]
    }
   ],
   "source": [
    "eval_test = [calc_prediction_error(outcome_dict[\"treatment_spending\"], \n",
    "                                   y[outcome_dict[\"idx\"]], g[outcome_dict[\"idx\"]], tau_true=tau_response[outcome_dict[\"idx\"]]) \n",
    "             for outcome_dict in predictions_test]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_test_dataframe = pd.concat([pd.DataFrame(x) for x in eval_test], axis=0, keys=range(len(eval_test)))\n",
    "eval_test_dataframe.index.rename([\"fold\",\"metric\"], inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_precision = eval_test_dataframe.groupby(\"metric\").mean().T\n",
    "#print(eval_test_dataframe.groupby(\"metric\").std().T)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_precision.index = pd.MultiIndex.from_tuples(eval_precision.index.str.split(\"_\", expand=True).tolist())\n",
    "eval_precision = eval_precision.rename(mapper={\"transformed_outcome_loss\": \"TOL\", \"root_mean_squared_error\": \"RMSE\", \"mean_absolute_error\": \"MAE\"}, axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>metric</th>\n",
       "      <th>TOL</th>\n",
       "      <th>RMSE</th>\n",
       "      <th>MAE</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">single-model</th>\n",
       "      <th>outcome</th>\n",
       "      <th>gbt</th>\n",
       "      <td>4180.37</td>\n",
       "      <td>3.11</td>\n",
       "      <td>2.32</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>hurdle</th>\n",
       "      <th>gbt</th>\n",
       "      <td>4180.13</td>\n",
       "      <td>3.08</td>\n",
       "      <td>2.29</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"4\" valign=\"top\">two-model</th>\n",
       "      <th rowspan=\"2\" valign=\"top\">outcome</th>\n",
       "      <th>linear</th>\n",
       "      <td>4183.06</td>\n",
       "      <td>3.44</td>\n",
       "      <td>2.35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>gbt</th>\n",
       "      <td>4181.10</td>\n",
       "      <td>3.39</td>\n",
       "      <td>2.44</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">hurdle</th>\n",
       "      <th>linear</th>\n",
       "      <td>4186.06</td>\n",
       "      <td>3.83</td>\n",
       "      <td>2.53</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>gbt</th>\n",
       "      <td>4183.05</td>\n",
       "      <td>3.40</td>\n",
       "      <td>2.45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">dr</th>\n",
       "      <th rowspan=\"2\" valign=\"top\">outcome</th>\n",
       "      <th>linear</th>\n",
       "      <td>4183.31</td>\n",
       "      <td>3.48</td>\n",
       "      <td>2.35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>gbt</th>\n",
       "      <td>4181.76</td>\n",
       "      <td>3.26</td>\n",
       "      <td>2.23</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>xbcf</th>\n",
       "      <th>outcome</th>\n",
       "      <th>xbcf</th>\n",
       "      <td>4177.79</td>\n",
       "      <td>2.71</td>\n",
       "      <td>1.92</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>oracle</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <td>4164.46</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ATE</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <td>4186.27</td>\n",
       "      <td>3.77</td>\n",
       "      <td>3.05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "metric                           TOL  RMSE   MAE\n",
       "single-model outcome gbt     4180.37  3.11  2.32\n",
       "             hurdle  gbt     4180.13  3.08  2.29\n",
       "two-model    outcome linear  4183.06  3.44  2.35\n",
       "                     gbt     4181.10  3.39  2.44\n",
       "             hurdle  linear  4186.06  3.83  2.53\n",
       "                     gbt     4183.05  3.40  2.45\n",
       "dr           outcome linear  4183.31  3.48  2.35\n",
       "                     gbt     4181.76  3.26  2.23\n",
       "xbcf         outcome xbcf    4177.79  2.71  1.92\n",
       "oracle                       4164.46  0.00  0.00\n",
       "ATE                          4186.27  3.77  3.05"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "eval_precision.round(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_precision[[\"RMSE\",\"TOL\"]].to_latex(buf=\"../results/treatment_prediction_quality.tex\", float_format=\"%.2f\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "SpearmanrResult(correlation=0.9818181818181818, pvalue=8.403066433955246e-08)"
      ]
     },
     "execution_count": 72,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import scipy.stats as stats\n",
    "stats.spearmanr(eval_precision[[\"RMSE\"]], eval_precision[[\"TOL\"]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 9,  8,  1,  0,  7,  3,  5,  2,  6, 10,  4])"
      ]
     },
     "execution_count": 75,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.argsort(eval_precision[[\"RMSE\"]].values.flatten())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 9,  8,  1,  0,  3,  7,  5,  2,  6,  4, 10])"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.argsort(eval_precision[[\"TOL\"]].values.flatten())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Profit Comparison"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "MARGIN_RATIO = 0.3\n",
    "OFFER_COST = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# # #### Comparison in terms of profit\n",
    "\n",
    "# treatment_conversion_model_map = {\n",
    "#         \"ATE_logit\":('ATE','logit'), \"oracle_gbt\":('oracle','gbt'), \"oracle_gbt\":('oracle','logit')\n",
    "#         \"dr_reg-logit\":('dr_reg','logit'), \"dr_gbt-gbt\":('dr_gbt','gbt'), \n",
    "#         # Two-Model\n",
    "#         \"two_model_rf-rf\":('two_model_rf','rf'), \"two_model_gbt-gbt\":('two_model_gbt','gbt'), \"two_model_reg-logit\":('two_model_reg','logit'),\n",
    "#         \"two_model_hurdle_rf\":('two_model_hurdle_rf','two_model_hurdle_rf'), \"two_model_hurdle_gbt\":('two_model_hurdle_gbt','two_model_hurdle_gbt'), \n",
    "#         \"two_model_hurdle_linear\":('two_model_hurdle_linear','two_model_hurdle_linear'),\n",
    "#         # Single model\n",
    "#         #\"single_model_bart-gbt\":('single_model_bart','gbt'),\n",
    "#         \"single_model_gbt-gbt\":('single_model_gbt','gbt'), \"single_model_rf-rf\":('single_model_gbt','rf'),\n",
    "#         \"single_model_hurdle_rf*2\":('hurdle_rf','hurdle_rf'), \"single_model_hurdle_gbt*2\":('hurdle_gbt','hurdle_gbt')\n",
    "# }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "def tune_threshold(treatment_dict, y_true, c_true, g, margin, contact_cost, offer_cost, prob_treatment=None):\n",
    "    if prob_treatment is None:\n",
    "        prob_treatment = g.mean()\n",
    "    \n",
    "    # Threshold candiates [1, 0.975,...,0]\n",
    "    step_size=25\n",
    "    quantiles = np.array(range(1000,-1,-step_size))/1000\n",
    "    \n",
    "    threshold_dict = {}\n",
    "    for treatment_model, treatment_pred in treatment_dict.items():\n",
    "        quantile_candidates = np.quantile(treatment_pred, quantiles)\n",
    "\n",
    "        best_profit = -np.inf\n",
    "        best_threshold = None\n",
    "        for threshold in quantile_candidates:\n",
    "            decision = (treatment_pred>threshold)*1\n",
    "            profit = expected_policy_profit(targeting_decision=decision, g=g, observed_profit= (y_true*margin-(offer_cost*decision*c_true)), prob_treatment=prob_treatment)\n",
    "            if profit > best_profit:\n",
    "                best_threshold = threshold\n",
    "                best_profit = profit\n",
    "        \n",
    "        threshold_dict[treatment_model] = best_threshold\n",
    "    \n",
    "    return threshold_dict\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calc_bayesian_policy(treatment_dict, conversion_dict, margin, contact_cost, offer_cost):\n",
    "    \n",
    "    policy={}\n",
    "    # Calculate targeting threshold according to expected value\n",
    "    for treatment_model, treatment_pred in treatment_dict.items():\n",
    "        for conversion_model, conversion_pred in conversion_dict.items():\n",
    "            policy[\"Bayesian+\"+str(treatment_model)+\"+\"+str(conversion_model)] = bayesian_targeting_policy(\n",
    "                                                         tau_pred=treatment_pred * margin, \n",
    "                                                         offer_accept_prob=conversion_pred,\n",
    "                                                         contact_cost=contact_cost, offer_cost=offer_cost\n",
    "                                                    )\n",
    "    return policy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calc_naive_policy(treatment_dict):\n",
    "    policy={}\n",
    "    # Calculate targeting threshold according to expected value\n",
    "    for treatment_model, treatment_pred in treatment_dict.items():\n",
    "        policy[\"Treat-all+\"] = np.ones(treatment_pred.shape[0], dtype=\"int\")\n",
    "        policy[\"Treat-none+\"] = np.zeros(treatment_pred.shape[0], dtype=\"int\")\n",
    "    \n",
    "    return policy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calc_threshold_policy(treatment_dict, threshold=0):\n",
    "    policy={}\n",
    "    # Calculate targeting threshold according to expected value\n",
    "    try:\n",
    "        for treatment_model, treatment_pred in treatment_dict.items():\n",
    "            # policy[\"Filter+\"+str(threshold[treatment_model].round(2))+\"+\"+str(treatment_model)]\n",
    "            policy[\"Threshold+\"+str(treatment_model)] = (treatment_pred>=threshold[treatment_model])*1       \n",
    "    except:\n",
    "        for treatment_model, treatment_pred in treatment_dict.items():\n",
    "            policy[\"Threshold\"+str(threshold)+\"+\"+str(treatment_model)] = (treatment_pred>=threshold)*1\n",
    "\n",
    "    return policy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "def calc_policy_profit(policy_dict, y_true, c_true, g, margin, contact_cost, offer_cost, prob_treatment=None):\n",
    "    if prob_treatment is None:\n",
    "        prob_treatment = g.mean()\n",
    "        \n",
    "    profit = {key:expected_policy_profit(targeting_decision=decision, g=g, observed_profit= (y_true*margin-(offer_cost*decision*c_true)), prob_treatment=prob_treatment).round(0) \n",
    "              for key,decision in policy_dict.items()}\n",
    "    ratio_treated = {key:decision.mean().round(2)\n",
    "              for key,decision in policy_dict.items()}\n",
    "    \n",
    "    return {\"profit\":profit, \"ratio_treated\":ratio_treated}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_profit = []\n",
    "for prediction_dict in predictions:\n",
    "    outcome_dict_train = prediction_dict[\"train\"]    \n",
    "    outcome_dict = prediction_dict[\"test\"]\n",
    "    \n",
    "    # Calculate policy decision\n",
    "    policy_dict = calc_bayesian_policy(\n",
    "        treatment_dict = outcome_dict[\"treatment_spending\"], \n",
    "        conversion_dict = outcome_dict[\"conversion\"],\n",
    "    margin =MARGIN_RATIO, contact_cost=0, offer_cost=OFFER_COST) \n",
    "    \n",
    "    policy_dict.update(calc_naive_policy(treatment_dict=outcome_dict[\"treatment_spending\"]))\n",
    "    policy_dict.update(calc_threshold_policy(treatment_dict=outcome_dict[\"treatment_spending\"]))\n",
    "    \n",
    "    # Threshold policy\n",
    "    # Tune threshold \n",
    "    threshold_dict = tune_threshold(outcome_dict_train[\"treatment_spending\"],        \n",
    "                   y_true = y[outcome_dict_train[\"idx\"]], c_true = c[outcome_dict_train[\"idx\"]], \n",
    "                   g=g[outcome_dict_train[\"idx\"]],\n",
    "                   margin =MARGIN_RATIO, contact_cost=0, offer_cost=OFFER_COST)\n",
    "    # Make policy\n",
    "    policy_dict.update(calc_threshold_policy(treatment_dict=outcome_dict[\"treatment_spending\"], \n",
    "                      threshold = threshold_dict))\n",
    "    \n",
    "    # Calculate profit\n",
    "    profit = calc_policy_profit(\n",
    "        policy_dict=policy_dict,\n",
    "        y_true = y[outcome_dict[\"idx\"]], c_true = c[outcome_dict[\"idx\"]], g=g[outcome_dict[\"idx\"]],\n",
    "        margin =MARGIN_RATIO, contact_cost=0, offer_cost=OFFER_COST)\n",
    "    \n",
    "    eval_profit.append(profit)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_profit = pd.concat([pd.DataFrame(x) for x in eval_profit], axis=0, keys=range(len(eval_profit)))\n",
    "eval_profit.index.rename([\"fold\",\"model\"], inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "#eval_profit = eval_profit.groupby(\"model\").agg([np.mean, 'sem'])\n",
    "eval_profit = eval_profit.groupby(\"model\").mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_profit.index = pd.MultiIndex.from_tuples(eval_profit.index.str.split(\"[+|_]\", expand=True).tolist())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_profit.index.names = [\"Policy\",\"Causal\",\"Stages\",\"Estimator\",\"conversion_architecture\",\"conversion_stages\",\"conversion_estimator\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_profit.reset_index(drop=False, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "#eval_profit.columns = ['_'.join(filter(None,[x,y])) for x,y in list(eval_profit.columns)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "#eval_profit['profit_sem'] = eval_profit['profit_sem'].round(0)\n",
    "#eval_profit['ratio_treated_sem'] = eval_profit['ratio_treated_sem'].round(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_profit = eval_profit.reindex(columns= [\"Policy\",\"Stages\",\"Causal\",\"Estimator\",\n",
    "                                            \"conversion_stages\",\"conversion_architecture\",\"conversion_estimator\",\n",
    "                                           \"profit\",\"ratio_treated\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "# eval_profit = eval_profit.reindex(columns= [\"Policy\",\"Stages\",\"Causal\",\"Estimator\",\n",
    "#                                             \"conversion_stages\",\"conversion_architecture\",\"conversion_estimator\",\n",
    "#                                            \"profit_mean\",\"profit_sem\",\"ratio_treated_mean\",\"ratio_treated_sem\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_profit.sort_values([\"Policy\",\"Stages\",\"Causal\",\"Estimator\", \"conversion_architecture\", \"conversion_estimator\"],\n",
    "                        ascending=[True, True, True, False, True, False], inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Conversion Tables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "temp = eval_profit.query(('(Policy in [\"Bayesian\",\"Threshold0\"] and Causal in [\"oracle\",\"ATE\"]) and conversion_stages!=\"outcome\"'\n",
    "                         'or (Policy in [\"Treat-all\", \"Treat-none\"])'),\n",
    "                  inplace=False).drop([\"Stages\", \"Estimator\",\"conversion_stages\"], axis=1)#.sort_values([\"Policy\",\"Causal\", \"conversion_architecture\", \"conversion_estimator\"])\n",
    "\n",
    "temp.to_latex(\"../results/profit_oracle_conversion_models.tex\", na_rep=\"\", index=False,formatters={'profit':'{:.0f}'.format,\"ratio_treated\":'{:.2f}'.format})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Policy</th>\n",
       "      <th>Causal</th>\n",
       "      <th>conversion_architecture</th>\n",
       "      <th>conversion_estimator</th>\n",
       "      <th>profit</th>\n",
       "      <th>ratio_treated</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>65</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>ATE</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td></td>\n",
       "      <td>55055.0</td>\n",
       "      <td>1.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>62</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>ATE</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>55140.8</td>\n",
       "      <td>0.728</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>63</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>ATE</td>\n",
       "      <td>two-model</td>\n",
       "      <td>linear</td>\n",
       "      <td>54097.0</td>\n",
       "      <td>0.726</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>64</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>ATE</td>\n",
       "      <td>two-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>54949.8</td>\n",
       "      <td>0.740</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>59</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>oracle</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td></td>\n",
       "      <td>60124.8</td>\n",
       "      <td>0.558</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>56</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>oracle</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>61346.6</td>\n",
       "      <td>0.578</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>57</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>oracle</td>\n",
       "      <td>two-model</td>\n",
       "      <td>linear</td>\n",
       "      <td>61162.6</td>\n",
       "      <td>0.572</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>58</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>oracle</td>\n",
       "      <td>two-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>61648.2</td>\n",
       "      <td>0.572</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>78</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>ATE</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55055.0</td>\n",
       "      <td>1.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>77</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>oracle</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55794.6</td>\n",
       "      <td>0.946</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>66</th>\n",
       "      <td>Treat-all</td>\n",
       "      <td></td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55055.0</td>\n",
       "      <td>1.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>67</th>\n",
       "      <td>Treat-none</td>\n",
       "      <td></td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>50668.2</td>\n",
       "      <td>0.000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        Policy  Causal conversion_architecture conversion_estimator   profit  \\\n",
       "65    Bayesian     ATE         Conversion-Rate                       55055.0   \n",
       "62    Bayesian     ATE            single-model                  gbt  55140.8   \n",
       "63    Bayesian     ATE               two-model               linear  54097.0   \n",
       "64    Bayesian     ATE               two-model                  gbt  54949.8   \n",
       "59    Bayesian  oracle         Conversion-Rate                       60124.8   \n",
       "56    Bayesian  oracle            single-model                  gbt  61346.6   \n",
       "57    Bayesian  oracle               two-model               linear  61162.6   \n",
       "58    Bayesian  oracle               two-model                  gbt  61648.2   \n",
       "78  Threshold0     ATE                     NaN                  NaN  55055.0   \n",
       "77  Threshold0  oracle                     NaN                  NaN  55794.6   \n",
       "66   Treat-all                             NaN                  NaN  55055.0   \n",
       "67  Treat-none                             NaN                  NaN  50668.2   \n",
       "\n",
       "    ratio_treated  \n",
       "65          1.000  \n",
       "62          0.728  \n",
       "63          0.726  \n",
       "64          0.740  \n",
       "59          0.558  \n",
       "56          0.578  \n",
       "57          0.572  \n",
       "58          0.572  \n",
       "78          1.000  \n",
       "77          0.946  \n",
       "66          1.000  \n",
       "67          0.000  "
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "temp"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Treatment Effect Architectures"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Average conversion rate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "temp = eval_profit.query('(conversion_architecture==\"Conversion-Rate\") or (Policy==\"Threshold0\")',\n",
    "                  inplace=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "temp.drop([\"conversion_stages\", \"conversion_estimator\"], axis=1).to_latex(\"../results/profit_treatment-models_avg-conversion.tex\", na_rep=\"-\", index=False,formatters={'profit':'{:.0f}'.format,\"ratio_treated\":'{:.2f}'.format})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Policy</th>\n",
       "      <th>Stages</th>\n",
       "      <th>Causal</th>\n",
       "      <th>Estimator</th>\n",
       "      <th>conversion_architecture</th>\n",
       "      <th>profit</th>\n",
       "      <th>ratio_treated</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>65</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td></td>\n",
       "      <td>ATE</td>\n",
       "      <td></td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>55055.0</td>\n",
       "      <td>1.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>59</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td></td>\n",
       "      <td>oracle</td>\n",
       "      <td></td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>60124.8</td>\n",
       "      <td>0.558</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>56212.8</td>\n",
       "      <td>0.460</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>two-model</td>\n",
       "      <td>linear</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>56928.2</td>\n",
       "      <td>0.512</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>35</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>two-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>57870.8</td>\n",
       "      <td>0.522</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>41</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>dr</td>\n",
       "      <td>linear</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>56445.6</td>\n",
       "      <td>0.552</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>47</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>dr</td>\n",
       "      <td>gbt</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>57972.4</td>\n",
       "      <td>0.552</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>57584.8</td>\n",
       "      <td>0.494</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>two-model</td>\n",
       "      <td>linear</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>56449.2</td>\n",
       "      <td>0.552</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>two-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>57853.6</td>\n",
       "      <td>0.536</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>53</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>xbcf</td>\n",
       "      <td>xbcf</td>\n",
       "      <td>Conversion-Rate</td>\n",
       "      <td>57578.2</td>\n",
       "      <td>0.550</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>78</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td></td>\n",
       "      <td>ATE</td>\n",
       "      <td></td>\n",
       "      <td>NaN</td>\n",
       "      <td>55055.0</td>\n",
       "      <td>1.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>77</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td></td>\n",
       "      <td>oracle</td>\n",
       "      <td></td>\n",
       "      <td>NaN</td>\n",
       "      <td>55794.6</td>\n",
       "      <td>0.946</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>69</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55031.2</td>\n",
       "      <td>1.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>72</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>two-model</td>\n",
       "      <td>linear</td>\n",
       "      <td>NaN</td>\n",
       "      <td>54750.8</td>\n",
       "      <td>0.974</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>73</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>two-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55299.8</td>\n",
       "      <td>0.972</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>74</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>outcome</td>\n",
       "      <td>dr</td>\n",
       "      <td>linear</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55448.2</td>\n",
       "      <td>0.948</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>outcome</td>\n",
       "      <td>dr</td>\n",
       "      <td>gbt</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55138.2</td>\n",
       "      <td>0.998</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>68</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>outcome</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>NaN</td>\n",
       "      <td>54961.0</td>\n",
       "      <td>0.998</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>70</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>outcome</td>\n",
       "      <td>two-model</td>\n",
       "      <td>linear</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55431.6</td>\n",
       "      <td>0.950</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>71</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>outcome</td>\n",
       "      <td>two-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55660.2</td>\n",
       "      <td>0.964</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>76</th>\n",
       "      <td>Threshold0</td>\n",
       "      <td>outcome</td>\n",
       "      <td>xbcf</td>\n",
       "      <td>xbcf</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55181.2</td>\n",
       "      <td>0.988</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        Policy   Stages        Causal Estimator conversion_architecture  \\\n",
       "65    Bayesian                    ATE                   Conversion-Rate   \n",
       "59    Bayesian                 oracle                   Conversion-Rate   \n",
       "11    Bayesian   hurdle  single-model       gbt         Conversion-Rate   \n",
       "29    Bayesian   hurdle     two-model    linear         Conversion-Rate   \n",
       "35    Bayesian   hurdle     two-model       gbt         Conversion-Rate   \n",
       "41    Bayesian  outcome            dr    linear         Conversion-Rate   \n",
       "47    Bayesian  outcome            dr       gbt         Conversion-Rate   \n",
       "5     Bayesian  outcome  single-model       gbt         Conversion-Rate   \n",
       "17    Bayesian  outcome     two-model    linear         Conversion-Rate   \n",
       "23    Bayesian  outcome     two-model       gbt         Conversion-Rate   \n",
       "53    Bayesian  outcome          xbcf      xbcf         Conversion-Rate   \n",
       "78  Threshold0                    ATE                               NaN   \n",
       "77  Threshold0                 oracle                               NaN   \n",
       "69  Threshold0   hurdle  single-model       gbt                     NaN   \n",
       "72  Threshold0   hurdle     two-model    linear                     NaN   \n",
       "73  Threshold0   hurdle     two-model       gbt                     NaN   \n",
       "74  Threshold0  outcome            dr    linear                     NaN   \n",
       "75  Threshold0  outcome            dr       gbt                     NaN   \n",
       "68  Threshold0  outcome  single-model       gbt                     NaN   \n",
       "70  Threshold0  outcome     two-model    linear                     NaN   \n",
       "71  Threshold0  outcome     two-model       gbt                     NaN   \n",
       "76  Threshold0  outcome          xbcf      xbcf                     NaN   \n",
       "\n",
       "     profit  ratio_treated  \n",
       "65  55055.0          1.000  \n",
       "59  60124.8          0.558  \n",
       "11  56212.8          0.460  \n",
       "29  56928.2          0.512  \n",
       "35  57870.8          0.522  \n",
       "41  56445.6          0.552  \n",
       "47  57972.4          0.552  \n",
       "5   57584.8          0.494  \n",
       "17  56449.2          0.552  \n",
       "23  57853.6          0.536  \n",
       "53  57578.2          0.550  \n",
       "78  55055.0          1.000  \n",
       "77  55794.6          0.946  \n",
       "69  55031.2          1.000  \n",
       "72  54750.8          0.974  \n",
       "73  55299.8          0.972  \n",
       "74  55448.2          0.948  \n",
       "75  55138.2          0.998  \n",
       "68  54961.0          0.998  \n",
       "70  55431.6          0.950  \n",
       "71  55660.2          0.964  \n",
       "76  55181.2          0.988  "
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "temp.drop([\"conversion_stages\", \"conversion_estimator\"], axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Model-based conversion rate"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "temp = eval_profit.query(('(Causal==conversion_architecture and Stages==conversion_stages and Estimator==conversion_estimator) or'\n",
    "                   '(Policy==\"Threshold\") or'\n",
    "                   '(Causal==\"dr\" and conversion_stages!=\"hurdle\" and Estimator==conversion_estimator) or'\n",
    "                   '(Causal==\"xbcf\" and conversion_stages!=\"hurdle\" and conversion_architecture==\"single-model\") or'\n",
    "                   '(Stages==\"outcome\" and conversion_stages==\"outcome\" and Estimator==conversion_estimator) or'\n",
    "                   '(Stages==\"outcome\" and conversion_stages==\"outcome\" and Estimator==\"reg\" and conversion_estimator==\"linear\")'),\n",
    "                  inplace=False)#.sort_values([\"Policy\",\"Stages\", \"Causal\", \"Estimator\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Policy</th>\n",
       "      <th>Stages</th>\n",
       "      <th>Causal</th>\n",
       "      <th>conversion_architecture</th>\n",
       "      <th>conversion_estimator</th>\n",
       "      <th>profit</th>\n",
       "      <th>ratio_treated</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>single-model</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>60456.6</td>\n",
       "      <td>0.616</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>two-model</td>\n",
       "      <td>two-model</td>\n",
       "      <td>linear</td>\n",
       "      <td>58673.0</td>\n",
       "      <td>0.650</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>34</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>two-model</td>\n",
       "      <td>two-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>58526.4</td>\n",
       "      <td>0.664</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>36</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>dr</td>\n",
       "      <td>single-model</td>\n",
       "      <td>linear</td>\n",
       "      <td>57717.8</td>\n",
       "      <td>0.580</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>43</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>dr</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>58075.6</td>\n",
       "      <td>0.708</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>single-model</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>57899.2</td>\n",
       "      <td>0.608</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>two-model</td>\n",
       "      <td>single-model</td>\n",
       "      <td>linear</td>\n",
       "      <td>57736.8</td>\n",
       "      <td>0.580</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>two-model</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>58686.2</td>\n",
       "      <td>0.624</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>48</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>xbcf</td>\n",
       "      <td>single-model</td>\n",
       "      <td>linear</td>\n",
       "      <td>58696.2</td>\n",
       "      <td>0.590</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>49</th>\n",
       "      <td>Bayesian</td>\n",
       "      <td>outcome</td>\n",
       "      <td>xbcf</td>\n",
       "      <td>single-model</td>\n",
       "      <td>gbt</td>\n",
       "      <td>58955.8</td>\n",
       "      <td>0.624</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>89</th>\n",
       "      <td>Threshold</td>\n",
       "      <td></td>\n",
       "      <td>ATE</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>55055.0</td>\n",
       "      <td>1.000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>88</th>\n",
       "      <td>Threshold</td>\n",
       "      <td></td>\n",
       "      <td>oracle</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>59942.6</td>\n",
       "      <td>0.520</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>80</th>\n",
       "      <td>Threshold</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>single-model</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>56603.4</td>\n",
       "      <td>0.536</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>83</th>\n",
       "      <td>Threshold</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>two-model</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>57349.4</td>\n",
       "      <td>0.634</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>84</th>\n",
       "      <td>Threshold</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>two-model</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>57631.8</td>\n",
       "      <td>0.552</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>85</th>\n",
       "      <td>Threshold</td>\n",
       "      <td>outcome</td>\n",
       "      <td>dr</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>56620.6</td>\n",
       "      <td>0.584</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>86</th>\n",
       "      <td>Threshold</td>\n",
       "      <td>outcome</td>\n",
       "      <td>dr</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>57334.4</td>\n",
       "      <td>0.456</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>79</th>\n",
       "      <td>Threshold</td>\n",
       "      <td>outcome</td>\n",
       "      <td>single-model</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>57401.8</td>\n",
       "      <td>0.544</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>81</th>\n",
       "      <td>Threshold</td>\n",
       "      <td>outcome</td>\n",
       "      <td>two-model</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>56854.6</td>\n",
       "      <td>0.616</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>82</th>\n",
       "      <td>Threshold</td>\n",
       "      <td>outcome</td>\n",
       "      <td>two-model</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>57767.0</td>\n",
       "      <td>0.524</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>87</th>\n",
       "      <td>Threshold</td>\n",
       "      <td>outcome</td>\n",
       "      <td>xbcf</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>57261.6</td>\n",
       "      <td>0.564</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       Policy   Stages        Causal conversion_architecture  \\\n",
       "8    Bayesian   hurdle  single-model            single-model   \n",
       "27   Bayesian   hurdle     two-model               two-model   \n",
       "34   Bayesian   hurdle     two-model               two-model   \n",
       "36   Bayesian  outcome            dr            single-model   \n",
       "43   Bayesian  outcome            dr            single-model   \n",
       "1    Bayesian  outcome  single-model            single-model   \n",
       "12   Bayesian  outcome     two-model            single-model   \n",
       "19   Bayesian  outcome     two-model            single-model   \n",
       "48   Bayesian  outcome          xbcf            single-model   \n",
       "49   Bayesian  outcome          xbcf            single-model   \n",
       "89  Threshold                    ATE                     NaN   \n",
       "88  Threshold                 oracle                     NaN   \n",
       "80  Threshold   hurdle  single-model                     NaN   \n",
       "83  Threshold   hurdle     two-model                     NaN   \n",
       "84  Threshold   hurdle     two-model                     NaN   \n",
       "85  Threshold  outcome            dr                     NaN   \n",
       "86  Threshold  outcome            dr                     NaN   \n",
       "79  Threshold  outcome  single-model                     NaN   \n",
       "81  Threshold  outcome     two-model                     NaN   \n",
       "82  Threshold  outcome     two-model                     NaN   \n",
       "87  Threshold  outcome          xbcf                     NaN   \n",
       "\n",
       "   conversion_estimator   profit  ratio_treated  \n",
       "8                   gbt  60456.6          0.616  \n",
       "27               linear  58673.0          0.650  \n",
       "34                  gbt  58526.4          0.664  \n",
       "36               linear  57717.8          0.580  \n",
       "43                  gbt  58075.6          0.708  \n",
       "1                   gbt  57899.2          0.608  \n",
       "12               linear  57736.8          0.580  \n",
       "19                  gbt  58686.2          0.624  \n",
       "48               linear  58696.2          0.590  \n",
       "49                  gbt  58955.8          0.624  \n",
       "89                  NaN  55055.0          1.000  \n",
       "88                  NaN  59942.6          0.520  \n",
       "80                  NaN  56603.4          0.536  \n",
       "83                  NaN  57349.4          0.634  \n",
       "84                  NaN  57631.8          0.552  \n",
       "85                  NaN  56620.6          0.584  \n",
       "86                  NaN  57334.4          0.456  \n",
       "79                  NaN  57401.8          0.544  \n",
       "81                  NaN  56854.6          0.616  \n",
       "82                  NaN  57767.0          0.524  \n",
       "87                  NaN  57261.6          0.564  "
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "temp.drop([\"conversion_stages\", \"Estimator\"], axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "temp.drop([\"conversion_stages\", \"Estimator\"], axis=1).to_latex(\"../results/profit_treatment_models.tex\", na_rep=\"-\", index=False, formatters={'profit':'{:.0f}'.format,\"ratio_treated\":'{:.2f}'.format})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Prediction distribution analysis"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.backends.backend_pdf import PdfPages\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### Evaluation of distribution of predicted treatment effects\n",
    "\n",
    "axes_limits = {\"treatment_spending\":[-20,20,0,0.4],\n",
    "               \"treatment_basket_value\":[-20,20,0,0.3],\n",
    "               \"treatment_conversion\":[-0.15, 0.25, 0, 60]}\n",
    "\n",
    "clip_limits = {\"treatment_spending\":[-50,50],\n",
    "               \"treatment_basket_value\":[-50,50],\n",
    "               \"treatment_conversion\":[-0.5, 0.5]}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "for treatment_level in [\"treatment_spending\",\"treatment_basket_value\",\"treatment_conversion\"]:\n",
    "    for fold_index in range(len(predictions_test)):\n",
    "        with PdfPages(f\"../results/{treatment_level}_distribution_fold{fold_index}.pdf\") as pdf:\n",
    "            for model in predictions_test[fold_index][treatment_level].keys():\n",
    "                if model not in [\"oracle__\", \"ATE__\"]:\n",
    "                    plt.figure()\n",
    "                    plt.title(model)\n",
    "                    plt.xlabel(\"Model Estimate\")\n",
    "                    plt.ylabel(\"Kernel Density\")\n",
    "                    plt.axis(axes_limits[treatment_level]) #\n",
    "                    try:\n",
    "                        sns.kdeplot(predictions_test[fold_index][treatment_level][\"oracle__\"])\n",
    "                    except:\n",
    "                        sns.kdeplot(predictions_test[fold_index][treatment_level][\"oracle\"])\n",
    "                    sns.kdeplot(predictions_test[fold_index][treatment_level][model])\n",
    "                    pdf.savefig()\n",
    "                    plt.close()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Look at range of treatment effect predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "single-model_outcome_gbt      -5.941065\n",
      "single-model_hurdle_gbt      -13.037524\n",
      "two-model_outcome_linear   -1554.766673\n",
      "two-model_outcome_gbt        -17.599552\n",
      "two-model_hurdle_linear     -311.487841\n",
      "two-model_hurdle_gbt         -26.214187\n",
      "dr_outcome_linear          -1527.970429\n",
      "dr_outcome_gbt              -381.344892\n",
      "xbcf_outcome_xbcf             -3.126965\n",
      "oracle__                      -3.897381\n",
      "ATE__                          4.520083\n",
      "dtype: float64\n",
      "single-model_outcome_gbt     35.984899\n",
      "single-model_hurdle_gbt      39.157361\n",
      "two-model_outcome_linear    174.870256\n",
      "two-model_outcome_gbt        40.255595\n",
      "two-model_hurdle_linear     575.209426\n",
      "two-model_hurdle_gbt         56.357401\n",
      "dr_outcome_linear           180.185574\n",
      "dr_outcome_gbt              327.047726\n",
      "xbcf_outcome_xbcf            22.660689\n",
      "oracle__                    117.525666\n",
      "ATE__                         4.633144\n",
      "dtype: float64\n"
     ]
    }
   ],
   "source": [
    "for treatment_level in [\"treatment_spending\"]:  \n",
    "    predictions_combined = pd.concat([pd.DataFrame(fold[treatment_level]) for fold in predictions_test])\n",
    "print(predictions_combined.min())\n",
    "print(predictions_combined.max())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### Merge all folds\n",
    "for treatment_level in [\"treatment_spending\",\"treatment_basket_value\",\"treatment_conversion\"]:  \n",
    "    predictions_combined = pd.concat([pd.DataFrame(fold[treatment_level]) for fold in predictions_test])\n",
    "\n",
    "    with PdfPages(f\"../results/{treatment_level}_distribution_combined.pdf\") as pdf:\n",
    "        for model in predictions_combined.columns.values:\n",
    "            if model not in [\"oracle__\", \"ATE__\"]:\n",
    "                plt.figure()\n",
    "                plt.title(model)\n",
    "                plt.xlabel(\"Model Estimate\")\n",
    "                plt.ylabel(\"Kernel Density\")\n",
    "                plt.axis(axes_limits[treatment_level]) #\n",
    "                try:\n",
    "                    sns.kdeplot(predictions_combined[[\"oracle__\"]].values.flatten(), linestyle=\"--\", color='grey')\n",
    "                except:\n",
    "                    sns.kdeplot(predictions_combined[[\"oracle\"]].values.flatten(), linestyle=\"--\", color='grey')\n",
    "                #sns.kdeplot(tau_response.values.flatten(), linestyle=\"--\", color='green')\n",
    "                sns.kdeplot(predictions_combined[[model]].values.flatten(), \n",
    "                            clip=clip_limits[treatment_level], color=\"blue\")\n",
    "                pdf.savefig()\n",
    "                plt.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### Merge all folds\n",
    "for treatment_level in [\"treatment_spending\",\"treatment_basket_value\",\"treatment_conversion\"]:  \n",
    "    predictions_combined = pd.concat([pd.DataFrame(fold[treatment_level]) for fold in predictions_test])\n",
    "\n",
    "    with PdfPages(f\"../results/{treatment_level}_scatter_combined.pdf\") as pdf:\n",
    "        for model in predictions_combined.columns.values:\n",
    "            if model not in [\"oracle__\", \"ATE__\"]:\n",
    "                plt.figure()\n",
    "                plt.title(model)\n",
    "                plt.xlabel(\"Model Estimate\")\n",
    "                plt.ylabel(\"ITE\")\n",
    "                try:\n",
    "                    sns.kdeplot(x=predictions_combined[[model]].values.flatten(),y=predictions_combined[[\"oracle__\"]].values.flatten(), clip=clip_limits[treatment_level])\n",
    "                except:\n",
    "                    sns.kdeplot(x=predictions_combined[[model]].values.flatten(),y=predictions_combined[[\"oracle\"]].values.flatten(), clip=clip_limits[treatment_level])\n",
    "                #sns.kdeplot(tau_response.values.flatten(), linestyle=\"--\", color='green')\n",
    "                pdf.savefig()\n",
    "                plt.close()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Hurdle Model Parts Prediction Accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "effect_stage = {}\n",
    "effect_stage_summary = {}\n",
    "for treatment_level in [\"treatment_spending\",\"treatment_basket_value\",\"treatment_conversion\"]:  \n",
    "    effect_stage[treatment_level] = pd.concat([pd.DataFrame(fold[treatment_level]) for fold in predictions_test])\n",
    "    effect_stage_summary[treatment_level] = (effect_stage[treatment_level].describe().transpose())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "effect_stage_summary = pd.concat(effect_stage_summary,axis=1).drop(\"ATE__\").drop(['count','std'],axis=1, level=1).round(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [],
   "source": [
    "for key,value in effect_stage.items():\n",
    "    effect_stage[key] = value.drop(\"ATE__\", axis=1).rename({'single-model_hurdle_gbt':'Single Model GBT',\n",
    "                                                                      'two-model_hurdle_linear':'Two Model Linear',\n",
    "                                                                      'two-model_hurdle_gbt':'Two Model GBT',\n",
    "                                                                      'oracle__':'Oracle'}, axis=1)\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead tr th {\n",
       "        text-align: left;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th colspan=\"6\" halign=\"left\">treatment_spending</th>\n",
       "      <th colspan=\"6\" halign=\"left\">treatment_basket_value</th>\n",
       "      <th colspan=\"6\" halign=\"left\">treatment_conversion</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th>mean</th>\n",
       "      <th>min</th>\n",
       "      <th>25%</th>\n",
       "      <th>50%</th>\n",
       "      <th>75%</th>\n",
       "      <th>max</th>\n",
       "      <th>mean</th>\n",
       "      <th>min</th>\n",
       "      <th>25%</th>\n",
       "      <th>50%</th>\n",
       "      <th>75%</th>\n",
       "      <th>max</th>\n",
       "      <th>mean</th>\n",
       "      <th>min</th>\n",
       "      <th>25%</th>\n",
       "      <th>50%</th>\n",
       "      <th>75%</th>\n",
       "      <th>max</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>single-model_outcome_gbt</th>\n",
       "      <td>4.30</td>\n",
       "      <td>-5.94</td>\n",
       "      <td>2.67</td>\n",
       "      <td>3.96</td>\n",
       "      <td>5.50</td>\n",
       "      <td>35.98</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>single-model_hurdle_gbt</th>\n",
       "      <td>4.34</td>\n",
       "      <td>-13.04</td>\n",
       "      <td>2.26</td>\n",
       "      <td>3.71</td>\n",
       "      <td>5.66</td>\n",
       "      <td>39.16</td>\n",
       "      <td>0.08</td>\n",
       "      <td>-7.14</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>7.05</td>\n",
       "      <td>0.04</td>\n",
       "      <td>-0.08</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.06</td>\n",
       "      <td>0.26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>two-model_outcome_linear</th>\n",
       "      <td>4.58</td>\n",
       "      <td>-1554.77</td>\n",
       "      <td>2.12</td>\n",
       "      <td>4.46</td>\n",
       "      <td>6.68</td>\n",
       "      <td>174.87</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>two-model_outcome_gbt</th>\n",
       "      <td>4.58</td>\n",
       "      <td>-17.60</td>\n",
       "      <td>2.44</td>\n",
       "      <td>4.27</td>\n",
       "      <td>6.26</td>\n",
       "      <td>40.26</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>two-model_hurdle_linear</th>\n",
       "      <td>4.64</td>\n",
       "      <td>-311.49</td>\n",
       "      <td>2.32</td>\n",
       "      <td>4.07</td>\n",
       "      <td>6.27</td>\n",
       "      <td>575.21</td>\n",
       "      <td>-0.13</td>\n",
       "      <td>-2211.32</td>\n",
       "      <td>-2.40</td>\n",
       "      <td>0.55</td>\n",
       "      <td>3.20</td>\n",
       "      <td>44.85</td>\n",
       "      <td>0.05</td>\n",
       "      <td>-1.00</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.07</td>\n",
       "      <td>0.94</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>two-model_hurdle_gbt</th>\n",
       "      <td>4.58</td>\n",
       "      <td>-26.21</td>\n",
       "      <td>2.63</td>\n",
       "      <td>4.15</td>\n",
       "      <td>6.02</td>\n",
       "      <td>56.36</td>\n",
       "      <td>-0.23</td>\n",
       "      <td>-44.66</td>\n",
       "      <td>-3.09</td>\n",
       "      <td>0.19</td>\n",
       "      <td>3.03</td>\n",
       "      <td>47.29</td>\n",
       "      <td>0.05</td>\n",
       "      <td>-0.15</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.06</td>\n",
       "      <td>0.38</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>dr_outcome_linear</th>\n",
       "      <td>4.58</td>\n",
       "      <td>-1527.97</td>\n",
       "      <td>2.10</td>\n",
       "      <td>4.46</td>\n",
       "      <td>6.69</td>\n",
       "      <td>180.19</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>dr_outcome_gbt</th>\n",
       "      <td>4.60</td>\n",
       "      <td>-381.34</td>\n",
       "      <td>2.91</td>\n",
       "      <td>4.57</td>\n",
       "      <td>5.56</td>\n",
       "      <td>327.05</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>xbcf_outcome_xbcf</th>\n",
       "      <td>4.64</td>\n",
       "      <td>-3.13</td>\n",
       "      <td>2.34</td>\n",
       "      <td>4.44</td>\n",
       "      <td>6.46</td>\n",
       "      <td>22.66</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>oracle__</th>\n",
       "      <td>4.80</td>\n",
       "      <td>-3.90</td>\n",
       "      <td>1.30</td>\n",
       "      <td>4.83</td>\n",
       "      <td>7.21</td>\n",
       "      <td>117.53</td>\n",
       "      <td>1.00</td>\n",
       "      <td>-10.00</td>\n",
       "      <td>-0.02</td>\n",
       "      <td>1.72</td>\n",
       "      <td>2.98</td>\n",
       "      <td>9.14</td>\n",
       "      <td>0.05</td>\n",
       "      <td>-0.02</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.08</td>\n",
       "      <td>0.15</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                         treatment_spending                             \\\n",
       "                                       mean      min   25%   50%   75%   \n",
       "single-model_outcome_gbt               4.30    -5.94  2.67  3.96  5.50   \n",
       "single-model_hurdle_gbt                4.34   -13.04  2.26  3.71  5.66   \n",
       "two-model_outcome_linear               4.58 -1554.77  2.12  4.46  6.68   \n",
       "two-model_outcome_gbt                  4.58   -17.60  2.44  4.27  6.26   \n",
       "two-model_hurdle_linear                4.64  -311.49  2.32  4.07  6.27   \n",
       "two-model_hurdle_gbt                   4.58   -26.21  2.63  4.15  6.02   \n",
       "dr_outcome_linear                      4.58 -1527.97  2.10  4.46  6.69   \n",
       "dr_outcome_gbt                         4.60  -381.34  2.91  4.57  5.56   \n",
       "xbcf_outcome_xbcf                      4.64    -3.13  2.34  4.44  6.46   \n",
       "oracle__                               4.80    -3.90  1.30  4.83  7.21   \n",
       "\n",
       "                                 treatment_basket_value                       \\\n",
       "                             max                   mean      min   25%   50%   \n",
       "single-model_outcome_gbt   35.98                    NaN      NaN   NaN   NaN   \n",
       "single-model_hurdle_gbt    39.16                   0.08    -7.14  0.00  0.00   \n",
       "two-model_outcome_linear  174.87                    NaN      NaN   NaN   NaN   \n",
       "two-model_outcome_gbt      40.26                    NaN      NaN   NaN   NaN   \n",
       "two-model_hurdle_linear   575.21                  -0.13 -2211.32 -2.40  0.55   \n",
       "two-model_hurdle_gbt       56.36                  -0.23   -44.66 -3.09  0.19   \n",
       "dr_outcome_linear         180.19                    NaN      NaN   NaN   NaN   \n",
       "dr_outcome_gbt            327.05                    NaN      NaN   NaN   NaN   \n",
       "xbcf_outcome_xbcf          22.66                    NaN      NaN   NaN   NaN   \n",
       "oracle__                  117.53                   1.00   -10.00 -0.02  1.72   \n",
       "\n",
       "                                      treatment_conversion                    \\\n",
       "                           75%    max                 mean   min   25%   50%   \n",
       "single-model_outcome_gbt   NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "single-model_hurdle_gbt   0.00   7.05                 0.04 -0.08  0.02  0.04   \n",
       "two-model_outcome_linear   NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "two-model_outcome_gbt      NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "two-model_hurdle_linear   3.20  44.85                 0.05 -1.00  0.02  0.04   \n",
       "two-model_hurdle_gbt      3.03  47.29                 0.05 -0.15  0.03  0.05   \n",
       "dr_outcome_linear          NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "dr_outcome_gbt             NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "xbcf_outcome_xbcf          NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "oracle__                  2.98   9.14                 0.05 -0.02  0.01  0.05   \n",
       "\n",
       "                                      \n",
       "                           75%   max  \n",
       "single-model_outcome_gbt   NaN   NaN  \n",
       "single-model_hurdle_gbt   0.06  0.26  \n",
       "two-model_outcome_linear   NaN   NaN  \n",
       "two-model_outcome_gbt      NaN   NaN  \n",
       "two-model_hurdle_linear   0.07  0.94  \n",
       "two-model_hurdle_gbt      0.06  0.38  \n",
       "dr_outcome_linear          NaN   NaN  \n",
       "dr_outcome_gbt             NaN   NaN  \n",
       "xbcf_outcome_xbcf          NaN   NaN  \n",
       "oracle__                  0.08  0.15  "
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "effect_stage_summary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "with PdfPages(f\"../results/treatment_effect_on_spending.pdf\") as pdf:\n",
    "    sns.violinplot(data=effect_stage[\"treatment_spending\"].clip(-10,20), scale='count', orient='h', color='lightgrey')\n",
    "    plt.xlim(-8,15)\n",
    "    plt.title(\"Treatment Effect on Customer Spending\")\n",
    "    pdf.savefig()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "with PdfPages(f\"../results/treatment_effect_on_conversion.pdf\") as pdf:\n",
    "    sns.violinplot(data=effect_stage[\"treatment_conversion\"].clip(-0.05,0.2), scale='count', orient='h', color='lightgrey')\n",
    "    plt.xlim(-0.05,0.2)\n",
    "    plt.title(\"Treatment Effect on Purchase Probability\")\n",
    "    pdf.savefig()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "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": [
    "with PdfPages(f\"../results/treatment_effect_on_value.pdf\") as pdf:\n",
    "    sns.violinplot(data=effect_stage[\"treatment_basket_value\"].clip(-20,20), scale='count', orient='h', color='lightgrey')\n",
    "    plt.xlim(-13,13)\n",
    "    plt.title(\"Treatment Effect on Purchase Value\")\n",
    "    pdf.savefig()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead tr th {\n",
       "        text-align: left;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th colspan=\"6\" halign=\"left\">treatment_spending</th>\n",
       "      <th colspan=\"6\" halign=\"left\">treatment_basket_value</th>\n",
       "      <th colspan=\"6\" halign=\"left\">treatment_conversion</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th></th>\n",
       "      <th>mean</th>\n",
       "      <th>min</th>\n",
       "      <th>25%</th>\n",
       "      <th>50%</th>\n",
       "      <th>75%</th>\n",
       "      <th>max</th>\n",
       "      <th>mean</th>\n",
       "      <th>min</th>\n",
       "      <th>25%</th>\n",
       "      <th>50%</th>\n",
       "      <th>75%</th>\n",
       "      <th>max</th>\n",
       "      <th>mean</th>\n",
       "      <th>min</th>\n",
       "      <th>25%</th>\n",
       "      <th>50%</th>\n",
       "      <th>75%</th>\n",
       "      <th>max</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>single-model_outcome_gbt</th>\n",
       "      <td>4.30</td>\n",
       "      <td>-5.94</td>\n",
       "      <td>2.67</td>\n",
       "      <td>3.96</td>\n",
       "      <td>5.50</td>\n",
       "      <td>35.98</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>single-model_hurdle_gbt</th>\n",
       "      <td>4.34</td>\n",
       "      <td>-13.04</td>\n",
       "      <td>2.26</td>\n",
       "      <td>3.71</td>\n",
       "      <td>5.66</td>\n",
       "      <td>39.16</td>\n",
       "      <td>0.08</td>\n",
       "      <td>-7.14</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.00</td>\n",
       "      <td>7.05</td>\n",
       "      <td>0.04</td>\n",
       "      <td>-0.08</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.06</td>\n",
       "      <td>0.26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>two-model_outcome_linear</th>\n",
       "      <td>4.58</td>\n",
       "      <td>-1554.77</td>\n",
       "      <td>2.12</td>\n",
       "      <td>4.46</td>\n",
       "      <td>6.68</td>\n",
       "      <td>174.87</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>two-model_outcome_gbt</th>\n",
       "      <td>4.58</td>\n",
       "      <td>-17.60</td>\n",
       "      <td>2.44</td>\n",
       "      <td>4.27</td>\n",
       "      <td>6.26</td>\n",
       "      <td>40.26</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>two-model_hurdle_linear</th>\n",
       "      <td>4.64</td>\n",
       "      <td>-311.49</td>\n",
       "      <td>2.32</td>\n",
       "      <td>4.07</td>\n",
       "      <td>6.27</td>\n",
       "      <td>575.21</td>\n",
       "      <td>-0.13</td>\n",
       "      <td>-2211.32</td>\n",
       "      <td>-2.40</td>\n",
       "      <td>0.55</td>\n",
       "      <td>3.20</td>\n",
       "      <td>44.85</td>\n",
       "      <td>0.05</td>\n",
       "      <td>-1.00</td>\n",
       "      <td>0.02</td>\n",
       "      <td>0.04</td>\n",
       "      <td>0.07</td>\n",
       "      <td>0.94</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>two-model_hurdle_gbt</th>\n",
       "      <td>4.58</td>\n",
       "      <td>-26.21</td>\n",
       "      <td>2.63</td>\n",
       "      <td>4.15</td>\n",
       "      <td>6.02</td>\n",
       "      <td>56.36</td>\n",
       "      <td>-0.23</td>\n",
       "      <td>-44.66</td>\n",
       "      <td>-3.09</td>\n",
       "      <td>0.19</td>\n",
       "      <td>3.03</td>\n",
       "      <td>47.29</td>\n",
       "      <td>0.05</td>\n",
       "      <td>-0.15</td>\n",
       "      <td>0.03</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.06</td>\n",
       "      <td>0.38</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>dr_outcome_linear</th>\n",
       "      <td>4.58</td>\n",
       "      <td>-1527.97</td>\n",
       "      <td>2.10</td>\n",
       "      <td>4.46</td>\n",
       "      <td>6.69</td>\n",
       "      <td>180.19</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>dr_outcome_gbt</th>\n",
       "      <td>4.60</td>\n",
       "      <td>-381.34</td>\n",
       "      <td>2.91</td>\n",
       "      <td>4.57</td>\n",
       "      <td>5.56</td>\n",
       "      <td>327.05</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>xbcf_outcome_xbcf</th>\n",
       "      <td>4.64</td>\n",
       "      <td>-3.13</td>\n",
       "      <td>2.34</td>\n",
       "      <td>4.44</td>\n",
       "      <td>6.46</td>\n",
       "      <td>22.66</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>oracle__</th>\n",
       "      <td>4.80</td>\n",
       "      <td>-3.90</td>\n",
       "      <td>1.30</td>\n",
       "      <td>4.83</td>\n",
       "      <td>7.21</td>\n",
       "      <td>117.53</td>\n",
       "      <td>1.00</td>\n",
       "      <td>-10.00</td>\n",
       "      <td>-0.02</td>\n",
       "      <td>1.72</td>\n",
       "      <td>2.98</td>\n",
       "      <td>9.14</td>\n",
       "      <td>0.05</td>\n",
       "      <td>-0.02</td>\n",
       "      <td>0.01</td>\n",
       "      <td>0.05</td>\n",
       "      <td>0.08</td>\n",
       "      <td>0.15</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                         treatment_spending                             \\\n",
       "                                       mean      min   25%   50%   75%   \n",
       "single-model_outcome_gbt               4.30    -5.94  2.67  3.96  5.50   \n",
       "single-model_hurdle_gbt                4.34   -13.04  2.26  3.71  5.66   \n",
       "two-model_outcome_linear               4.58 -1554.77  2.12  4.46  6.68   \n",
       "two-model_outcome_gbt                  4.58   -17.60  2.44  4.27  6.26   \n",
       "two-model_hurdle_linear                4.64  -311.49  2.32  4.07  6.27   \n",
       "two-model_hurdle_gbt                   4.58   -26.21  2.63  4.15  6.02   \n",
       "dr_outcome_linear                      4.58 -1527.97  2.10  4.46  6.69   \n",
       "dr_outcome_gbt                         4.60  -381.34  2.91  4.57  5.56   \n",
       "xbcf_outcome_xbcf                      4.64    -3.13  2.34  4.44  6.46   \n",
       "oracle__                               4.80    -3.90  1.30  4.83  7.21   \n",
       "\n",
       "                                 treatment_basket_value                       \\\n",
       "                             max                   mean      min   25%   50%   \n",
       "single-model_outcome_gbt   35.98                    NaN      NaN   NaN   NaN   \n",
       "single-model_hurdle_gbt    39.16                   0.08    -7.14  0.00  0.00   \n",
       "two-model_outcome_linear  174.87                    NaN      NaN   NaN   NaN   \n",
       "two-model_outcome_gbt      40.26                    NaN      NaN   NaN   NaN   \n",
       "two-model_hurdle_linear   575.21                  -0.13 -2211.32 -2.40  0.55   \n",
       "two-model_hurdle_gbt       56.36                  -0.23   -44.66 -3.09  0.19   \n",
       "dr_outcome_linear         180.19                    NaN      NaN   NaN   NaN   \n",
       "dr_outcome_gbt            327.05                    NaN      NaN   NaN   NaN   \n",
       "xbcf_outcome_xbcf          22.66                    NaN      NaN   NaN   NaN   \n",
       "oracle__                  117.53                   1.00   -10.00 -0.02  1.72   \n",
       "\n",
       "                                      treatment_conversion                    \\\n",
       "                           75%    max                 mean   min   25%   50%   \n",
       "single-model_outcome_gbt   NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "single-model_hurdle_gbt   0.00   7.05                 0.04 -0.08  0.02  0.04   \n",
       "two-model_outcome_linear   NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "two-model_outcome_gbt      NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "two-model_hurdle_linear   3.20  44.85                 0.05 -1.00  0.02  0.04   \n",
       "two-model_hurdle_gbt      3.03  47.29                 0.05 -0.15  0.03  0.05   \n",
       "dr_outcome_linear          NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "dr_outcome_gbt             NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "xbcf_outcome_xbcf          NaN    NaN                  NaN   NaN   NaN   NaN   \n",
       "oracle__                  2.98   9.14                 0.05 -0.02  0.01  0.05   \n",
       "\n",
       "                                      \n",
       "                           75%   max  \n",
       "single-model_outcome_gbt   NaN   NaN  \n",
       "single-model_hurdle_gbt   0.06  0.26  \n",
       "two-model_outcome_linear   NaN   NaN  \n",
       "two-model_outcome_gbt      NaN   NaN  \n",
       "two-model_hurdle_linear   0.07  0.94  \n",
       "two-model_hurdle_gbt      0.06  0.38  \n",
       "dr_outcome_linear          NaN   NaN  \n",
       "dr_outcome_gbt             NaN   NaN  \n",
       "xbcf_outcome_xbcf          NaN   NaN  \n",
       "oracle__                  0.08  0.15  "
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "effect_stage_summary.to_latex(\"../results/effect_summary_per_level.tex\", na_rep=\"-\", index=True)\n",
    "effect_stage_summary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "results = {}\n",
    "\n",
    "for treatment_level in [\"treatment_basket_value\",\"treatment_conversion\"]:  \n",
    "    predictions_combined = pd.concat([pd.DataFrame(fold[treatment_level]) for fold in predictions_test])\n",
    "    results[treatment_level] = {}\n",
    "    for model in predictions_combined.columns.values:\n",
    "        results[treatment_level][model] = np.sqrt(mean_absolute_error(y_pred=predictions_combined[[model]], \n",
    "                                                                     y_true=predictions_combined[[\"oracle__\"]]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [],
   "source": [
    "results = pd.DataFrame(results)\n",
    "results.index = pd.MultiIndex.from_tuples(results.index.str.split(\"[+|_]\", expand=True).tolist())\n",
    "results.columns = [\"RMSE Basket Value\", \"RMSE Conversion\"]\n",
    "results.index.names = [\"Causal\",\"Stages\",\"Estimator\"]\n",
    "\n",
    "results.drop('oracle', level=\"Causal\", inplace=True)\n",
    "results.reset_index(drop=False, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>Causal</th>\n",
       "      <th>Stages</th>\n",
       "      <th>Estimator</th>\n",
       "      <th>RMSE Basket Value</th>\n",
       "      <th>RMSE Conversion</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>single-model</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>gbt</td>\n",
       "      <td>1.586720</td>\n",
       "      <td>0.151528</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>two-model</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>linear</td>\n",
       "      <td>1.990707</td>\n",
       "      <td>0.158453</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>two-model</td>\n",
       "      <td>hurdle</td>\n",
       "      <td>gbt</td>\n",
       "      <td>2.094563</td>\n",
       "      <td>0.155283</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>ATE</td>\n",
       "      <td></td>\n",
       "      <td></td>\n",
       "      <td>1.829925</td>\n",
       "      <td>0.175813</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         Causal  Stages Estimator  RMSE Basket Value  RMSE Conversion\n",
       "0  single-model  hurdle       gbt           1.586720         0.151528\n",
       "1     two-model  hurdle    linear           1.990707         0.158453\n",
       "2     two-model  hurdle       gbt           2.094563         0.155283\n",
       "3           ATE                             1.829925         0.175813"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "results.to_latex(\"../results/stepwise_performance_hurdle_models.tex\", na_rep=\"-\", index=False, formatters={'RMSE Basket Value':'{:.2f}'.format,\"RMSE Conversion\":'{:.4f}'.format})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "treatment_level= \"treatment_basket_value\"\n",
    "predictions_combined = pd.concat([pd.DataFrame(fold[treatment_level]) for fold in predictions_test])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "two-model_hurdle_linear   -0.133867\n",
       "dtype: float64"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predictions_combined[['two-model_hurdle_linear']].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "two-model_hurdle_linear    0.54638\n",
       "Name: 0.5, dtype: float64"
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predictions_combined[['two-model_hurdle_linear']].quantile()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "two-model_hurdle_gbt   -0.232688\n",
       "dtype: float64"
      ]
     },
     "execution_count": 68,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predictions_combined[['two-model_hurdle_gbt']].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "oracle__    1.000135\n",
       "dtype: float64"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predictions_combined[['oracle__']].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>oracle__</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0.00</th>\n",
       "      <td>-10.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.05</th>\n",
       "      <td>-5.574609</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.50</th>\n",
       "      <td>1.717308</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.95</th>\n",
       "      <td>4.425391</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1.00</th>\n",
       "      <td>9.142449</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       oracle__\n",
       "0.00 -10.000000\n",
       "0.05  -5.574609\n",
       "0.50   1.717308\n",
       "0.95   4.425391\n",
       "1.00   9.142449"
      ]
     },
     "execution_count": 70,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predictions_combined[['oracle__']].quantile(q=[0,0.05,0.5,0.95,1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:uplift]",
   "language": "python",
   "name": "conda-env-uplift-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}