main.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from models.BinaryMI import BinaryMI\n",
    "from models.DebiasClassifier import DebiasClassifier\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.metrics import roc_curve, auc, roc_auc_score\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib as mpl\n",
    "import matplotlib.patches as patches\n",
    "from matplotlib import rc, rcParams\n",
    "from matplotlib.collections import LineCollection\n",
    "from scipy.interpolate import UnivariateSpline\n",
    "import random"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# The label ‘signal’ being ‘1’ for signal events, ‘0’ for background events\n",
    "# the signal events are simulated, while background events are real data.\n",
    "training = pd.read_csv(\"data/training.csv\")\n",
    "# The test dataset has all the columns that training.csv has, except\n",
    "# mass, production, min_ANNmuon, and signal.\n",
    "test = pd.read_csv(\"data/test.csv\")\n",
    "# Simulated and real events from the channel Ds → φπ to evaluate\n",
    "# the performance for simulated-real data\n",
    "# It contains the same columns as test.csv and weight column. \n",
    "agreement = pd.read_csv(\"data/check_agreement.csv\")\n",
    "# Only real background events recorded at LHCb to evaluate\n",
    "# the mass values locally.\n",
    "# It contains the same columns as test.csv and mass column to check correlation\n",
    "correlation = pd.read_csv(\"data/check_correlation.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Agreement data # source domain: 8205, # target domain: 322942\n",
      "Training data # source domain: 41674, # target domain: 25879\n"
     ]
    }
   ],
   "source": [
    "# simulation (source) (signal==1) contains 8205 examples \n",
    "# real data (target) (signal==0) contains 322942 examples with weight values\n",
    "counts = np.unique(agreement[\"signal\"], return_counts=True)\n",
    "counts_train = np.unique(training[\"signal\"], return_counts=True)\n",
    "\n",
    "print(f\"Agreement data # source domain: {counts[1][1]}, # target domain: {counts[1][0]}\")\n",
    "print(f\"Training data # source domain: {counts_train[1][1]}, # target domain: {counts_train[1][0]}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# prepare the data\n",
    "\n",
    "# split the data for agreement test at the end\n",
    "agreement_train, agreement_test = \\\n",
    "    train_test_split(agreement, test_size=0.666)\n",
    "\n",
    "# we take equal values from both groups\n",
    "agreement_train_s0 = agreement_train[agreement_train[\"signal\"]==0].sample(n=int(len(agreement_train[agreement_train[\"signal\"]==1])/2))\n",
    "agreement_train_s1 = agreement_train[agreement_train[\"signal\"]==1].sample(n=int(len(agreement_train[agreement_train[\"signal\"]==1])/2))\n",
    "agreement = pd.concat([agreement_train_s0, agreement_train_s1])\n",
    "\n",
    "# prepare the data for the domain adaptation task\n",
    "s = np.concatenate([agreement_train[\"signal\"].to_numpy(), training[\"signal\"].to_numpy()])\n",
    "agreement_train = agreement_train.drop(columns=[\"id\", \"signal\", \"SPDhits\", \"weight\"], axis = 1)\n",
    "\n",
    "# from the agreement data we only have background so y=0\n",
    "y = np.concatenate([np.zeros(len(agreement_train)), training[\"signal\"].to_numpy()])\n",
    "\n",
    "# Note: for this example we dont use the mass\n",
    "# mass_train = training[\"mass\"]\n",
    "\n",
    "# create the features to train on\n",
    "x = training.drop(columns=[\"id\", \"production\", \"min_ANNmuon\", \"signal\", \"mass\", \"SPDhits\"], axis = 1)\n",
    "x = pd.concat([agreement_train, x], ignore_index=True)\n",
    "\n",
    "# split the data into train and test set\n",
    "X_train, X_test, y_train, y_test, S_train, S_test = \\\n",
    "    train_test_split(x, y, s, test_size=0.333)\n",
    "\n",
    "# scale data\n",
    "scaler = StandardScaler()\n",
    "X_train = scaler.fit_transform(X_train)\n",
    "X_test = scaler.transform(X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"BinaryMI\"</span>\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1mModel: \"BinaryMI\"\u001b[0m\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\"> Layer (type)                    </span>┃<span style=\"font-weight: bold\"> Output Shape           </span>┃<span style=\"font-weight: bold\">       Param # </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ x (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)                  │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">45</span>)             │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_0 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">2,944</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ qact_t_0 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">QActivation</span>)          │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)             │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">2,080</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_2 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">2</span>)              │            <span style=\"color: #00af00; text-decoration-color: #00af00\">66</span> │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape          \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ x (\u001b[38;5;33mInputLayer\u001b[0m)                  │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m45\u001b[0m)             │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_0 (\u001b[38;5;33mDense\u001b[0m)                     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m64\u001b[0m)             │         \u001b[38;5;34m2,944\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ qact_t_0 (\u001b[38;5;33mQActivation\u001b[0m)          │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m64\u001b[0m)             │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_1 (\u001b[38;5;33mDense\u001b[0m)                     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m32\u001b[0m)             │         \u001b[38;5;34m2,080\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_2 (\u001b[38;5;33mDense\u001b[0m)                     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m2\u001b[0m)              │            \u001b[38;5;34m66\u001b[0m │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">5,090</span> (19.88 KB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m5,090\u001b[0m (19.88 KB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">5,090</span> (19.88 KB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m5,090\u001b[0m (19.88 KB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 466us/step - loss: 0.5262 - qact_t_0_acc: 0.6760 - t_2_acc: 0.7701 - val_loss: 0.4590 - val_qact_t_0_acc: 0.5866 - val_t_2_acc: 0.8281\n",
      "Epoch 2/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 457us/step - loss: 0.4555 - qact_t_0_acc: 0.5768 - t_2_acc: 0.8269 - val_loss: 0.4490 - val_qact_t_0_acc: 0.5490 - val_t_2_acc: 0.8370\n",
      "Epoch 3/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 412us/step - loss: 0.4466 - qact_t_0_acc: 0.5452 - t_2_acc: 0.8318 - val_loss: 0.4388 - val_qact_t_0_acc: 0.5350 - val_t_2_acc: 0.8419\n",
      "Epoch 4/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 419us/step - loss: 0.4396 - qact_t_0_acc: 0.5300 - t_2_acc: 0.8351 - val_loss: 0.4316 - val_qact_t_0_acc: 0.5202 - val_t_2_acc: 0.8449\n",
      "Epoch 5/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 465us/step - loss: 0.4351 - qact_t_0_acc: 0.5232 - t_2_acc: 0.8371 - val_loss: 0.4340 - val_qact_t_0_acc: 0.5195 - val_t_2_acc: 0.8469\n",
      "Epoch 6/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 391us/step - loss: 0.4359 - qact_t_0_acc: 0.5199 - t_2_acc: 0.8380 - val_loss: 0.4293 - val_qact_t_0_acc: 0.5171 - val_t_2_acc: 0.8401\n",
      "Epoch 7/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 388us/step - loss: 0.4332 - qact_t_0_acc: 0.5146 - t_2_acc: 0.8395 - val_loss: 0.4334 - val_qact_t_0_acc: 0.5112 - val_t_2_acc: 0.8450\n",
      "Epoch 8/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 403us/step - loss: 0.4315 - qact_t_0_acc: 0.5109 - t_2_acc: 0.8401 - val_loss: 0.4260 - val_qact_t_0_acc: 0.5054 - val_t_2_acc: 0.8455\n",
      "Epoch 9/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 389us/step - loss: 0.4300 - qact_t_0_acc: 0.5073 - t_2_acc: 0.8428 - val_loss: 0.4345 - val_qact_t_0_acc: 0.5135 - val_t_2_acc: 0.8505\n",
      "Epoch 10/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 415us/step - loss: 0.4301 - qact_t_0_acc: 0.5042 - t_2_acc: 0.8432 - val_loss: 0.4283 - val_qact_t_0_acc: 0.5009 - val_t_2_acc: 0.8474\n"
     ]
    }
   ],
   "source": [
    "# train the transfer model\n",
    "BMI = BinaryMI(\n",
    "    kernel_regularizer = 0.0,\n",
    "    batch_size=32,\n",
    "    hidden_layers = [64,32],\n",
    "    validation_size=0.1,\n",
    "    quantized_position=[True, False],\n",
    "    input_shape=(X_train.shape[1],),\n",
    "    verbose=1,\n",
    "    gamma=0.3,\n",
    "    print_summary=True,\n",
    "    last_layer_size=2,\n",
    "    run_eagerly=False\n",
    ")\n",
    "train_hist_BMI = BMI.fit(x_train=X_train, y_train=y_train, s_train=S_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"BinaryMI\"</span>\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1mModel: \"BinaryMI\"\u001b[0m\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\"> Layer (type)                    </span>┃<span style=\"font-weight: bold\"> Output Shape           </span>┃<span style=\"font-weight: bold\">       Param # </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ x (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)                  │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">45</span>)             │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_0 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">2,944</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ qact_t_0 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">QActivation</span>)          │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)             │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">2,080</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_2 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">2</span>)              │            <span style=\"color: #00af00; text-decoration-color: #00af00\">66</span> │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape          \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ x (\u001b[38;5;33mInputLayer\u001b[0m)                  │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m45\u001b[0m)             │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_0 (\u001b[38;5;33mDense\u001b[0m)                     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m64\u001b[0m)             │         \u001b[38;5;34m2,944\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ qact_t_0 (\u001b[38;5;33mQActivation\u001b[0m)          │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m64\u001b[0m)             │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_1 (\u001b[38;5;33mDense\u001b[0m)                     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m32\u001b[0m)             │         \u001b[38;5;34m2,080\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ t_2 (\u001b[38;5;33mDense\u001b[0m)                     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m2\u001b[0m)              │            \u001b[38;5;34m66\u001b[0m │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">5,090</span> (19.88 KB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m5,090\u001b[0m (19.88 KB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">5,090</span> (19.88 KB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m5,090\u001b[0m (19.88 KB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 432us/step - loss: 0.3959 - qact_t_0_acc: 0.7412 - t_2_acc: 0.8194 - val_loss: 0.3132 - val_qact_t_0_acc: 0.7438 - val_t_2_acc: 0.8561\n",
      "Epoch 2/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 421us/step - loss: 0.3133 - qact_t_0_acc: 0.7451 - t_2_acc: 0.8578 - val_loss: 0.2958 - val_qact_t_0_acc: 0.7425 - val_t_2_acc: 0.8705\n",
      "Epoch 3/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 409us/step - loss: 0.3042 - qact_t_0_acc: 0.7374 - t_2_acc: 0.8645 - val_loss: 0.2871 - val_qact_t_0_acc: 0.7284 - val_t_2_acc: 0.8739\n",
      "Epoch 4/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 425us/step - loss: 0.2901 - qact_t_0_acc: 0.7271 - t_2_acc: 0.8716 - val_loss: 0.2786 - val_qact_t_0_acc: 0.7193 - val_t_2_acc: 0.8766\n",
      "Epoch 5/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 399us/step - loss: 0.2920 - qact_t_0_acc: 0.7182 - t_2_acc: 0.8707 - val_loss: 0.2753 - val_qact_t_0_acc: 0.7156 - val_t_2_acc: 0.8818\n",
      "Epoch 6/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 410us/step - loss: 0.2846 - qact_t_0_acc: 0.7155 - t_2_acc: 0.8749 - val_loss: 0.2744 - val_qact_t_0_acc: 0.7091 - val_t_2_acc: 0.8791\n",
      "Epoch 7/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 402us/step - loss: 0.2823 - qact_t_0_acc: 0.7093 - t_2_acc: 0.8762 - val_loss: 0.2733 - val_qact_t_0_acc: 0.7086 - val_t_2_acc: 0.8805\n",
      "Epoch 8/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 419us/step - loss: 0.2802 - qact_t_0_acc: 0.7092 - t_2_acc: 0.8767 - val_loss: 0.2696 - val_qact_t_0_acc: 0.7117 - val_t_2_acc: 0.8800\n",
      "Epoch 9/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 413us/step - loss: 0.2805 - qact_t_0_acc: 0.7074 - t_2_acc: 0.8771 - val_loss: 0.2731 - val_qact_t_0_acc: 0.7083 - val_t_2_acc: 0.8803\n",
      "Epoch 10/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 443us/step - loss: 0.2779 - qact_t_0_acc: 0.7056 - t_2_acc: 0.8780 - val_loss: 0.2668 - val_qact_t_0_acc: 0.7157 - val_t_2_acc: 0.8831\n"
     ]
    }
   ],
   "source": [
    "# train a simple NN\n",
    "NN = BinaryMI(\n",
    "    kernel_regularizer = 0.0,\n",
    "    batch_size=32,\n",
    "    hidden_layers = [64,32],\n",
    "    validation_size=0.1,\n",
    "    quantized_position=[True, False],\n",
    "    input_shape=(X_train.shape[1],),\n",
    "    verbose=1,\n",
    "    gamma=0.0,\n",
    "    print_summary=True,\n",
    "    last_layer_size=2,\n",
    "    run_eagerly=False\n",
    ")\n",
    "train_hist_NN = NN.fit(x_train=X_train, y_train=y_train, s_train=S_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"nn\"</span>\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1mModel: \"nn\"\u001b[0m\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\"> Layer (type)                    </span>┃<span style=\"font-weight: bold\"> Output Shape           </span>┃<span style=\"font-weight: bold\">       Param # </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ input (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)              │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">45</span>)             │             <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ nn_nn_0 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                 │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">64</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">2,944</span> │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ nn_nn_1 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)                 │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>)             │         <span style=\"color: #00af00; text-decoration-color: #00af00\">2,080</span> │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)                   \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape          \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m      Param #\u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━┩\n",
       "│ input (\u001b[38;5;33mInputLayer\u001b[0m)              │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m45\u001b[0m)             │             \u001b[38;5;34m0\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ nn_nn_0 (\u001b[38;5;33mDense\u001b[0m)                 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m64\u001b[0m)             │         \u001b[38;5;34m2,944\u001b[0m │\n",
       "├─────────────────────────────────┼────────────────────────┼───────────────┤\n",
       "│ nn_nn_1 (\u001b[38;5;33mDense\u001b[0m)                 │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m32\u001b[0m)             │         \u001b[38;5;34m2,080\u001b[0m │\n",
       "└─────────────────────────────────┴────────────────────────┴───────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">5,024</span> (19.62 KB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m5,024\u001b[0m (19.62 KB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">5,024</span> (19.62 KB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m5,024\u001b[0m (19.62 KB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\">Model: \"DebiasClassifier\"</span>\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1mModel: \"DebiasClassifier\"\u001b[0m\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓\n",
       "┃<span style=\"font-weight: bold\"> Layer (type)        </span>┃<span style=\"font-weight: bold\"> Output Shape      </span>┃<span style=\"font-weight: bold\">    Param # </span>┃<span style=\"font-weight: bold\"> Connected to      </span>┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩\n",
       "│ x0 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">InputLayer</span>)     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">45</span>)        │          <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │ -                 │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ nn (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Functional</span>)     │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>)        │      <span style=\"color: #00af00; text-decoration-color: #00af00\">5,024</span> │ x0[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>]          │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ grad_reverse_1      │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">32</span>)        │          <span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> │ nn[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>]          │\n",
       "│ (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">GradReverse</span>)       │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ nn_bias_0 (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)   │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">10</span>)        │        <span style=\"color: #00af00; text-decoration-color: #00af00\">330</span> │ grad_reverse_1[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>… │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ cls_part (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>)    │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">1</span>)         │         <span style=\"color: #00af00; text-decoration-color: #00af00\">33</span> │ nn[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>]          │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ nn_out_bias (<span style=\"color: #0087ff; text-decoration-color: #0087ff\">Dense</span>) │ (<span style=\"color: #00d7ff; text-decoration-color: #00d7ff\">None</span>, <span style=\"color: #00af00; text-decoration-color: #00af00\">1</span>)         │         <span style=\"color: #00af00; text-decoration-color: #00af00\">11</span> │ nn_bias_0[<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>][<span style=\"color: #00af00; text-decoration-color: #00af00\">0</span>]   │\n",
       "└─────────────────────┴───────────────────┴────────────┴───────────────────┘\n",
       "</pre>\n"
      ],
      "text/plain": [
       "┏━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━┓\n",
       "┃\u001b[1m \u001b[0m\u001b[1mLayer (type)       \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mOutput Shape     \u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1m   Param #\u001b[0m\u001b[1m \u001b[0m┃\u001b[1m \u001b[0m\u001b[1mConnected to     \u001b[0m\u001b[1m \u001b[0m┃\n",
       "┡━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━┩\n",
       "│ x0 (\u001b[38;5;33mInputLayer\u001b[0m)     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m45\u001b[0m)        │          \u001b[38;5;34m0\u001b[0m │ -                 │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ nn (\u001b[38;5;33mFunctional\u001b[0m)     │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m32\u001b[0m)        │      \u001b[38;5;34m5,024\u001b[0m │ x0[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m]          │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ grad_reverse_1      │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m32\u001b[0m)        │          \u001b[38;5;34m0\u001b[0m │ nn[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m]          │\n",
       "│ (\u001b[38;5;33mGradReverse\u001b[0m)       │                   │            │                   │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ nn_bias_0 (\u001b[38;5;33mDense\u001b[0m)   │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m10\u001b[0m)        │        \u001b[38;5;34m330\u001b[0m │ grad_reverse_1[\u001b[38;5;34m0\u001b[0m… │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ cls_part (\u001b[38;5;33mDense\u001b[0m)    │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m1\u001b[0m)         │         \u001b[38;5;34m33\u001b[0m │ nn[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m]          │\n",
       "├─────────────────────┼───────────────────┼────────────┼───────────────────┤\n",
       "│ nn_out_bias (\u001b[38;5;33mDense\u001b[0m) │ (\u001b[38;5;45mNone\u001b[0m, \u001b[38;5;34m1\u001b[0m)         │         \u001b[38;5;34m11\u001b[0m │ nn_bias_0[\u001b[38;5;34m0\u001b[0m][\u001b[38;5;34m0\u001b[0m]   │\n",
       "└─────────────────────┴───────────────────┴────────────┴───────────────────┘\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Total params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">5,398</span> (21.09 KB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Total params: \u001b[0m\u001b[38;5;34m5,398\u001b[0m (21.09 KB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">5,398</span> (21.09 KB)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Trainable params: \u001b[0m\u001b[38;5;34m5,398\u001b[0m (21.09 KB)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\"><span style=\"font-weight: bold\"> Non-trainable params: </span><span style=\"color: #00af00; text-decoration-color: #00af00\">0</span> (0.00 B)\n",
       "</pre>\n"
      ],
      "text/plain": [
       "\u001b[1m Non-trainable params: \u001b[0m\u001b[38;5;34m0\u001b[0m (0.00 B)\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 429us/step - cls_part_AUC: 0.9035 - loss: 0.3406 - nn_out_bias_acc: 0.8103 - val_cls_part_AUC: 0.9409 - val_loss: 0.2627 - val_nn_out_bias_acc: 0.8830\n",
      "Epoch 2/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 390us/step - cls_part_AUC: 0.9370 - loss: 0.2690 - nn_out_bias_acc: 0.8795 - val_cls_part_AUC: 0.9470 - val_loss: 0.2502 - val_nn_out_bias_acc: 0.8885\n",
      "Epoch 3/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 406us/step - cls_part_AUC: 0.9425 - loss: 0.2582 - nn_out_bias_acc: 0.8843 - val_cls_part_AUC: 0.9498 - val_loss: 0.2442 - val_nn_out_bias_acc: 0.8915\n",
      "Epoch 4/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 375us/step - cls_part_AUC: 0.9454 - loss: 0.2524 - nn_out_bias_acc: 0.8873 - val_cls_part_AUC: 0.9515 - val_loss: 0.2406 - val_nn_out_bias_acc: 0.8935\n",
      "Epoch 5/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 401us/step - cls_part_AUC: 0.9473 - loss: 0.2485 - nn_out_bias_acc: 0.8887 - val_cls_part_AUC: 0.9527 - val_loss: 0.2379 - val_nn_out_bias_acc: 0.8951\n",
      "Epoch 6/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 396us/step - cls_part_AUC: 0.9487 - loss: 0.2457 - nn_out_bias_acc: 0.8895 - val_cls_part_AUC: 0.9536 - val_loss: 0.2359 - val_nn_out_bias_acc: 0.8958\n",
      "Epoch 7/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 388us/step - cls_part_AUC: 0.9498 - loss: 0.2434 - nn_out_bias_acc: 0.8904 - val_cls_part_AUC: 0.9543 - val_loss: 0.2342 - val_nn_out_bias_acc: 0.8972\n",
      "Epoch 8/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 397us/step - cls_part_AUC: 0.9506 - loss: 0.2416 - nn_out_bias_acc: 0.8913 - val_cls_part_AUC: 0.9550 - val_loss: 0.2328 - val_nn_out_bias_acc: 0.8968\n",
      "Epoch 9/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 377us/step - cls_part_AUC: 0.9513 - loss: 0.2400 - nn_out_bias_acc: 0.8922 - val_cls_part_AUC: 0.9556 - val_loss: 0.2316 - val_nn_out_bias_acc: 0.8974\n",
      "Epoch 10/10\n",
      "\u001b[1m3343/3343\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m1s\u001b[0m 392us/step - cls_part_AUC: 0.9520 - loss: 0.2387 - nn_out_bias_acc: 0.8928 - val_cls_part_AUC: 0.9560 - val_loss: 0.2305 - val_nn_out_bias_acc: 0.8981\n"
     ]
    }
   ],
   "source": [
    "# train a DebiasClassifier\n",
    "DCls = DebiasClassifier(\n",
    "    kernel_regularizer=0.0,\n",
    "    batch_size=32,\n",
    "    validation_size=0.1,\n",
    "    hidden_layers=[64,32],\n",
    "    bias_layers=[10],\n",
    "    verbose=1,\n",
    "    gamma=0.2,\n",
    "    print_summary=True,\n",
    "    run_eagerly=False\n",
    ")\n",
    "train_hist_DCls = DCls.fit(x=X_train, y=y_train, s=S_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [],
   "source": [
    "# following code is based on: https://www.kaggle.com/code/matteomonzali/nn-model\n",
    "\n",
    "def __rolling_window(data, window_size):\n",
    "    \"\"\"\n",
    "    Rolling window: take window with definite size through the array\n",
    "\n",
    "    :param data: array-like\n",
    "    :param window_size: size\n",
    "    :return: the sequence of windows\n",
    "\n",
    "    Example: data = array(1, 2, 3, 4, 5, 6), window_size = 4\n",
    "        Then this function return array(array(1, 2, 3, 4), array(2, 3, 4, 5), array(3, 4, 5, 6))\n",
    "    \"\"\"\n",
    "    shape = data.shape[:-1] + (data.shape[-1] - window_size + 1, window_size)\n",
    "    strides = data.strides + (data.strides[-1],)\n",
    "    return np.lib.stride_tricks.as_strided(data, shape=shape, strides=strides)\n",
    "\n",
    "\n",
    "def __cvm(subindices, total_events):\n",
    "    \"\"\"\n",
    "    Compute Cramer-von Mises metric.\n",
    "    Compared two distributions, where first is subset of second one.\n",
    "    Assuming that second is ordered by ascending\n",
    "\n",
    "    :param subindices: indices of events which will be associated with the first distribution\n",
    "    :param total_events: count of events in the second distribution\n",
    "    :return: cvm metric\n",
    "    \"\"\"\n",
    "    target_distribution = np.arange(1, total_events + 1, dtype='float') / total_events\n",
    "    subarray_distribution = np.cumsum(np.bincount(subindices, minlength=total_events), dtype='float')\n",
    "    subarray_distribution /= 1.0 * subarray_distribution[-1]\n",
    "    return np.mean((target_distribution - subarray_distribution) ** 2)\n",
    "\n",
    "\n",
    "def compute_cvm(predictions, masses, n_neighbours=200, step=50):\n",
    "    \"\"\"\n",
    "    Computing Cramer-von Mises (cvm) metric on background events: take average of cvms calculated for each mass bin.\n",
    "    In each mass bin global prediction's cdf is compared to prediction's cdf in mass bin.\n",
    "\n",
    "    :param predictions: array-like, predictions\n",
    "    :param masses: array-like, in case of Kaggle tau23mu this is reconstructed mass\n",
    "    :param n_neighbours: count of neighbours for event to define mass bin\n",
    "    :param step: step through sorted mass-array to define next center of bin\n",
    "    :return: average cvm value\n",
    "    \"\"\"\n",
    "    predictions = np.array(predictions)\n",
    "    masses = np.array(masses)\n",
    "    assert len(predictions) == len(masses)\n",
    "\n",
    "    # First, reorder by masses\n",
    "    predictions = predictions[np.argsort(masses)]\n",
    "\n",
    "    # Second, replace probabilities with order of probability among other events\n",
    "    predictions = np.argsort(np.argsort(predictions, kind='mergesort'), kind='mergesort')\n",
    "\n",
    "    # Now, each window forms a group, and we can compute contribution of each group to CvM\n",
    "    cvms = []\n",
    "    for window in __rolling_window(predictions, window_size=n_neighbours)[::step]:\n",
    "        cvms.append(__cvm(subindices=window, total_events=len(predictions)))\n",
    "    return np.mean(cvms)\n",
    "\n",
    "\n",
    "def __roc_curve_splitted(data_zero, data_one, sample_weights_zero, sample_weights_one):\n",
    "    \"\"\"\n",
    "    Compute roc curve\n",
    "\n",
    "    :param data_zero: 0-labeled data\n",
    "    :param data_one:  1-labeled data\n",
    "    :param sample_weights_zero: weights for 0-labeled data\n",
    "    :param sample_weights_one:  weights for 1-labeled data\n",
    "    :return: roc curve\n",
    "    \"\"\"\n",
    "    labels = [0] * len(data_zero) + [1] * len(data_one)\n",
    "    weights = np.concatenate([sample_weights_zero, sample_weights_one])\n",
    "    data_all = np.concatenate([data_zero, data_one])\n",
    "    fpr, tpr, _ = roc_curve(labels, data_all, sample_weight=weights)\n",
    "    return fpr, tpr\n",
    "\n",
    "\n",
    "def compute_ks(data_prediction, mc_prediction, weights_data, weights_mc):\n",
    "    \"\"\"\n",
    "    Compute Kolmogorov-Smirnov (ks) distance between real data predictions cdf and Monte Carlo one.\n",
    "\n",
    "    :param data_prediction: array-like, real data predictions\n",
    "    :param mc_prediction: array-like, Monte Carlo data predictions\n",
    "    :param weights_data: array-like, real data weights\n",
    "    :param weights_mc: array-like, Monte Carlo weights\n",
    "    :return: ks value\n",
    "    \"\"\"\n",
    "    assert len(data_prediction) == len(weights_data), 'Data length and weight one must be the same'\n",
    "    assert len(mc_prediction) == len(weights_mc), 'Data length and weight one must be the same'\n",
    "\n",
    "    data_prediction, mc_prediction = np.array(data_prediction), np.array(mc_prediction)\n",
    "    weights_data, weights_mc = np.array(weights_data), np.array(weights_mc)\n",
    "\n",
    "    assert np.all(data_prediction >= 0.) and np.all(data_prediction <= 1.), 'Data predictions are out of range [0, 1]'\n",
    "    assert np.all(mc_prediction >= 0.) and np.all(mc_prediction <= 1.), 'MC predictions are out of range [0, 1]'\n",
    "\n",
    "    weights_data /= np.sum(weights_data)\n",
    "    weights_mc /= np.sum(weights_mc)\n",
    "\n",
    "    fpr, tpr = __roc_curve_splitted(data_prediction, mc_prediction, weights_data, weights_mc)\n",
    "\n",
    "    Dnm = np.max(np.abs(fpr - tpr))\n",
    "    return Dnm\n",
    "\n",
    "\n",
    "def roc_auc_truncated(labels, predictions, tpr_thresholds=(0.2, 0.4, 0.6, 0.8),\n",
    "                      roc_weights=(4, 3, 2, 1, 0)):\n",
    "    \"\"\"\n",
    "    Compute weighted area under ROC curve.\n",
    "\n",
    "    :param labels: array-like, true labels\n",
    "    :param predictions: array-like, predictions\n",
    "    :param tpr_thresholds: array-like, true positive rate thresholds delimiting the ROC segments\n",
    "    :param roc_weights: array-like, weights for true positive rate segments\n",
    "    :return: weighted AUC\n",
    "    \"\"\"\n",
    "    assert np.all(predictions >= 0.) and np.all(predictions <= 1.), 'Data predictions are out of range [0, 1]'\n",
    "    assert len(tpr_thresholds) + 1 == len(roc_weights), 'Incompatible lengths of thresholds and weights'\n",
    "    fpr, tpr, _ = roc_curve(labels, predictions)\n",
    "    area = 0.\n",
    "    tpr_thresholds = [0.] + list(tpr_thresholds) + [1.]\n",
    "    for index in range(1, len(tpr_thresholds)):\n",
    "        tpr_cut = np.minimum(tpr, tpr_thresholds[index])\n",
    "        tpr_previous = np.minimum(tpr, tpr_thresholds[index - 1])\n",
    "        area += roc_weights[index - 1] * (auc(fpr, tpr_cut, reorder=True) - auc(fpr, tpr_previous, reorder=True))\n",
    "    tpr_thresholds = np.array(tpr_thresholds)\n",
    "    # roc auc normalization to be 1 for an ideal classifier\n",
    "    area /= np.sum((tpr_thresholds[1:] - tpr_thresholds[:-1]) * np.array(roc_weights))\n",
    "    return area"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m1854/1854\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m0s\u001b[0m 258us/step\n",
      "AUC BMI: 0.9006468155639382\n",
      "AUC NN: 0.9311125763105179\n",
      "AUC DCls: 0.9508931006528414\n"
     ]
    }
   ],
   "source": [
    "# get model performance\n",
    "pre_bmi = BMI.predict_proba(X_test)\n",
    "pre_nn = NN.predict_proba(X_test)\n",
    "pre_DCls = DCls.predict_proba(X_test)\n",
    "print(f\"AUC BMI: {roc_auc_score(y_test, pre_bmi[:,1])}\")\n",
    "print(f\"AUC NN: {roc_auc_score(y_test, pre_nn[:,1])}\")\n",
    "print(f\"AUC DCls: {roc_auc_score(y_test, pre_DCls)}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[1m6892/6892\u001b[0m \u001b[32m━━━━━━━━━━━━━━━━━━━━\u001b[0m\u001b[37m\u001b[0m \u001b[1m2s\u001b[0m 264us/step\n"
     ]
    }
   ],
   "source": [
    "# prepare agreement data\n",
    "agreement_weight = agreement_test[\"weight\"]\n",
    "agreement_signal = agreement_test[\"signal\"]\n",
    "agreement_test_feature = agreement_test.drop(columns=[\"id\", \"signal\", \"SPDhits\", \"weight\"])\n",
    "\n",
    "agreement_probs_BMI = BMI.predict_proba(agreement_test_feature)\n",
    "agreement_probs_NN = NN.predict_proba(agreement_test_feature)\n",
    "agreement_probs_DCls = DCls.predict_proba(agreement_test_feature)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "KS metric 0.014166066560546434 True\n",
      "KS metric 0.04060012283674952 True\n",
      "KS metric 0.048405255957215676 True\n"
     ]
    }
   ],
   "source": [
    "ks_bmi = compute_ks(\n",
    "    agreement_probs_BMI[:,1][agreement_signal == 0],\n",
    "    agreement_probs_BMI[:,1][agreement_signal == 1],\n",
    "    agreement_weight[agreement_signal == 0],\n",
    "    agreement_weight[agreement_signal == 1]\n",
    ")\n",
    "\n",
    "ks_nn = compute_ks(\n",
    "    agreement_probs_NN[:,1][agreement_signal == 0],\n",
    "    agreement_probs_NN[:,1][agreement_signal == 1],\n",
    "    agreement_weight[agreement_signal == 0],\n",
    "    agreement_weight[agreement_signal == 1]\n",
    ")\n",
    "\n",
    "ks_DCls = compute_ks(\n",
    "    agreement_probs_DCls[agreement_signal == 0].reshape(-1),\n",
    "    agreement_probs_DCls[agreement_signal == 1].reshape(-1),\n",
    "    agreement_weight[agreement_signal == 0],\n",
    "    agreement_weight[agreement_signal == 1]\n",
    ")\n",
    "\n",
    "print ('KS metric', ks_bmi, ks_bmi < 0.09)\n",
    "print ('KS metric', ks_nn, ks_nn < 0.09)\n",
    "print ('KS metric', ks_DCls, ks_DCls < 0.09)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/sd/hnbd9dm54xl_zh0fctx18ls40000gn/T/ipykernel_80885/2491414440.py:130: UserWarning: There are no gridspecs with layoutgrids. Possibly did not call parent GridSpec with the \"figure\" keyword\n",
      "  plt.savefig(\"source_target.pdf\")\n",
      "/var/folders/sd/hnbd9dm54xl_zh0fctx18ls40000gn/T/ipykernel_80885/2491414440.py:130: UserWarning: There are no gridspecs with layoutgrids. Possibly did not call parent GridSpec with the \"figure\" keyword\n",
      "  plt.savefig(\"source_target.pdf\")\n",
      "/opt/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:152: UserWarning: There are no gridspecs with layoutgrids. Possibly did not call parent GridSpec with the \"figure\" keyword\n",
      "  fig.canvas.print_figure(bytes_io, **kw)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 444x377.872 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# make some plots\n",
    "fsize = 12\n",
    "fsize_annotate = 10\n",
    "\n",
    "std_figsize = (1.2 * 3.7, 1.6 * 2.3617)\n",
    "std_axes_form = [0.16, 0.15, 0.81, 0.76]\n",
    "\n",
    "electron_color = (0.85, 0.39, 0.14)\n",
    "positron_color = (0.16, 0.37, 0.65)\n",
    "# ax_colors = \"#CFCFCF\"\n",
    "ax_colors = \"darkgrey\"\n",
    "text_color = \"grey\"\n",
    "\n",
    "def std_fig(ax_form=std_axes_form, figsize=std_figsize, rasterized=False):\n",
    "    rcparams = {\n",
    "        \"axes.labelsize\": fsize,\n",
    "        \"xtick.labelsize\": fsize,\n",
    "        \"ytick.labelsize\": fsize,\n",
    "        \"figure.figsize\": std_figsize,\n",
    "        \"legend.frameon\": False,\n",
    "        \"legend.loc\": \"best\",\n",
    "    }\n",
    "    plt.rcParams[\"text.latex.preamble\"] = r\"\\usepackage{amsmath}\\usepackage{amssymb}\"\n",
    "    rc(\"font\", **{\"family\": \"serif\", \"serif\": [\"Computer Modern\"]})\n",
    "    rc(\"text\", usetex=True)\n",
    "    rcParams.update(rcparams)\n",
    "    mpl.rcParams[\"hatch.linewidth\"] = 0.25\n",
    "    fig = plt.figure(figsize=figsize, constrained_layout=True)\n",
    "    ax = fig.add_axes(ax_form, rasterized=rasterized)\n",
    "    ax.patch.set_alpha(0.0)\n",
    "\n",
    "    return fig, ax\n",
    "\n",
    "fig, ax = std_fig()\n",
    "BINS = 10000\n",
    "WIDTH = 0.6\n",
    "LW = 1.3\n",
    "\n",
    "kwargs = {\n",
    "    \"lw\": LW,\n",
    "    \"linestyle\": \"--\",\n",
    "    \"histtype\": \"step\",\n",
    "    \"rwidth\": WIDTH,\n",
    "    \"facecolor\": \"deeppink\",\n",
    "}\n",
    "kwargs[\"label\"] = r\"simulation ($\\gamma=0$)\"\n",
    "ax.hist(\n",
    "    agreement_probs_NN[agreement_signal == 0][:,1],\n",
    "    bins=BINS,\n",
    "    density=True,\n",
    "    weights=agreement_weight[agreement_signal == 0],\n",
    "    edgecolor=\"deeppink\",\n",
    "    **kwargs\n",
    ")\n",
    "kwargs[\"label\"] = r\"data ($\\gamma=0$)\"\n",
    "ax.hist(\n",
    "    agreement_probs_NN[agreement_signal == 1][:,1],\n",
    "    bins=BINS,\n",
    "    density=True,\n",
    "    weights=agreement_weight[agreement_signal == 1],\n",
    "    edgecolor=\"steelblue\",\n",
    "    **kwargs\n",
    ")\n",
    "\n",
    "kwargs = {\n",
    "    \"lw\": LW,\n",
    "    \"histtype\": \"step\",\n",
    "    \"rwidth\": WIDTH,\n",
    "    \"facecolor\": \"deeppink\",\n",
    "}\n",
    "kwargs[\"label\"] = r\"simulation ($\\gamma=0.2$)\"\n",
    "ax.hist(\n",
    "    agreement_probs_BMI[agreement_signal == 0][:,1],\n",
    "    bins=BINS,\n",
    "    density=True,\n",
    "    weights=agreement_weight[agreement_signal == 0],\n",
    "    edgecolor=\"deeppink\",\n",
    "    **kwargs\n",
    ")\n",
    "kwargs[\"label\"] = r\"data ($\\gamma=0.2$)\"\n",
    "ax.hist(\n",
    "    agreement_probs_BMI[agreement_signal == 1][:,1],\n",
    "    bins=BINS,\n",
    "    density=True,\n",
    "    weights=agreement_weight[agreement_signal == 1],\n",
    "    edgecolor=\"steelblue\",\n",
    "    **kwargs\n",
    ")\n",
    "\n",
    "kwargs = {\n",
    "    \"lw\": LW,\n",
    "    \"linestyle\": \":\",\n",
    "    \"histtype\": \"step\",\n",
    "    \"rwidth\": WIDTH,\n",
    "    \"facecolor\": \"deeppink\",\n",
    "}\n",
    "kwargs[\"label\"] = r\"simulation (Adv. Cls.)\"\n",
    "ax.hist(\n",
    "    agreement_probs_DCls[agreement_signal == 0].reshape(-1),\n",
    "    bins=BINS,\n",
    "    density=True,\n",
    "    weights=agreement_weight[agreement_signal == 0],\n",
    "    edgecolor=\"deeppink\",\n",
    "    **kwargs\n",
    ")\n",
    "kwargs[\"label\"] = r\"data (Adv. Cls.)\"\n",
    "ax.hist(\n",
    "    agreement_probs_DCls[agreement_signal == 1].reshape(-1),\n",
    "    bins=BINS,\n",
    "    density=True,\n",
    "    weights=agreement_weight[agreement_signal == 1],\n",
    "    edgecolor=\"steelblue\",\n",
    "    **kwargs\n",
    ")\n",
    "\n",
    "ax.set_xlabel(r\"output\", fontsize=12)\n",
    "ax.set_ylabel(r\"\\#\", fontsize=12)\n",
    "ax.grid(alpha=1, lw=0.25, zorder=-10, color=\"grey\")\n",
    "ax.set_xscale(\"log\")\n",
    "\n",
    "leg1 = ax.legend(\n",
    "    frameon=False,\n",
    "    fontsize=10,\n",
    "    ncol=1,\n",
    "    handlelength=1,\n",
    "    columnspacing=2,\n",
    "    markerfirst=True,\n",
    "    loc=\"upper right\",\n",
    ")\n",
    "plt.savefig(\"source_target.pdf\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "metadata": {},
   "outputs": [],
   "source": [
    "# prepare correlation data\n",
    "check_correlation_features = correlation.drop(['id', 'mass', 'SPDhits'], axis=1)\n",
    "correlation_probs_BMI = BMI.predict_proba(check_correlation_features)\n",
    "correlation_probs_NN = NN.predict_proba(check_correlation_features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 154,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CvM metric 0.0008627846393993524 True\n",
      "CvM metric 0.0008137530980162971 True\n"
     ]
    }
   ],
   "source": [
    "cvm_BMI = compute_cvm(correlation_probs_BMI[:,1], correlation['mass'])\n",
    "cvm_NN = compute_cvm(correlation_probs_NN[:,1], correlation['mass'])\n",
    "print ('CvM metric', cvm_BMI, cvm_BMI < 0.002)\n",
    "print ('CvM metric', cvm_NN, cvm_NN < 0.002)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 191,
   "metadata": {},
   "outputs": [],
   "source": [
    "# performe some gamma experiments\n",
    "trained_bmi = []\n",
    "for gamma in np.linspace(0, 1, 20):\n",
    "    BMI = BinaryMI(\n",
    "        kernel_regularizer = 0.0,\n",
    "        batch_size=32,\n",
    "        hidden_layers = [64,32],\n",
    "        quantized_position=[True, False],\n",
    "        input_shape=(X_train.shape[1],),\n",
    "        verbose=0,\n",
    "        gamma=gamma,\n",
    "        print_summary=False,\n",
    "        last_layer_size=2,\n",
    "        run_eagerly=False\n",
    "    )\n",
    "    train_hist_BMI = BMI.fit(x_train=X_train, y_train=y_train, s_train=S_train)\n",
    "    trained_bmi.append(BMI)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 192,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get evaluation\n",
    "auc_list = []\n",
    "ks_list = []\n",
    "for bmi in trained_bmi:\n",
    "    pre_bmi = bmi.predict_proba(X_test)\n",
    "    agreement_probs_BMI = bmi.predict_proba(agreement_test_feature)\n",
    "    auc_list.append(roc_auc_score(y_test, pre_bmi[:,1]))\n",
    "    ks_list.append(compute_ks(\n",
    "        agreement_probs_BMI[:,1][agreement_signal == 0],\n",
    "        agreement_probs_BMI[:,1][agreement_signal == 1],\n",
    "        agreement_weight[agreement_signal == 0],\n",
    "        agreement_weight[agreement_signal == 1]\n",
    "    ))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 195,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/var/folders/sd/hnbd9dm54xl_zh0fctx18ls40000gn/T/ipykernel_67096/3033064254.py:46: UserWarning: There are no gridspecs with layoutgrids. Possibly did not call parent GridSpec with the \"figure\" keyword\n",
      "  plt.savefig(\"gamma.pdf\")\n",
      "/opt/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:152: UserWarning: There are no gridspecs with layoutgrids. Possibly did not call parent GridSpec with the \"figure\" keyword\n",
      "  fig.canvas.print_figure(bytes_io, **kw)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 444x377.872 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "def get_color_line(x, y, z, ax_kwargs, s, linestyle='solid', norm=True):\n",
    "    xnew = np.linspace(min(x), max(x), 3*len(z))\n",
    "    cs = UnivariateSpline(x, y, s=s)\n",
    "    ysmooth= cs(xnew)\n",
    "\n",
    "    ax_kwargs.scatter(x, y, zorder=1)\n",
    "\n",
    "    znew = []\n",
    "    for v in z:\n",
    "        for i in range(3):\n",
    "            znew.append(v)\n",
    "    x = xnew\n",
    "    y = ysmooth\n",
    "    z = znew\n",
    "\n",
    "    points = np.array([x, y]).T.reshape(-1, 1, 2)\n",
    "    segments = np.concatenate([points[:-1], points[1:]], axis=1)\n",
    "    if norm:\n",
    "        norm = plt.Normalize(min(z), max(z))\n",
    "        lc = LineCollection(segments, cmap='viridis', norm=norm, linestyle=linestyle)\n",
    "    else:\n",
    "        lc = LineCollection(segments, cmap='viridis', linestyle=linestyle)\n",
    "\n",
    "    # Set the values used for colormapping\n",
    "    lc.set_array(z)\n",
    "    lc.set_linewidth(2)\n",
    "    line = ax_kwargs.add_collection(lc)\n",
    "    return line\n",
    "\n",
    "fig, ax = std_fig()\n",
    "x = np.array(auc_list)\n",
    "y = 1-np.array(ks_list)\n",
    "z = np.linspace(0, 1, 20)\n",
    "\n",
    "y = y[x>0.6]\n",
    "z = z[x>0.6]\n",
    "x = x[x>0.6]\n",
    "x, y, z = zip(*sorted(zip(x, y, z)))\n",
    "line = get_color_line(x, y, z, ax, s=0.1, linestyle=\"-\", norm=False)\n",
    "\n",
    "plt.xlabel(\"AUC\")\n",
    "plt.ylabel(f\"1-KS\")\n",
    "cbar = fig.colorbar(line)\n",
    "cbar.set_label(f\"$\\gamma$\")\n",
    "cbar.formatter.set_powerlimits((0, 0))\n",
    "plt.savefig(\"gamma.pdf\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Simulation of the Bernoulli distribution\n",
    "=====\n",
    "```VHDL\n",
    "library ieee;\n",
    "use ieee.std_logic_1164.all;\n",
    "\n",
    "entity linear_shift is\n",
    "generic (\n",
    "    g_min : positive := 174;\n",
    "    g_max : positive := 1048509;\n",
    "    g_m : positive := 20;\n",
    "    g_poly : std_logic_vector := \"11000000001100000000\" -- x^7+x^6+1\n",
    ");\n",
    "port (\n",
    "    i_sync_reset    : in    std_logic;\n",
    "    i_seed          : in    std_logic_vector (g_m-1 downto 0);\n",
    "    i_en            : in    std_logic;\n",
    "    i_activation    : in    std_logic_vector (9 downto 0);\n",
    "    o_bernoulli     : out   std_logic;\n",
    "    o_lfsr          : out   std_logic_vector (g_m-1 downto 0);\n",
    "\n",
    "    i_reset_n       : in    std_logic;\n",
    "    i_clk           : in    std_logic--;\n",
    ");\n",
    "end entity;\n",
    "\n",
    "architecture rtl of linear_shift is\n",
    "\n",
    "    signal r_lfsr : std_logic_vector (g_m downto 1);\n",
    "    signal w_mask : std_logic_vector (g_m downto 1);\n",
    "    signal w_poly : std_logic_vector (g_m downto 1);\n",
    "\n",
    "begin\n",
    "\n",
    "    o_lfsr <= r_lfsr(g_m downto 1);\n",
    "    w_poly <= g_poly;\n",
    "    g_mask : for k in g_m downto 1 generate\n",
    "        w_mask(k) <= w_poly(k) and r_lfsr(1);\n",
    "    end generate;\n",
    "\n",
    "    process(i_clk, i_reset_n)\n",
    "    begin\n",
    "    if ( i_reset_n /= '1' ) then\n",
    "        r_lfsr <= (others => '1');\n",
    "    elsif rising_edge(i_clk) then\n",
    "        if ( i_sync_reset = '1' ) then\n",
    "            r_lfsr <= i_seed;\n",
    "        elsif (i_en = '1') then\n",
    "            r_lfsr <= '0' & r_lfsr(g_m downto 2) xor w_mask;\n",
    "        end if;\n",
    "    end if;\n",
    "    end process;\n",
    "\n",
    "end architecture;\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 98,
   "metadata": {},
   "outputs": [],
   "source": [
    "# run some analysis\n",
    "def run_lfsr(w_poly, r_lfsr, g_m, i_seed, min_value=0, max_value=0, activation=0):\n",
    "    bernoulli = []\n",
    "    states = []\n",
    "    counter = 0\n",
    "    while True:\n",
    "        # get the outer asignement first\n",
    "        w_mask_list = []\n",
    "        for bit in range(g_m):\n",
    "            w_mask_list.append(((w_poly >> bit) & 1) and ((r_lfsr >> 1) & 1))\n",
    "        w_mask = int(\"\".join(map(str, reversed(w_mask_list))), 2)\n",
    "\n",
    "        # do the actual shift\n",
    "        r_lfsr = r_lfsr >> 1\n",
    "        r_lfsr_list = []\n",
    "        for bit in range(g_m):\n",
    "            r_lfsr_list.append(((r_lfsr >> bit) & 1) ^ ((w_mask >> bit) & 1))\n",
    "        r_lfsr = int(\"\".join(map(str, reversed(r_lfsr_list))), 2)\n",
    "\n",
    "        # sometimes we dont find a good LFSR then we stop\n",
    "        if counter == 100000:\n",
    "            break\n",
    "        if r_lfsr == i_seed:\n",
    "            break\n",
    "        states.append(r_lfsr)\n",
    "\n",
    "        counter += 1\n",
    "        if max_value - min_value == 0: continue\n",
    "\n",
    "        # we use a ROM here to get the pre-computed normed values \n",
    "        if activation <= ((r_lfsr - min_value) / (max_value - min_value)):\n",
    "            bernoulli.append(0)\n",
    "        else:\n",
    "            bernoulli.append(1)\n",
    "\n",
    "    return states, bernoulli\n",
    "\n",
    "g_poly_list = [1,1,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,0]\n",
    "i_seed_list = [0,0,1,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0]\n",
    "x = []\n",
    "y = []\n",
    "yerr = []\n",
    "for i in range(7, 20):\n",
    "    # set default values\n",
    "    g_m = i\n",
    "    g_poly = int(\"\".join(map(str, g_poly_list[:i])), 2)\n",
    "    w_poly = g_poly\n",
    "    i_seed = int(\"\".join(map(str, i_seed_list[:i])), 2)\n",
    "    r_lfsr = i_seed\n",
    "\n",
    "    # run the LFSR\n",
    "    states, _ = run_lfsr(w_poly, r_lfsr, g_m, i_seed)\n",
    "    xi = []\n",
    "    yi = []\n",
    "    zi = []\n",
    "    for i in range(10):\n",
    "        activation = random.uniform(0, 1)\n",
    "        _, bernoulli = run_lfsr(w_poly, r_lfsr, g_m, i_seed, min(states), max(states), activation=activation)\n",
    "\n",
    "        # run random generator\n",
    "        bernoulli_random = []\n",
    "        for i in states:\n",
    "            if activation <= random.uniform(0, 1):\n",
    "                bernoulli_random.append(0)\n",
    "            else:\n",
    "                bernoulli_random.append(1)\n",
    "        xi.append(sum(bernoulli)/len(bernoulli))\n",
    "        yi.append(sum(bernoulli_random)/len(bernoulli_random))\n",
    "        zi.append(len(states))\n",
    "    x.append(np.mean(zi))\n",
    "    y.append(np.mean(np.array(xi)-np.array(yi)))\n",
    "    yerr.append(np.std(np.array(xi)-np.array(yi)))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/opt/anaconda3/lib/python3.11/site-packages/IPython/core/pylabtools.py:152: UserWarning: There are no gridspecs with layoutgrids. Possibly did not call parent GridSpec with the \"figure\" keyword\n",
      "  fig.canvas.print_figure(bytes_io, **kw)\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 444x377.872 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# plot\n",
    "fig, ax = std_fig()\n",
    "\n",
    "ax.scatter(x, y)\n",
    "ax.errorbar(x, y, yerr=yerr, alpha=0.3, fmt='none', color=\"gray\")\n",
    "ax.set_xscale(\"log\")\n",
    "\n",
    "ax.set_xlabel(r\"LFSR states\", fontsize=12)\n",
    "ax.set_ylabel(r\"LFSR bernoulli - random bernoulli\", fontsize=12)\n",
    "plt.savefig(\"lfsr.pdf\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "base",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}