{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true,
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "# Applying data minimization to a trained regression ML model"
   ]
  },
  {
   "cell_type": "markdown",
   "source": [
    "In this tutorial we will show how to perform data minimization for regression ML models using the minimization module.\n",
    "\n",
    "We will show you applying data minimization to a different trained regression models."
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Load data\n",
    "QI parameter determines which features will be minimized."
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "outputs": [],
   "source": [
    "from sklearn.datasets import load_diabetes\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "dataset = load_diabetes()\n",
    "X_train, X_test, y_train, y_test = train_test_split(dataset.data, dataset.target, test_size=0.5, random_state=14)\n",
    "\n",
    "features = ['age', 'sex', 'bmi', 'bp',\n",
    "                's1', 's2', 's3', 's4', 's5', 's6']\n",
    "QI = [0, 2, 5, 8, 9]"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Train DecisionTreeRegressor model"
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Base model accuracy (R2 score):  0.15014421352446072\n"
     ]
    }
   ],
   "source": [
    "from apt.minimization import GeneralizeToRepresentative\n",
    "from sklearn.tree import DecisionTreeRegressor\n",
    "\n",
    "model1 = DecisionTreeRegressor(random_state=10, min_samples_split=2)\n",
    "model1.fit(X_train, y_train)\n",
    "print('Base model accuracy (R2 score): ', model1.score(X_test, y_test))"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Run minimization\n",
    "We will try to run minimization with only a subset of the features."
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initial accuracy of model on generalized data, relative to original model predictions (base generalization derived from tree, before improvements): 0.108922\n",
      "Improving accuracy\n",
      "feature to remove: s5\n",
      "Removed feature: s5, new relative accuracy: 0.505498\n",
      "feature to remove: s6\n",
      "Removed feature: s6, new relative accuracy: 0.404757\n",
      "feature to remove: bmi\n",
      "Removed feature: bmi, new relative accuracy: 0.718978\n",
      "Accuracy on minimized data:  0.11604533946025941\n",
      "generalizations:  {'ranges': {'age': [-0.07090024650096893, -0.043656209483742714, -0.041839939542114735, -0.03639113181270659, -0.01459590089507401, -0.012779632292222232, -0.009147093165665865, -0.0036982858437113464, 0.03989217430353165, 0.039892176166176796, 0.05623859912157059, 0.06713621318340302], 's2': [-0.0550188384950161, -0.0285577941685915, -0.024643437936902046, -0.02135537937283516, -0.013683241792023182, -0.006480826530605555, 0.009176596067845821, 0.023111702874302864, 0.02420772146433592, 0.02655633445829153, 0.039082273840904236]}, 'categories': {}, 'untouched': ['s3', 'bmi', 's6', 'bp', 's4', 's5', 'sex', 's1']}\n"
     ]
    }
   ],
   "source": [
    "# note that is_regression param is True\n",
    "\n",
    "minimizer1 = GeneralizeToRepresentative(model1, target_accuracy=0.7, features=features, is_regression=True,\n",
    "                                    features_to_minimize=QI)\n",
    "\n",
    "# Fitting the minimizar can be done either on training or test data. Doing it with test data is better as the\n",
    "# resulting accuracy on test data will be closer to the desired target accuracy (when working with training\n",
    "# data it could result in a larger gap)\n",
    "# Don't forget to leave a hold-out set for final validation!\n",
    "X_generalizer_train1, x_test1, y_generalizer_train1, y_test1 = train_test_split(X_test, y_test,\n",
    "                                                                test_size = 0.4, random_state = 38)\n",
    "\n",
    "x_train_predictions1 = model1.predict(X_generalizer_train1)\n",
    "minimizer1.fit(X_generalizer_train1, x_train_predictions1)\n",
    "transformed1 = minimizer1.transform(x_test1)\n",
    "print('Accuracy on minimized data: ', model1.score(transformed1, y_test1))\n",
    "print('generalizations: ',minimizer1.generalizations_)#%% md"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Train linear regression model"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "source": [
    "from sklearn.linear_model import LinearRegression\n",
    "from apt.minimization import GeneralizeToRepresentative\n",
    "\n",
    "model2 = LinearRegression()\n",
    "model2.fit(X_train, y_train)\n",
    "print('Base model accuracy (R2 score): ', model2.score(X_test, y_test))"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": null,
   "outputs": []
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Run minimization\n",
    "We will try to run minimization with only a subset of the features."
   ],
   "metadata": {
    "collapsed": false
   }
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initial accuracy of model on generalized data, relative to original model predictions (base generalization derived from tree, before improvements): 0.225782\n",
      "Improving accuracy\n",
      "feature to remove: age\n",
      "Removed feature: age, new relative accuracy: 0.223565\n",
      "feature to remove: s2\n",
      "Removed feature: s2, new relative accuracy: 0.759788\n",
      "Accuracy on minimized data:  0.4414329261774286\n",
      "generalizations:  {'ranges': {'bmi': [-0.0660245232284069, -0.06171327643096447, -0.048779530450701714, -0.036923596635460854, -0.022912041284143925, -0.015906263142824173, -0.009978296235203743, 0.007266696775332093, 0.022356065921485424, 0.028822937980294228, 0.04499012045562267, 0.053073709830641747, 0.10103634744882584], 's5': [-0.08940735459327698, -0.07823517918586731, -0.07310866191983223, -0.07022909820079803, -0.06740894541144371, -0.06558558344841003, -0.041897499933838844, -0.04049498960375786, -0.03781316243112087, -0.033939776942133904, -0.03263746201992035, -0.02538660168647766, -0.023219254799187183, -0.017585186287760735, -0.016525186598300934, -0.008522996446117759, 0.0015758189256303012, 0.012934560421854258, 0.014069339726120234, 0.015929921995848417, 0.01947084255516529, 0.028651678003370762, 0.03358383011072874, 0.03639278281480074, 0.041416410356760025, 0.06386702693998814], 's6': [-0.07356456853449345, -0.052854035049676895, -0.048711927607655525, -0.0383566590026021, -0.02800139266764745, -0.021788232028484344, -0.007290858076885343, -0.007290857844054699, 0.017561784014105797, 0.02377494378015399, 0.02791705122217536, 0.02998810407007113, 0.054840744473040104]}, 'categories': {}, 'untouched': ['s2', 's3', 'bp', 's4', 'age', 'sex', 's1']}\n"
     ]
    }
   ],
   "source": [
    "# note that is_regression param is True\n",
    "\n",
    "minimizer2 = GeneralizeToRepresentative(model2, target_accuracy=0.7, features=features, is_regression=True,\n",
    "                                    features_to_minimize=QI)\n",
    "\n",
    "# Fitting the minimizar can be done either on training or test data. Doing it with test data is better as the\n",
    "# resulting accuracy on test data will be closer to the desired target accuracy (when working with training\n",
    "# data it could result in a larger gap)\n",
    "# Don't forget to leave a hold-out set for final validation!\n",
    "X_generalizer_train2, x_test2, y_generalizer_train2, y_test2 = train_test_split(X_test, y_test,\n",
    "                                                                test_size = 0.4, random_state = 38)\n",
    "\n",
    "x_train_predictions2 = model2.predict(X_generalizer_train2)\n",
    "minimizer2.fit(X_generalizer_train2, x_train_predictions2)\n",
    "transformed2 = minimizer2.transform(x_test2)\n",
    "print('Accuracy on minimized data: ', model2.score(transformed2, y_test2))\n",
    "print('generalizations: ',minimizer2.generalizations_)"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}