InterpretableML.with.LIME.TabularData.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "<img src=\"https://www.th-koeln.de/img/logo.svg\" style=\"float:right;\" width=\"200\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Use Case: <font color=\"#C70039\">Interpretable Machine Learning with LIME for tabular data</font>\n",
    "* Course: AML\n",
    "* Lecturer: <a href=\"https://www.gernotheisenberg.de/\">Gernot Heisenberg</a>\n",
    "* Author of notebook: <a href=\"https://www.gernotheisenberg.de/\">Gernot Heisenberg</a>\n",
    "* Date:   28.08.2023\n",
    "\n",
    "<img src=\"https://miro.medium.com/max/664/1*J1V-RIBHIcX-Aej0x7UXnA.png\" style=\"float: center;\" width=\"600\">\n",
    "\n",
    "---------------------------------\n",
    "\n",
    "### Description \n",
    "This is one implementation example for LIME interpreting a random forest ensemble model that predicts tablua data. \n",
    "PLease try to understand the implementation.\n",
    "\n",
    "---------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "L1MnrZSd06ED",
    "tags": []
   },
   "source": [
    "### Imports\n",
    "Import all necessary python utilities for loading, preprocessing and predicting the data. \n",
    "\n",
    "#### NOTE\n",
    "In some cases you need to downgrade to scikit-learn-1.2.2 version to get it running"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 54
    },
    "id": "uq9q-_zXgqtV",
    "outputId": "3a4be885-98a1-4d53-843e-8b9ca05c0ab5",
    "scrolled": true,
    "tags": []
   },
   "outputs": [],
   "source": [
    "from utils import DataLoader\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.metrics import f1_score, accuracy_score\n",
    "from interpret.blackbox import LimeTabular\n",
    "from interpret import show\n",
    "\n",
    "# get rid of those LIME warnings\n",
    "import warnings\n",
    "warnings.filterwarnings('ignore')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Load and preprocess the data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "owMeom1L-Afp"
   },
   "source": [
    "## load the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "data_loader = DataLoader() \n",
    "data_loader.load_dataset() \n",
    "data_loader.preprocess_data()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Split the data for evaluation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "X_train, X_test, y_train, y_test = data_loader.get_data_split()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Oversample the train data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train.shape (7776, 21)\n",
      "X_test.shape (1022, 21)\n"
     ]
    }
   ],
   "source": [
    "X_train, y_train = data_loader.oversample(X_train, y_train) \n",
    "print(\"X_train.shape\", X_train.shape) \n",
    "print(\"X_test.shape\", X_test.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "X_84p1vkjSJa"
   },
   "source": [
    "### Fit the blackbox model (Random Forest Classifier) to the stroke data\n",
    "For the case of image explanations, perturbations will be generated by turning on and off some of the superpixels in the image."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 Score 0.5153574066156178\n",
      "Accuracy 0.9403131115459883\n"
     ]
    }
   ],
   "source": [
    "rf = RandomForestClassifier()\n",
    "rf.fit(X_train, y_train)\n",
    "y_pred = rf.predict(X_test)\n",
    "\n",
    "print(f\"F1 Score {f1_score(y_test, y_pred, average='macro')}\")\n",
    "print(f\"Accuracy {accuracy_score(y_test, y_pred)}\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Apply LIME\n",
    "Initilize Lime for being used with tabular data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "lime = LimeTabular(predict_fn=rf.predict_proba, data=X_train, random_state=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Get local explanations and interpret the results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<!-- http://127.0.0.1:7001/2653934568064/ -->\n",
       "<iframe src=\"http://127.0.0.1:7001/2653934568064/\" width=100% height=800 frameBorder=\"0\"></iframe>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "lime_local = lime.explain_local(X_test[-10:], y_test[-10:], name='LIME')\n",
    "show(lime_local)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "colab": {
   "collapsed_sections": [],
   "name": "lime_image.ipynb",
   "provenance": []
  },
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "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": "3.9.17"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}