3_xai_and_dimensionality_reduction.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:03:58.269802Z",
     "end_time": "2023-04-15T19:04:04.756165Z"
    }
   },
   "outputs": [],
   "source": [
    "# TO BE DELETED\n",
    "\n",
    "import sys\n",
    "from typing import Iterable, Tuple\n",
    "\n",
    "import numpy as np\n",
    "\n",
    "sys.path.append(\"src\")\n",
    "\n",
    "import utils"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "outputs": [],
   "source": [
    "data = utils.read_data(\"train\")"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:04:04.756814Z",
     "end_time": "2023-04-15T19:04:10.697812Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "outputs": [],
   "source": [
    "min_, mean_, max_, var_, skw_, kurt_ = [np.empty((utils.meta.train_instances, utils.meta.n_columns)) for _ in range(6)]\n",
    "\n",
    "for i_d, d in enumerate(data.values()):\n",
    "    ts = d[\"ts\"]\n",
    "    min_[i_d] = ts.min(axis=0)\n",
    "    mean_[i_d] = ts.mean(axis=0)\n",
    "    max_[i_d] = ts.max(axis=0)\n",
    "    var_[i_d] = ts.var(axis=0)\n",
    "    skw_[i_d] = ts.skew(axis=0)\n",
    "    kurt_[i_d] = ts.kurt(axis=0)"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:04:10.699612Z",
     "end_time": "2023-04-15T19:04:11.345074Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "outputs": [],
   "source": [
    "classes = [1 if d[\"class\"] == \"a\" else 0 for d in data.values()]"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:04:11.346247Z",
     "end_time": "2023-04-15T19:04:11.347594Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9419354838709677\n",
      "0.9433962264150944\n"
     ]
    }
   ],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import f1_score\n",
    "\n",
    "X = np.hstack((min_, mean_, max_, var_, skw_, kurt_))\n",
    "y = classes\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)\n",
    "\n",
    "model = LogisticRegression(max_iter=1000).fit(X_train, y_train)\n",
    "print(model.score(X_test, y_test))\n",
    "print(f1_score(y_test, model.predict(X_test)))"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:04:11.349567Z",
     "end_time": "2023-04-15T19:04:11.771643Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "outputs": [],
   "source": [
    "def idx_to_sensor_statistic(idx: int, moments: Iterable[str] = (\n",
    "        \"minimum\", \"mean\", \"maximum\", \"variance\", \"skewness\", \"kurtosis\")) -> Tuple[str, str]:\n",
    "    sensor_id = idx % utils.meta.n_columns\n",
    "    sensor_name = utils.meta.column_names[sensor_id]\n",
    "    moment_id = idx // utils.meta.n_columns\n",
    "    moment_name = moments[moment_id]\n",
    "    return sensor_name, moment_name"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:04:11.771878Z",
     "end_time": "2023-04-15T19:04:11.775495Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "top:    1, coef:  0.2059, sensor:   F4, stat: variance\n",
      "top:    2, coef: -0.1772, sensor:  AF1, stat: variance\n",
      "top:    3, coef:  0.1724, sensor:  FC5, stat: variance\n",
      "top:    4, coef:  0.1647, sensor:  FP2, stat: variance\n",
      "top:    5, coef:  0.1637, sensor:   P8, stat: variance\n",
      "top:    6, coef:  0.1564, sensor:  PO2, stat: variance\n",
      "top:    7, coef:  0.1487, sensor:   F8, stat: variance\n",
      "top:    8, coef: -0.1471, sensor:   C2, stat: maximum\n",
      "top:    9, coef: -0.1447, sensor:  PO8, stat: variance\n",
      "top:   10, coef: -0.1430, sensor:  AF2, stat: variance\n",
      "top:   11, coef:  0.1410, sensor:   F8, stat: maximum\n",
      "top:   12, coef:  0.1385, sensor:  AF7, stat: variance\n",
      "top:   13, coef: -0.1354, sensor:  FPZ, stat: variance\n",
      "top:   14, coef: -0.1318, sensor:   CZ, stat: maximum\n",
      "top:   15, coef: -0.1317, sensor:   O2, stat: variance\n",
      "top:   16, coef: -0.1304, sensor:  AFZ, stat: variance\n",
      "top:   17, coef:  0.1293, sensor:  CP6, stat: variance\n",
      "top:   18, coef: -0.1280, sensor:  FT8, stat: variance\n",
      "top:   19, coef: -0.1261, sensor:   P5, stat: variance\n",
      "top:   20, coef: -0.1246, sensor:   P3, stat: variance\n"
     ]
    }
   ],
   "source": [
    "idcs_sorted = np.argsort(np.abs(model.coef_))\n",
    "for i_c, idx in enumerate(idcs_sorted.T[::-1]):\n",
    "    if i_c >= 20:\n",
    "        break\n",
    "    sensor, moment = idx_to_sensor_statistic(int(idx))\n",
    "    print(f\"top: {i_c + 1:4d}, coef: {float(model.coef_[0, idx]): 2.4f}, sensor: {sensor:>4}, stat: {moment}\")"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:04:11.775721Z",
     "end_time": "2023-04-15T19:04:11.817573Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9225806451612903\n",
      "0.9240506329113924\n"
     ]
    }
   ],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import f1_score\n",
    "\n",
    "X = var_\n",
    "y = classes\n",
    "\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)\n",
    "\n",
    "model = LogisticRegression(max_iter=int(1e5)).fit(X_train, y_train)\n",
    "print(model.score(X_test, y_test))\n",
    "print(f1_score(y_test, model.predict(X_test)))"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:04:11.798885Z",
     "end_time": "2023-04-15T19:04:12.071691Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "outputs": [],
   "source": [
    "from sklearn.feature_selection import SelectFromModel\n",
    "\n",
    "f1_hist = []\n",
    "\n",
    "for n_features in range(1, X.shape[1]):\n",
    "    selector = SelectFromModel(estimator=LogisticRegression(max_iter=int(1e5)), threshold=0,\n",
    "                               max_features=n_features).fit(X, y)\n",
    "    indices = selector.get_support()\n",
    "    indices = np.where(indices)[0]\n",
    "    X_n_train, X_n_test = X_train[:, indices], X_test[:, indices]\n",
    "    model = LogisticRegression(max_iter=int(1e5)).fit(X_n_train, y_train)\n",
    "    f1_hist.append(f1_score(y_test, model.predict(X_n_test)))"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:04:12.070974Z",
     "end_time": "2023-04-15T19:04:57.349810Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "outputs": [
    {
     "data": {
      "text/plain": "<Figure size 640x480 with 1 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of features yielding solution of best quality (argmax): 29\n"
     ]
    }
   ],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "\n",
    "plt.plot(f1_hist)\n",
    "plt.xlabel(\"Number of features\")\n",
    "plt.ylabel(\"F1\")\n",
    "plt.show()\n",
    "\n",
    "print(f\"Number of features yielding solution of best quality (argmax): {np.argmax(f1_hist)}\")"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:04:57.349718Z",
     "end_time": "2023-04-15T19:04:57.629019Z"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "accuracy: 0.9548387096774194\n",
      "f1: 0.9554140127388535\n",
      "===\n",
      "top:    1, coef: -0.9789, sensor:   C3, stat: variance\n",
      "top:    2, coef:  0.8989, sensor:  CP4, stat: variance\n",
      "top:    3, coef: -0.8882, sensor:  AF8, stat: variance\n",
      "top:    4, coef: -0.8131, sensor:   C2, stat: variance\n",
      "top:    5, coef: -0.8049, sensor:   F7, stat: variance\n",
      "top:    6, coef:  0.7357, sensor:   C4, stat: variance\n",
      "top:    7, coef:  0.7249, sensor:   C6, stat: variance\n",
      "top:    8, coef:  0.7135, sensor:  CP1, stat: variance\n",
      "top:    9, coef: -0.7012, sensor:   F4, stat: variance\n",
      "top:   10, coef:  0.6367, sensor:   F8, stat: variance\n",
      "top:   11, coef:  0.5833, sensor:   C1, stat: variance\n",
      "top:   12, coef: -0.5662, sensor:  AF2, stat: variance\n",
      "top:   13, coef: -0.5347, sensor:   F1, stat: variance\n",
      "top:   14, coef:  0.5064, sensor:  FC2, stat: variance\n",
      "top:   15, coef:  0.3535, sensor:   CZ, stat: variance\n",
      "top:   16, coef: -0.3440, sensor:   F5, stat: variance\n",
      "top:   17, coef:  0.3345, sensor:  AFZ, stat: variance\n",
      "top:   18, coef: -0.2198, sensor:   F6, stat: variance\n",
      "top:   19, coef:  0.1807, sensor:  FC3, stat: variance\n",
      "top:   20, coef: -0.1588, sensor:  CP2, stat: variance\n",
      "top:   21, coef:  0.1449, sensor:   F2, stat: variance\n",
      "top:   22, coef: -0.1402, sensor:  FC1, stat: variance\n",
      "top:   23, coef: -0.1338, sensor:   F3, stat: variance\n",
      "top:   24, coef: -0.1198, sensor:  CP6, stat: variance\n",
      "top:   25, coef:  0.1042, sensor:  AF7, stat: variance\n",
      "top:   26, coef: -0.0919, sensor:  CP3, stat: variance\n",
      "top:   27, coef: -0.0706, sensor:  CPZ, stat: variance\n",
      "top:   28, coef: -0.0326, sensor:  CP5, stat: variance\n",
      "top:   29, coef: -0.0232, sensor:   C5, stat: variance\n",
      "top:   30, coef: -0.0029, sensor:  AF1, stat: variance\n"
     ]
    }
   ],
   "source": [
    "selector = SelectFromModel(estimator=LogisticRegression(max_iter=int(1e5)), threshold=0,\n",
    "                           max_features=np.argmax(f1_hist) + 1).fit(X, y)\n",
    "indices = selector.get_support()\n",
    "indices = np.where(indices)[0]\n",
    "X_n_train, X_n_test = X_train[:, indices], X_test[:, indices]\n",
    "model = LogisticRegression(max_iter=int(1e5)).fit(X_n_train, y_train)\n",
    "\n",
    "print(f\"accuracy: {model.score(X_n_test, y_test)}\")\n",
    "print(f\"f1: {f1_score(y_test, model.predict(X_n_test))}\\n===\")\n",
    "\n",
    "idcs_sorted = np.argsort(np.abs(model.coef_))\n",
    "for i_c, idx in enumerate(idcs_sorted.T[::-1]):\n",
    "    sensor, moment = idx_to_sensor_statistic(int(idx), moments=[\"variance\"])\n",
    "    print(f\"top: {i_c + 1:4d}, coef: {float(model.coef_[0, idx]): 2.4f}, sensor: {sensor:>4}, stat: {moment}\")"
   ],
   "metadata": {
    "collapsed": false,
    "ExecuteTime": {
     "start_time": "2023-04-15T19:04:57.631271Z",
     "end_time": "2023-04-15T19:04:58.453866Z"
    }
   }
  }
 ],
 "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
}