train_classifier.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "inA7sCOffNuc"
   },
   "source": [
    "# Train IDS classifier\n",
    "Notebook for documentation of classifier training outputs (in addition to saved training logs). \n",
    "\n",
    "The classifier is used for evaluating the generated flows by the GAN."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train classifier including BENIGN flows"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Args: Namespace(n_estimators=100, save_dir='./models/classifier', seed=0)\n",
      "Loading dataset...\n",
      "\n",
      "Making Classifier...\n",
      "\n",
      "Starting training...\n",
      "[Parallel(n_jobs=-1)]: Using backend ThreadingBackend with 4 concurrent workers.\n",
      "building tree 1 of 100\n",
      "building tree 2 of 100\n",
      "building tree 3 of 100\n",
      "building tree 4 of 100\n",
      "building tree 5 of 100\n",
      "building tree 6 of 100\n",
      "building tree 7 of 100\n",
      "building tree 8 of 100\n",
      "building tree 9 of 100\n",
      "building tree 10 of 100\n",
      "building tree 11 of 100\n",
      "building tree 12 of 100\n",
      "building tree 13 of 100\n",
      "building tree 14 of 100\n",
      "building tree 15 of 100\n",
      "building tree 16 of 100\n",
      "building tree 17 of 100\n",
      "building tree 18 of 100\n",
      "building tree 19 of 100\n",
      "building tree 20 of 100\n",
      "building tree 21 of 100\n",
      "building tree 22 of 100\n",
      "building tree 23 of 100\n",
      "building tree 24 of 100\n",
      "building tree 25 of 100\n",
      "building tree 26 of 100\n",
      "building tree 27 of 100\n",
      "building tree 28 of 100\n",
      "building tree 29 of 100\n",
      "building tree 30 of 100\n",
      "building tree 31 of 100\n",
      "building tree 32 of 100\n",
      "building tree 33 of 100\n",
      "building tree 34 of 100\n",
      "building tree 35 of 100\n",
      "building tree 36 of 100\n",
      "building tree 37 of 100\n",
      "[Parallel(n_jobs=-1)]: Done  33 tasks      | elapsed:  4.9min\n",
      "building tree 38 of 100\n",
      "building tree 39 of 100\n",
      "building tree 40 of 100\n",
      "building tree 41 of 100\n",
      "building tree 42 of 100\n",
      "building tree 43 of 100\n",
      "building tree 44 of 100\n",
      "building tree 45 of 100\n",
      "building tree 46 of 100\n",
      "building tree 47 of 100\n",
      "building tree 48 of 100\n",
      "building tree 49 of 100\n",
      "building tree 50 of 100\n",
      "building tree 51 of 100\n",
      "building tree 52 of 100\n",
      "building tree 53 of 100\n",
      "building tree 54 of 100\n",
      "building tree 55 of 100\n",
      "building tree 56 of 100\n",
      "building tree 57 of 100\n",
      "building tree 58 of 100\n",
      "building tree 59 of 100\n",
      "building tree 60 of 100\n",
      "building tree 61 of 100\n",
      "building tree 62 of 100\n",
      "building tree 63 of 100\n",
      "building tree 64 of 100\n",
      "building tree 65 of 100\n",
      "building tree 66 of 100\n",
      "building tree 67 of 100\n",
      "building tree 68 of 100\n",
      "building tree 69 of 100\n",
      "building tree 70 of 100\n",
      "building tree 71 of 100\n",
      "building tree 72 of 100\n",
      "building tree 73 of 100\n",
      "building tree 74 of 100\n",
      "building tree 75 of 100\n",
      "building tree 76 of 100\n",
      "building tree 77 of 100\n",
      "building tree 78 of 100\n",
      "building tree 79 of 100\n",
      "building tree 80 of 100\n",
      "building tree 81 of 100\n",
      "building tree 82 of 100\n",
      "building tree 83 of 100\n",
      "building tree 84 of 100\n",
      "building tree 85 of 100\n",
      "building tree 86 of 100\n",
      "building tree 87 of 100\n",
      "building tree 88 of 100\n",
      "building tree 89 of 100\n",
      "building tree 90 of 100\n",
      "building tree 91 of 100\n",
      "building tree 92 of 100\n",
      "building tree 93 of 100\n",
      "building tree 94 of 100\n",
      "building tree 95 of 100\n",
      "building tree 96 of 100\n",
      "building tree 97 of 100\n",
      "building tree 98 of 100\n",
      "building tree 99 of 100\n",
      "building tree 100 of 100\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed: 14.2min finished\n",
      "\n",
      "Saving classifier to models/classifier/23-04-2021_11h56m\n",
      "\n",
      "Evaluating...\n",
      "\n",
      "--------------- Test statistics ---------------\n",
      "[Parallel(n_jobs=4)]: Using backend ThreadingBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=4)]: Done  33 tasks      | elapsed:    0.3s\n",
      "[Parallel(n_jobs=4)]: Done 100 out of 100 | elapsed:    0.9s finished\n",
      "/Users/thomas/anaconda3/lib/python3.7/site-packages/sklearn/metrics/_classification.py:1221: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
      "  _warn_prf(average, modifier, msg_start, len(result))\n",
      "                            precision    recall  f1-score   support\n",
      "\n",
      "                       Bot       0.97      0.98      0.97        98\n",
      "                      DDoS       1.00      1.00      1.00      6401\n",
      "             DoS GoldenEye       1.00      1.00      1.00       515\n",
      "                  DoS Hulk       1.00      1.00      1.00     11506\n",
      "          DoS Slowhttptest       0.99      0.99      0.99       275\n",
      "             DoS slowloris       1.00      0.99      0.99       290\n",
      "               FTP-Patator       1.00      1.00      1.00       397\n",
      "              Infiltration       1.00      1.00      1.00         2\n",
      "                  PortScan       1.00      1.00      1.00      7940\n",
      "               SSH-Patator       1.00      1.00      1.00       295\n",
      "  Web Attack  Brute Force       0.92      0.87      0.89        75\n",
      "Web Attack  Sql Injection       0.00      0.00      0.00         1\n",
      "          Web Attack  XSS       0.71      0.73      0.72        33\n",
      "                   zBENIGN       1.00      1.00      1.00    113566\n",
      "\n",
      "                  accuracy                           1.00    141394\n",
      "                 macro avg       0.90      0.90      0.90    141394\n",
      "              weighted avg       1.00      1.00      1.00    141394\n",
      "\n",
      "/Users/thomas/anaconda3/lib/python3.7/site-packages/matplotlib/backends/backend_agg.py:214: RuntimeWarning: Glyph 150 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "/Users/thomas/anaconda3/lib/python3.7/site-packages/matplotlib/backends/backend_agg.py:183: RuntimeWarning: Glyph 150 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n",
      "Figure(640x480)\n",
      "Figure(640x480)\n",
      "\n",
      "--------------- Feature ranking ---------------\n",
      "                     Feature  Importance score\n",
      "Rank                                          \n",
      "1     Bwd Packet Length Mean            0.0510\n",
      "2          Packet Length Std            0.0501\n",
      "3      Bwd Packet Length Std            0.0501\n",
      "4        Average Packet Size            0.0470\n",
      "5       Avg Bwd Segment Size            0.0434\n",
      "...                      ...               ...\n",
      "75        Bwd Avg Bytes/Bulk            0.0000\n",
      "76         Fwd Avg Bulk Rate            0.0000\n",
      "77      Fwd Avg Packets/Bulk            0.0000\n",
      "78             Bwd URG Flags            0.0000\n",
      "79        Fwd Avg Bytes/Bulk            0.0000\n",
      "\n",
      "[79 rows x 2 columns]\n"
     ]
    }
   ],
   "source": [
    "! python3 train_classifier.py"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train classifier without BENIGN flows"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Args: Namespace(cross_validate=True, data_path='./data/cic-ids-2017_splits_with_benign/seed_0/', n_estimators=100, remove_benign=True, save_dir='./models/classifier', seed=0)\n",
      "Loading dataset...\n",
      "\n",
      "Making Classifier...\n",
      "\n",
      "Cross validating...\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done  42 tasks      | elapsed:   20.4s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:   46.7s finished\n",
      "[Parallel(n_jobs=4)]: Using backend ThreadingBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=4)]: Done  42 tasks      | elapsed:    0.3s\n",
      "[Parallel(n_jobs=4)]: Done 100 out of 100 | elapsed:    0.6s finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done  42 tasks      | elapsed:   18.3s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:   41.0s finished\n",
      "[Parallel(n_jobs=4)]: Using backend ThreadingBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=4)]: Done  42 tasks      | elapsed:    0.2s\n",
      "[Parallel(n_jobs=4)]: Done 100 out of 100 | elapsed:    0.6s finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done  42 tasks      | elapsed:   18.7s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:   41.1s finished\n",
      "[Parallel(n_jobs=4)]: Using backend ThreadingBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=4)]: Done  42 tasks      | elapsed:    0.2s\n",
      "[Parallel(n_jobs=4)]: Done 100 out of 100 | elapsed:    0.5s finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done  42 tasks      | elapsed:   18.5s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:   41.2s finished\n",
      "[Parallel(n_jobs=4)]: Using backend ThreadingBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=4)]: Done  42 tasks      | elapsed:    0.3s\n",
      "[Parallel(n_jobs=4)]: Done 100 out of 100 | elapsed:    0.6s finished\n",
      "[Parallel(n_jobs=-1)]: Using backend LokyBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done  42 tasks      | elapsed:   20.2s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:   42.9s finished\n",
      "[Parallel(n_jobs=4)]: Using backend ThreadingBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=4)]: Done  42 tasks      | elapsed:    0.3s\n",
      "[Parallel(n_jobs=4)]: Done 100 out of 100 | elapsed:    0.6s finished\n",
      "\n",
      "--------------- Cross validation statistics ---------------\n",
      "Mean accuracy: 0.9990505515444827 | Mean balanced-accuracy: 0.9216691285162891 | Mean f1-macro: 0.9350912336337978 | Mean f1-weighted: 0.9990304821813425\n",
      "\n",
      "Starting training...\n",
      "[Parallel(n_jobs=-1)]: Using backend ThreadingBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=-1)]: Done  42 tasks      | elapsed:   21.6s\n",
      "[Parallel(n_jobs=-1)]: Done 100 out of 100 | elapsed:   49.9s finished\n",
      "\n",
      "Saving classifier to models/classifier/20-05-2021_12h01m\n",
      "\n",
      "--------------- Test statistics ---------------\n",
      "[Parallel(n_jobs=4)]: Using backend ThreadingBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=4)]: Done  42 tasks      | elapsed:    0.1s\n",
      "[Parallel(n_jobs=4)]: Done 100 out of 100 | elapsed:    0.1s finished\n",
      "                            precision    recall  f1-score   support\n",
      "\n",
      "                       Bot       1.00      1.00      1.00        98\n",
      "                      DDoS       1.00      1.00      1.00      6401\n",
      "             DoS GoldenEye       1.00      1.00      1.00       515\n",
      "                  DoS Hulk       1.00      1.00      1.00     11506\n",
      "          DoS Slowhttptest       1.00      0.99      1.00       275\n",
      "             DoS slowloris       1.00      0.99      0.99       290\n",
      "               FTP-Patator       1.00      1.00      1.00       397\n",
      "              Infiltration       1.00      1.00      1.00         2\n",
      "                  PortScan       1.00      1.00      1.00      7940\n",
      "               SSH-Patator       1.00      1.00      1.00       295\n",
      "  Web Attack  Brute Force       0.90      0.85      0.88        75\n",
      "Web Attack  Sql Injection       1.00      1.00      1.00         1\n",
      "          Web Attack  XSS       0.70      0.79      0.74        33\n",
      "\n",
      "                  accuracy                           1.00     27828\n",
      "                 macro avg       0.97      0.97      0.97     27828\n",
      "              weighted avg       1.00      1.00      1.00     27828\n",
      "\n",
      "/Users/thomas/anaconda3/lib/python3.7/site-packages/matplotlib/backends/backend_agg.py:214: RuntimeWarning: Glyph 150 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "/Users/thomas/anaconda3/lib/python3.7/site-packages/matplotlib/backends/backend_agg.py:183: RuntimeWarning: Glyph 150 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n",
      "Figure(640x480)\n",
      "Figure(640x480)\n",
      "\n",
      "--------------- Feature ranking ---------------\n",
      "                          Feature  Importance score\n",
      "Rank                                               \n",
      "1           Bwd Packet Length Min            0.0608\n",
      "2     Total Length of Fwd Packets            0.0410\n",
      "3                   Fwd IAT Total            0.0397\n",
      "4                    Fwd IAT Mean            0.0363\n",
      "5          Fwd Packet Length Mean            0.0338\n",
      "...                           ...               ...\n",
      "75           Fwd Avg Packets/Bulk            0.0000\n",
      "76             Fwd Avg Bytes/Bulk            0.0000\n",
      "77                 ECE Flag Count            0.0000\n",
      "78                 CWE Flag Count            0.0000\n",
      "79           Bwd Avg Packets/Bulk            0.0000\n",
      "\n",
      "[79 rows x 2 columns]\n"
     ]
    }
   ],
   "source": [
    "! python3 train_classifier.py --remove_benign --cross_validate"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Feed unscaled testing data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [],
   "source": [
    "import joblib \n",
    "from cic_ids_17_dataset import CIC17Dataset\n",
    "from scikitplot.metrics import plot_confusion_matrix\n",
    "\n",
    "data_path = \"./data/cic-ids-2017_splits/seed_0/\"\n",
    "test_dataset = CIC17Dataset(data_path + \"test_dataset_scaled.pt\", is_scaled=True)\n",
    "classifier = joblib.load(\"./models/classifier/20-05-2021_12h01m/classifier.gz\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "[Parallel(n_jobs=4)]: Using backend ThreadingBackend with 4 concurrent workers.\n",
      "[Parallel(n_jobs=4)]: Done  42 tasks      | elapsed:    0.0s\n",
      "[Parallel(n_jobs=4)]: Done 100 out of 100 | elapsed:    0.1s finished\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7ff76ab08da0>"
      ]
     },
     "execution_count": 118,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/thomas/anaconda3/lib/python3.7/site-packages/matplotlib/backends/backend_agg.py:214: RuntimeWarning: Glyph 150 missing from current font.\n",
      "  font.set_text(s, 0.0, flags=flags)\n",
      "/Users/thomas/anaconda3/lib/python3.7/site-packages/matplotlib/backends/backend_agg.py:183: RuntimeWarning: Glyph 150 missing from current font.\n",
      "  font.set_text(s, 0, flags=flags)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "X = test_dataset.X[:500]\n",
    "y = test_dataset.y[:500]\n",
    "X_unscaled = test_dataset.scaler.inverse_transform(X)\n",
    "y_preds = classifier.predict(X_unscaled)\n",
    "\n",
    "y = test_dataset.label_encoder.inverse_transform(y)\n",
    "y_preds = test_dataset.label_encoder.inverse_transform(y_preds)\n",
    "# i.e., unscaling is not the probelm\n",
    "plot_confusion_matrix(y, y_preds, x_tick_rotation=90)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Extracting most important features per class"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 166,
   "metadata": {},
   "outputs": [],
   "source": [
    "import joblib \n",
    "import torch\n",
    "import json\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "from sklearn.preprocessing import scale\n",
    "\n",
    "from cic_ids_17_dataset import CIC17Dataset\n",
    "from scikitplot.metrics import plot_confusion_matrix\n",
    "\n",
    "data_path = \"./data/cic-ids-2017_splits/seed_0/\"\n",
    "train_dataset = CIC17Dataset(data_path + \"train_dataset.pt\", is_scaled=False)\n",
    "column_names = train_dataset.column_names[:-1]\n",
    "class_names = train_dataset.class_names[:-1]\n",
    "\n",
    "classifier = joblib.load(\"./models/classifier/20-05-2021_12h01m/classifier.gz\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 167,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>importances</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>Bwd Packet Length Min</th>\n",
       "      <td>0.060767</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Total Length of Fwd Packets</th>\n",
       "      <td>0.041007</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fwd IAT Total</th>\n",
       "      <td>0.039727</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fwd IAT Mean</th>\n",
       "      <td>0.036309</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fwd Packet Length Mean</th>\n",
       "      <td>0.033846</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Init_Win_bytes_forward</th>\n",
       "      <td>0.033296</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Total Fwd Packets</th>\n",
       "      <td>0.032648</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Fwd Packets/s</th>\n",
       "      <td>0.032176</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Subflow Fwd Packets</th>\n",
       "      <td>0.032028</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Packet Length Mean</th>\n",
       "      <td>0.031072</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                             importances\n",
       "Bwd Packet Length Min           0.060767\n",
       "Total Length of Fwd Packets     0.041007\n",
       "Fwd IAT Total                   0.039727\n",
       "Fwd IAT Mean                    0.036309\n",
       "Fwd Packet Length Mean          0.033846\n",
       "Init_Win_bytes_forward          0.033296\n",
       "Total Fwd Packets               0.032648\n",
       "Fwd Packets/s                   0.032176\n",
       "Subflow Fwd Packets             0.032028\n",
       "Packet Length Mean              0.031072"
      ]
     },
     "execution_count": 167,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pd.DataFrame({\"importances\": classifier.feature_importances_}, index=column_names).sort_values('importances', ascending=False)[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 168,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/thomas/anaconda3/lib/python3.7/site-packages/sklearn/preprocessing/_data.py:174: UserWarning: Numerical issues were encountered when centering the data and might not be solved. Dataset may contain too large values. You may need to prescale your features.\n",
      "  warnings.warn(\"Numerical issues were encountered \"\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{'Bot': {'Fwd Packet Length Mean': 0.04505365788633558,\n",
       "  'Avg Fwd Segment Size': 0.03441541524103997,\n",
       "  'Fwd IAT Total': -0.027022414843465428,\n",
       "  'Packet Length Mean': -0.023437493144880913},\n",
       " 'DoS GoldenEye': {'Init_Win_bytes_forward': 0.03886307632021916,\n",
       "  'Fwd IAT Min': 0.03709092998057118,\n",
       "  'Fwd IAT Mean': 0.037030882618887664,\n",
       "  'Flow IAT Mean': 0.023107578776238456},\n",
       " 'FTP-Patator': {'Packet Length Mean': -0.025996938895438013,\n",
       "  'Fwd IAT Total': -0.02451583468507913,\n",
       "  'Init_Win_bytes_forward': 0.021013389300412435,\n",
       "  'Fwd IAT Mean': -0.01878393258850729},\n",
       " 'DoS Slowhttptest': {'Fwd Packet Length Mean': 0.0682874494006301,\n",
       "  'Avg Fwd Segment Size': 0.05216315471661115,\n",
       "  'Flow Bytes/s': 0.04770643045640853,\n",
       "  'Init_Win_bytes_forward': 0.042840058084072354},\n",
       " 'DDoS': {'Packet Length Mean': 0.019187094556667107,\n",
       "  'Average Packet Size': 0.012076205208127692,\n",
       "  'Fwd IAT Total': -0.011854575667924862,\n",
       "  'Fwd Packet Length Std': -0.011165366516365087},\n",
       " 'SSH-Patator': {'Packet Length Mean': -0.023844559778381774,\n",
       "  'Fwd IAT Total': -0.022513678962132546,\n",
       "  'Init_Win_bytes_forward': 0.02145915467036674,\n",
       "  'Fwd IAT Mean': -0.01923075859434996},\n",
       " 'Web Attack \\x96 Sql Injection': {'Fwd Packet Length Std': 0.027661145384327505,\n",
       "  'Fwd IAT Total': -0.026637443077039205,\n",
       "  'Init_Win_bytes_forward': 0.02504975078444427,\n",
       "  'Fwd Packet Length Mean': 0.02264910730627846},\n",
       " 'DoS slowloris': {'Fwd IAT Mean': 0.07013045590787957,\n",
       "  'Fwd IAT Min': 0.05389723234614378,\n",
       "  'Init_Win_bytes_forward': 0.034874037388539884,\n",
       "  'Fwd IAT Total': 0.025755929323887563},\n",
       " 'Web Attack \\x96 XSS': {'Init_Win_bytes_backward': 0.3094235665331582,\n",
       "  'Init_Win_bytes_forward': 0.05965415450713518,\n",
       "  'Packet Length Mean': -0.025070141702924244,\n",
       "  'Fwd IAT Total': -0.02100628430089765},\n",
       " 'Infiltration': {'Total Length of Fwd Packets': 2.2279346316775452,\n",
       "  'Subflow Fwd Bytes': 1.6063265909173443,\n",
       "  'act_data_pkt_fwd': 1.5008217777160273,\n",
       "  'Total Fwd Packets': 1.4112561905137595},\n",
       " 'PortScan': {'Fwd IAT Total': -0.027275396824900426,\n",
       "  'Packet Length Mean': -0.02640313375049776,\n",
       "  'Init_Win_bytes_forward': 0.022500113172157075,\n",
       "  'PSH Flag Count': 0.021892531346280587},\n",
       " 'Heartbleed': {'Subflow Bwd Bytes': 5.387259782343267,\n",
       "  'Fwd Header Length.1': 5.048018145737692,\n",
       "  'Total Length of Bwd Packets': 5.021839742100469,\n",
       "  'Total Fwd Packets': 4.50712012195582},\n",
       " 'DoS Hulk': {'Fwd IAT Total': 0.026371332481425612,\n",
       "  'Flow IAT Max': 0.01798619104786563,\n",
       "  'Fwd IAT Mean': 0.016901855413630398,\n",
       "  'PSH Flag Count': -0.01624351643851001},\n",
       " 'Web Attack \\x96 Brute Force': {'Init_Win_bytes_backward': 0.2677787332508583,\n",
       "  'Init_Win_bytes_forward': 0.054594917838036695,\n",
       "  'Packet Length Mean': -0.02502421168191023,\n",
       "  'Fwd IAT Total': -0.02142380896475967}}"
      ]
     },
     "execution_count": 168,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# sklearn's RF does not expose them per class, only overall\n",
    "# extract feature importances per class, according to:\n",
    "# https://stackoverflow.com/questions/50201913/using-scikit-learn-to-determine-feature-importances-per-class-in-a-rf-model\n",
    "\n",
    "def class_feature_importance(X, Y, feature_importances):\n",
    "    N, M = X.shape\n",
    "    X = scale(X)\n",
    "\n",
    "    out = {}\n",
    "    for c in set(Y):\n",
    "        out[c] = dict(\n",
    "            zip(range(M), np.mean(X[Y==c, :], axis=0) * feature_importances)\n",
    "        )\n",
    "\n",
    "    return out\n",
    "\n",
    "def get_top_n_features_per_class(df, top_n=4):\n",
    "    top_n_features = dict()\n",
    "    df_dict = df.to_dict()\n",
    "    \n",
    "    for label in df.columns:\n",
    "        features = df[label].abs().nlargest(top_n).keys()\n",
    "        top_n_features[label] = {feature: df_dict[label][feature] for feature in features}\n",
    "    return top_n_features\n",
    "\n",
    "\n",
    "\n",
    "importances = classifier.feature_importances_\n",
    "std = np.std([tree.feature_importances_ for tree in classifier.estimators_], axis=0)\n",
    "indices = np.argsort(importances)[::-1]\n",
    "feature_list = [column_names[indices[f]] for f in range(len(column_names))]\n",
    "ff = np.array(feature_list)\n",
    "\n",
    "labels_transformed = train_dataset.label_encoder.inverse_transform(train_dataset.y)\n",
    "result = class_feature_importance(train_dataset.X, labels_transformed, importances)\n",
    "df_result = pd.DataFrame(result)\n",
    "df_result.index = column_names\n",
    "\n",
    "top_n_features = get_top_n_features_per_class(df_result)\n",
    "top_n_features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 172,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "({'Average Packet Size',\n",
       "  'Avg Fwd Segment Size',\n",
       "  'Flow Bytes/s',\n",
       "  'Flow IAT Max',\n",
       "  'Flow IAT Mean',\n",
       "  'Fwd Header Length.1',\n",
       "  'Fwd IAT Mean',\n",
       "  'Fwd IAT Min',\n",
       "  'Fwd IAT Total',\n",
       "  'Fwd Packet Length Mean',\n",
       "  'Fwd Packet Length Std',\n",
       "  'Init_Win_bytes_backward',\n",
       "  'Init_Win_bytes_forward',\n",
       "  'PSH Flag Count',\n",
       "  'Packet Length Mean',\n",
       "  'Subflow Bwd Bytes',\n",
       "  'Subflow Fwd Bytes',\n",
       "  'Total Fwd Packets',\n",
       "  'Total Length of Bwd Packets',\n",
       "  'Total Length of Fwd Packets',\n",
       "  'act_data_pkt_fwd'},\n",
       " 21)"
      ]
     },
     "execution_count": 172,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "unique_feature_cols = {k for top_n in top_n_features.values() for k in top_n.keys()}\n",
    "unique_feature_cols, len(unique_feature_cols)"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "name": "train_classifier.ipynb",
   "provenance": [],
   "toc_visible": true
  },
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}