Generating_adversarial_samples.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "## TASK2 GENERATING ADVERSARIAL SAMPLES"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.decomposition import PCA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "DONE\n"
     ]
    }
   ],
   "source": [
    "headers = ['timestamp','duration','protocol', 'src_ip', 'src_port', 'direction', 'dst_ip' , 'dst_port'\n",
    "           ,'state', 'src_type', 'dst_type' , 'total_packets', 'bytes_bothdir', 'bytes_onedir', 'label']\n",
    "\n",
    "df_train_original = pd.read_csv(\"training_data_with_labels.csv\", names=headers, header=None).sample(frac=0.05, replace=False, random_state=42).fillna(-1)\n",
    "#df_val_notsampled = pd.read_csv(\"A2_2/valid_data_with_labels.csv\", names=val_headers, header=None).fillna(-1)\n",
    "df_test_original = pd.read_csv(\"test_data_with_labels.csv\", names=headers, header=None).fillna(-1)\n",
    "\n",
    "print('DONE')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "98908\n"
     ]
    }
   ],
   "source": [
    "print(len(df_train_original))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_headers = ['timestamp','duration','protocol', 'src_ip', 'src_port', 'direction', 'dst_ip' , 'dst_port'\n",
    "           ,'state', 'src_type', 'dst_type' , 'total_packets', 'bytes_bothdir', 'bytes_onedir']\n",
    "\n",
    "df_train_original['label']= df_train_original['label'].apply(str)\n",
    "df_train_original.loc[df_train_original['label'].str.contains(\"From-Botnet\"), 'label'] = -1\n",
    "df_train_original.loc[df_train_original['label'].str.contains(\"From-Botnet\")== False, 'label'] = 1\n",
    "df_train_original.label = df_train_original.label.astype(int)\n",
    "\n",
    "df_test_original['label']= df_test_original['label'].apply(str)\n",
    "df_test_original.loc[df_test_original['label'].str.contains(\"From-Botnet\"), 'label'] = -1\n",
    "df_test_original.loc[df_test_original['label'].str.contains(\"From-Botnet\")== False, 'label'] = 1\n",
    "df_test_original.label = df_test_original.label.astype(int)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "98908\n",
      "574383\n"
     ]
    }
   ],
   "source": [
    "#X_train = df_train_sampled\n",
    "X_train_ori = df_train_original\n",
    "X_test_ori = df_test_original\n",
    "print(len(X_train_ori))\n",
    "print(len(X_test_ori))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Preprocessing for calculation, and preparing for onehot \n",
    "def convert_to_str(df):\n",
    "    convert_to_str_list = ['src_port', 'dst_port']\n",
    "    for column in convert_to_str_list: \n",
    "        df[column]= df[column].apply(str)\n",
    "\n",
    "    df_obj = df.select_dtypes(['object'])\n",
    "    df[df_obj.columns] = df_obj.apply(lambda x: x.str.strip())\n",
    "    \n",
    "def apply_stat(df):\n",
    "    df['pps'] = df['total_packets']/ df['duration']\n",
    "    df['bps_onedir'] = df['bytes_onedir']/ df['total_packets']\n",
    "    df['bpp_onedir'] = df['bytes_onedir']/ df['duration']\n",
    "    df['bps_bothdir'] = df['bytes_bothdir']/ df['total_packets']\n",
    "    df['bpp_bothdir'] = df['bytes_bothdir']/ df['duration']  \n",
    "    \n",
    "def preprocess(flag, df_original):\n",
    "    groupby_columns = ['src_ip', 'dst_ip', 'src_port', 'dst_port', 'protocol']\n",
    "    df_original[\"stream_id\"] = df_original.groupby(groupby_columns).grouper.group_info[0]\n",
    "    df_original = df_original.sort_values(by=['stream_id','timestamp'])\n",
    "    df_original['timestamp']=df_original['timestamp'].apply(pd.to_datetime)\n",
    "    #df['endtime'] = df['timestamp'] + pd.to_timedelta(df['duration'], unit='s')\n",
    "    df_original.loc[df_original['duration']==0, 'duration'] = 0.000001\n",
    "    #apply_addinfo(df)    \n",
    "    convert_to_str(df_original)    \n",
    "    stream_id_grouped = df_original.groupby(['stream_id'])    \n",
    "    stream_id = stream_id_grouped.grouper.group_info[0]\n",
    "    stream_id = set(stream_id)\n",
    "    newdf = pd.DataFrame(stream_id, columns=['stream_id'])\n",
    "    newdf['src_ip'] = stream_id_grouped['src_ip'].first()\n",
    "    newdf['dst_ip'] = stream_id_grouped['dst_ip'].first()\n",
    "    newdf['src_port'] = stream_id_grouped['src_port'].first()\n",
    "    newdf['dst_port'] = stream_id_grouped['dst_port'].first()\n",
    "    newdf['protocol'] = stream_id_grouped['protocol'].first()\n",
    "    \n",
    "    if flag == 'val':\n",
    "        newdf['label'] = stream_id_grouped['label'].mean()\n",
    "\n",
    "    return newdf, df_original"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Wall time: 0 ns\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "## FOR FEATURE3: average cumulative features\n",
    "\n",
    "src_time_diff_dict = {}\n",
    "src_cnt_dict = {}\n",
    "dst_time_diff_dict = {}\n",
    "dst_cnt_dict = {}\n",
    "srcdst_time_diff_dict = {}\n",
    "srcdst_cnt_dict = {}\n",
    "\n",
    "def get_time_diff(df_original):\n",
    "\n",
    "    df_original['src_time_diff']= df_original.groupby('src_ip')['timestamp'].apply(lambda s: s.diff().dt.seconds < 60)\n",
    "    df_original['dst_time_diff']= df_original.groupby('dst_ip')['timestamp'].apply(lambda s: s.diff().dt.seconds < 60)\n",
    "    df_original['srcdst_time_diff']= df_original.groupby(['src_ip', 'dst_port'])['timestamp'].apply(lambda s: s.diff().dt.seconds < 30) \n",
    "    \n",
    "def groupby_sum(df, df_original):\n",
    "    fields = ['bytes_bothdir','bytes_onedir', 'total_packets', 'duration']\n",
    "\n",
    "    for field in fields:\n",
    "        df[field] = df_original.groupby(['stream_id'])[field].sum()\n",
    "    \n",
    "    return df\n",
    "\n",
    "def apply_avg(df, df_original):\n",
    "    df['pps'] = df['total_packets']/df['duration']\n",
    "    df['bps_onedir'] = df['bytes_onedir']/ df['total_packets']\n",
    "    df['bpp_onedir'] = df['bytes_onedir']/ df['duration']\n",
    "    df['bps_bothdir'] = df['bytes_bothdir']/ df['total_packets']\n",
    "    df['bpp_bothdir'] = df['bytes_bothdir']/ df['duration']    \n",
    "    return df\n",
    "\n",
    "def cnt_timed_src(df, df_original):  \n",
    "    src_time_diff_dict = df_original.groupby('src_ip')['src_time_diff'].sum().to_dict()\n",
    "    #dst_cnt_dict = df_original.groupby('src_ip')['dst_ip'].count().to_dict()\n",
    "\n",
    "    df['cnt_timed_src']=df.apply(lambda x: src_time_diff_dict[x.src_ip], axis=1)\n",
    "    #df['cnt_dst']=df.apply(lambda x: dst_cnt_dict[x.src_ip], axis=1)\n",
    "    return df\n",
    "\n",
    "def cnt_timed_dst(df, df_original ):\n",
    "    dst_time_diff_dict = df_original.groupby('dst_ip')['dst_time_diff'].sum().to_dict()\n",
    "    #src_cnt_dict = df_original.groupby('dst_ip')['src_ip'].count().to_dict()\n",
    "    df['cnt_timed_dst']=df.apply(lambda x: dst_time_diff_dict[x.dst_ip], axis=1)\n",
    "    #df['cnt_src']=df.apply(lambda x: src_cnt_dict[x.dst_ip], axis=1)\n",
    "    return df\n",
    "\n",
    "def get_dict(dictionary):\n",
    "    newdict={}\n",
    "    for (k1, k2), count in dictionary.items():\n",
    "        if k1 in newdict:\n",
    "            if count>0:\n",
    "                newdict[k1] +=1\n",
    "        else:\n",
    "            newdict[k1] = 0\n",
    "    return newdict\n",
    "    \n",
    "def cnt_timed_srcdst(df, df_original):    \n",
    "    srcdst_time_diff_dict = df_original.groupby(['src_ip', 'dst_port' ])['srcdst_time_diff'].sum().to_dict() \n",
    "    srcdst_time_diff_dict = get_dict(srcdst_time_diff_dict)\n",
    "    #srcdst_cnt_dict = df_original.groupby(['src_ip', 'dst_port'])['src_ip'].count().to_dict()\n",
    "    #srcdst_cnt_dict = get_dict(srcdst_cnt_dict)\n",
    "    df['cnt_timed_srcdst']=df.apply(lambda x: srcdst_time_diff_dict[x.src_ip], axis=1)\n",
    "    #df['cnt_srcdst']=df.apply(lambda x: srcdst_cnt_dict[x.src_ip], axis=1)\n",
    "    return df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "scaler = StandardScaler()\n",
    "\n",
    "def apply_scaler(df):\n",
    "    df_columns_to_scale = df[columns_to_scale]\n",
    "    train_scaled_features  = scaler.fit_transform(df_columns_to_scale) \n",
    "    train_scaled_features_df = pd.DataFrame( train_scaled_features, index=df_columns_to_scale.index, columns=columns_to_scale)\n",
    "\n",
    "    return train_scaled_features_df\n",
    "\n",
    "def get_pc(X, num_components=2, tsne=False):\n",
    "\n",
    "    if tsne:\n",
    "        reduced_data = TSNE(n_components=num_components).fit_transform(X)\n",
    "    else:\n",
    "        pca = PCA(n_components=num_components, random_state=42).fit(X)\n",
    "        print(\"PCA.explained variance_ratio_ :\",sum(pca.explained_variance_ratio_) )\n",
    "        reduced_data = pca.transform(X)\n",
    "    \n",
    "    return reduced_data\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Wall time: 0 ns\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "## cumulative featrues mostly by grouped by [src_ip, dst_ip, src_port, dst_port, protocol]\n",
    "def get_feature3(flag, df_original):           \n",
    "\n",
    "    df, df_original = preprocess(flag, df_original)\n",
    "    df[['starttime', 'endtime']] = df_original.groupby(['stream_id']).agg(starttime=('timestamp', 'min'), endtime=('timestamp', 'max'))    \n",
    "    df = groupby_sum(df, df_original)    \n",
    "    df = apply_avg(df, df_original)  \n",
    "    get_time_diff(df_original)\n",
    "    df = cnt_timed_src(df, df_original)\n",
    "    df = cnt_timed_dst(df, df_original)\n",
    "    df = cnt_timed_srcdst(df, df_original)   \n",
    "    print(flag , 'DONE')\n",
    "    \n",
    "    return df\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "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>bytes_bothdir</th>\n",
       "      <th>bytes_onedir</th>\n",
       "      <th>total_packets</th>\n",
       "      <th>duration</th>\n",
       "      <th>pps</th>\n",
       "      <th>bps_onedir</th>\n",
       "      <th>bpp_onedir</th>\n",
       "      <th>bps_bothdir</th>\n",
       "      <th>bpp_bothdir</th>\n",
       "      <th>cnt_timed_src</th>\n",
       "      <th>cnt_timed_dst</th>\n",
       "      <th>cnt_timed_srcdst</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>-0.03916</td>\n",
       "      <td>-0.026629</td>\n",
       "      <td>-0.041634</td>\n",
       "      <td>-0.211271</td>\n",
       "      <td>0.657007</td>\n",
       "      <td>-0.424921</td>\n",
       "      <td>0.381812</td>\n",
       "      <td>-0.636037</td>\n",
       "      <td>1.063197</td>\n",
       "      <td>-0.880234</td>\n",
       "      <td>-1.004642</td>\n",
       "      <td>-1.677192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>-0.03916</td>\n",
       "      <td>-0.026629</td>\n",
       "      <td>-0.041634</td>\n",
       "      <td>-0.211271</td>\n",
       "      <td>-0.275575</td>\n",
       "      <td>-0.424921</td>\n",
       "      <td>-0.297139</td>\n",
       "      <td>-0.636037</td>\n",
       "      <td>-0.301503</td>\n",
       "      <td>-0.880234</td>\n",
       "      <td>-1.004642</td>\n",
       "      <td>-1.677192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>-0.03916</td>\n",
       "      <td>-0.026629</td>\n",
       "      <td>-0.041634</td>\n",
       "      <td>-0.211271</td>\n",
       "      <td>-0.264796</td>\n",
       "      <td>-0.424921</td>\n",
       "      <td>-0.289291</td>\n",
       "      <td>-0.636037</td>\n",
       "      <td>-0.285729</td>\n",
       "      <td>-0.880234</td>\n",
       "      <td>-1.004642</td>\n",
       "      <td>-1.677192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>-0.03916</td>\n",
       "      <td>-0.026629</td>\n",
       "      <td>-0.041634</td>\n",
       "      <td>-0.211271</td>\n",
       "      <td>-0.269122</td>\n",
       "      <td>-0.424921</td>\n",
       "      <td>-0.292441</td>\n",
       "      <td>-0.636037</td>\n",
       "      <td>-0.292060</td>\n",
       "      <td>-0.880234</td>\n",
       "      <td>-1.004642</td>\n",
       "      <td>-1.677192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>-0.03916</td>\n",
       "      <td>-0.026629</td>\n",
       "      <td>-0.041634</td>\n",
       "      <td>-0.211271</td>\n",
       "      <td>-0.247980</td>\n",
       "      <td>-0.424921</td>\n",
       "      <td>-0.277048</td>\n",
       "      <td>-0.636037</td>\n",
       "      <td>-0.261121</td>\n",
       "      <td>-0.880234</td>\n",
       "      <td>-1.004642</td>\n",
       "      <td>-1.677192</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   bytes_bothdir  bytes_onedir  total_packets  duration       pps  bps_onedir  \\\n",
       "0       -0.03916     -0.026629      -0.041634 -0.211271  0.657007   -0.424921   \n",
       "1       -0.03916     -0.026629      -0.041634 -0.211271 -0.275575   -0.424921   \n",
       "2       -0.03916     -0.026629      -0.041634 -0.211271 -0.264796   -0.424921   \n",
       "3       -0.03916     -0.026629      -0.041634 -0.211271 -0.269122   -0.424921   \n",
       "4       -0.03916     -0.026629      -0.041634 -0.211271 -0.247980   -0.424921   \n",
       "\n",
       "   bpp_onedir  bps_bothdir  bpp_bothdir  cnt_timed_src  cnt_timed_dst  \\\n",
       "0    0.381812    -0.636037     1.063197      -0.880234      -1.004642   \n",
       "1   -0.297139    -0.636037    -0.301503      -0.880234      -1.004642   \n",
       "2   -0.289291    -0.636037    -0.285729      -0.880234      -1.004642   \n",
       "3   -0.292441    -0.636037    -0.292060      -0.880234      -1.004642   \n",
       "4   -0.277048    -0.636037    -0.261121      -0.880234      -1.004642   \n",
       "\n",
       "   cnt_timed_srcdst  \n",
       "0         -1.677192  \n",
       "1         -1.677192  \n",
       "2         -1.677192  \n",
       "3         -1.677192  \n",
       "4         -1.677192  "
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generating features... \n",
      "val DONE\n",
      "train len:  70584\n",
      "DONE\n",
      "Wall time: 29 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "print('Generating features... ')\n",
    "df_train_f3 = get_feature3('val', X_train_ori) \n",
    "print('train len: ', len(df_train_f3))\n",
    "Y = X_train_ori['label']\n",
    "print('DONE')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "DONE\n"
     ]
    }
   ],
   "source": [
    "feature3 = ['bytes_bothdir', 'bytes_onedir', 'total_packets', 'duration', \n",
    "             'pps', 'bps_onedir', 'bpp_onedir', 'bps_bothdir', 'bpp_bothdir', \n",
    "             'cnt_timed_src', 'cnt_timed_dst', 'cnt_timed_srcdst']\n",
    "\n",
    "selected_features = feature3\n",
    "columns_to_scale = feature3\n",
    "\n",
    "df_train_f3[columns_to_scale] = apply_scaler(df_train_f3)\n",
    "X_train = df_train_f3[selected_features]\n",
    "#X = get_pc(X_train, 3)\n",
    "X = X_train\n",
    "Y = df_train_f3['label']\n",
    "\n",
    "print('DONE')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0        1\n",
       "1        1\n",
       "2        1\n",
       "3        1\n",
       "4        1\n",
       "        ..\n",
       "70579    1\n",
       "70580    1\n",
       "70581    1\n",
       "70582    1\n",
       "70583    1\n",
       "Name: label, Length: 70584, dtype: int32"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "class logistic_regression(object):\n",
    "    def fit(self, X, Y, learning_rate=0.0000003, epoch=1000):\n",
    "        X = np.array(X, dtype=\"float32\")\n",
    "        Y = np.array(Y, dtype=\"float32\")\n",
    "\n",
    "        N, D = X.shape\n",
    "        Y = Y.reshape(N, 1)\n",
    "\n",
    "        dlt = np.zeros([N, 1], dtype=\"float32\")\n",
    "        dW = np.zeros([1, D], dtype=\"float32\")\n",
    "        db = 0\n",
    "        self.weight = np.zeros([D, 1], dtype=\"float32\")\n",
    "        self.beta = 0\n",
    "\n",
    "        c = []\n",
    "\n",
    "        for n in range(epoch):\n",
    "            dlt = self.predict(X).T - Y\n",
    "            dW = np.matmul(dlt.T, X).T\n",
    "            db = dlt.sum() / N\n",
    "\n",
    "            self.weight -= learning_rate * dW\n",
    "            self.beta -= learning_rate * db\n",
    "            '''\n",
    "            if n % 200 == 0:\n",
    "                c_new = cost(self.predict(X).T, Y)\n",
    "                c.append(c_new)\n",
    "                err = error_rate(self.predict_class(X).T, Y)\n",
    "                print(\"epoch:\", n, \"cost:\", c_new, \"error rate:\", err)\n",
    "            '''\n",
    "    def predict(self, X):\n",
    "        z = np.matmul(self.weight.T, X.T) + self.beta\n",
    "        return 1 / (1 + np.exp(-z))\n",
    "\n",
    "    def predict_class(self, X):\n",
    "        predictclass = self.predict(X)\n",
    "        return (predictclass >= 0.5) * 1\n",
    "                \n",
    "    def predict(self, X):\n",
    "        z = np.matmul(self.weight.T, X.T) + self.beta\n",
    "        return 1 / (1 + np.exp(-z))\n",
    "\n",
    "    def predict_class(self, X):\n",
    "        predictclass = self.predict(X)\n",
    "        return (predictclass >= 0.5) * 1\n",
    "                \n",
    "def error_rate(P, Y):\n",
    "    return np.mean(Y != P)\n",
    "\n",
    "def adversial_samples(X, Y, model, epsilon=0.00001):\n",
    "    dlt = model.predict(X).T - Y.reshape(X.shape[0], 1)\n",
    "    direction = np.sign(np.matmul(dlt, model.weight.T))\n",
    "    return X + epsilon * direction, Y\n",
    "\n",
    "def cost(Y_hat, Y):\n",
    "    Y_hat = Y_hat.flatten()\n",
    "    Y = Y.flatten()\n",
    "    cost1 = 0\n",
    "    elp = 0.0000000000000000000000000000000000000000000001\n",
    "    for i in range(len(Y)):\n",
    "        cost1 -= Y[i] * np.log(Y_hat[i] + elp) + (1 - Y[i]) * np.log(1 - Y_hat[i] + elp)\n",
    "    return cost1\n",
    "\n",
    "\n",
    "model = logistic_regression()\n",
    "#model = LogisticRegression()\n",
    "#model.fit(X, Y)\n",
    "X = np.array(X)\n",
    "Y = np.array(Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[-0.03916046, -0.02662853, -0.04163367, ..., -0.88023427,\n",
       "        -1.00464185, -1.67719155],\n",
       "       [-0.03916046, -0.02662853, -0.04163367, ..., -0.88023427,\n",
       "        -1.00464185, -1.67719155],\n",
       "       [-0.03916046, -0.02662853, -0.04163367, ..., -0.88023427,\n",
       "        -1.00464185, -1.67719155],\n",
       "       ...,\n",
       "       [-0.03500942, -0.02413222, -0.02837064, ..., -0.93701282,\n",
       "        -1.00364406, -1.67719155],\n",
       "       [-0.03483064, -0.01978513, -0.02173912, ..., -0.93701282,\n",
       "        -1.00364406, -1.67719155],\n",
       "       [-0.02660855, -0.00600252,  0.00611324, ..., -0.93701282,\n",
       "        -1.00464185, -1.67719155]])"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.array(X)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [],
   "source": [
    "def show_adversarials(epoch_num, X, Y):\n",
    "    print('Epoch: ', epoch_num)\n",
    "    model.fit(X, Y, learning_rate=0.0005,  epoch=epoch_num) #20000\n",
    "    print('Error Rate without adversarial examples:', error_rate(model.predict_class(X).T, Y.reshape(X.shape[0], 1)))\n",
    "    print()\n",
    "    \n",
    "    Xadv, Y = adversial_samples(X, Y, model, epsilon=0.2)\n",
    "    print('Error Rate with adversarial examples, epsilon = 0.2:', error_rate(model.predict(Xadv).T, Y.reshape(Xadv.shape[0], 1)))\n",
    "    Xadv2, Y = adversial_samples(X, Y, model, epsilon=0.5)\n",
    "    print('Error Rate with adversarial examples, epsilon = 0.5:', error_rate(model.predict(Xadv2).T, Y.reshape(Xadv2.shape[0], 1)))\n",
    "\n",
    "    plt.scatter(X[:9999, 0], X[:9999, 1], color='blue', alpha=0.4, label='Original Data')\n",
    "    plt.scatter(Xadv[:9999, 0], Xadv[:9999, 1], color='green', alpha=0.4, label='Adversarial Examples, Epsilon = 0.2')\n",
    "    plt.scatter(Xadv2[:9999, 0], Xadv2[:9999, 1], color='purple', alpha=0.4, label='Adversarial Examples, Epsilon = 0.5')\n",
    "    plt.scatter(X[10000:, 0], X[10000:, 1], color='red', alpha=0.4, label='Original Data')\n",
    "    plt.scatter(Xadv[10000:, 0], Xadv[10000:, 1], color='orange', alpha=0.4, label='Adversarial Examples, Epsilon = 0.2')\n",
    "    plt.scatter(Xadv2[10000:, 0], Xadv2[10000:, 1], color='yellow', alpha=0.4, label='Adversarial Examples, Epsilon = 0.5')\n",
    "    plt.legend(loc='best')\n",
    "    plt.show()\n",
    "    return Xadv, Xadv2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch:  1\n",
      "Error Rate without adversarial examples: 0.43446106766405984\n",
      "\n",
      "Error Rate with adversarial examples, epsilon = 0.2: 0.9842598889266689\n",
      "Error Rate with adversarial examples, epsilon = 0.5: 0.9842598889266689\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "Xadv, Xadv2 = show_adversarials(1, X, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[-0.23916046, -0.22662853, -0.24163367, ..., -1.08023427,\n",
       "        -1.20464185, -1.47719155],\n",
       "       [-0.23916046, -0.22662853, -0.24163367, ..., -1.08023427,\n",
       "        -1.20464185, -1.47719155],\n",
       "       [-0.23916046, -0.22662853, -0.24163367, ..., -1.08023427,\n",
       "        -1.20464185, -1.47719155],\n",
       "       ...,\n",
       "       [-0.23500942, -0.22413222, -0.22837064, ..., -1.13701282,\n",
       "        -1.20364406, -1.47719155],\n",
       "       [-0.23483064, -0.21978513, -0.22173912, ..., -1.13701282,\n",
       "        -1.20364406, -1.47719155],\n",
       "       [-0.22660855, -0.20600252, -0.19388676, ..., -1.13701282,\n",
       "        -1.20464185, -1.47719155]])"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "Xadv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "adv_df = pd.DataFrame(data=Xadv, columns=selected_features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "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>bytes_bothdir</th>\n",
       "      <th>bytes_onedir</th>\n",
       "      <th>total_packets</th>\n",
       "      <th>duration</th>\n",
       "      <th>pps</th>\n",
       "      <th>bps_onedir</th>\n",
       "      <th>bpp_onedir</th>\n",
       "      <th>bps_bothdir</th>\n",
       "      <th>bpp_bothdir</th>\n",
       "      <th>cnt_timed_src</th>\n",
       "      <th>cnt_timed_dst</th>\n",
       "      <th>cnt_timed_srcdst</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>-0.239160</td>\n",
       "      <td>-0.226629</td>\n",
       "      <td>-0.241634</td>\n",
       "      <td>-0.411271</td>\n",
       "      <td>0.857007</td>\n",
       "      <td>-0.624921</td>\n",
       "      <td>0.581812</td>\n",
       "      <td>-0.836037</td>\n",
       "      <td>1.263197</td>\n",
       "      <td>-1.080234</td>\n",
       "      <td>-1.204642</td>\n",
       "      <td>-1.477192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>-0.239160</td>\n",
       "      <td>-0.226629</td>\n",
       "      <td>-0.241634</td>\n",
       "      <td>-0.411271</td>\n",
       "      <td>-0.075575</td>\n",
       "      <td>-0.624921</td>\n",
       "      <td>-0.097139</td>\n",
       "      <td>-0.836037</td>\n",
       "      <td>-0.101503</td>\n",
       "      <td>-1.080234</td>\n",
       "      <td>-1.204642</td>\n",
       "      <td>-1.477192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>-0.239160</td>\n",
       "      <td>-0.226629</td>\n",
       "      <td>-0.241634</td>\n",
       "      <td>-0.411271</td>\n",
       "      <td>-0.064796</td>\n",
       "      <td>-0.624921</td>\n",
       "      <td>-0.089291</td>\n",
       "      <td>-0.836037</td>\n",
       "      <td>-0.085729</td>\n",
       "      <td>-1.080234</td>\n",
       "      <td>-1.204642</td>\n",
       "      <td>-1.477192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>-0.239160</td>\n",
       "      <td>-0.226629</td>\n",
       "      <td>-0.241634</td>\n",
       "      <td>-0.411271</td>\n",
       "      <td>-0.069122</td>\n",
       "      <td>-0.624921</td>\n",
       "      <td>-0.092441</td>\n",
       "      <td>-0.836037</td>\n",
       "      <td>-0.092060</td>\n",
       "      <td>-1.080234</td>\n",
       "      <td>-1.204642</td>\n",
       "      <td>-1.477192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>-0.239160</td>\n",
       "      <td>-0.226629</td>\n",
       "      <td>-0.241634</td>\n",
       "      <td>-0.411271</td>\n",
       "      <td>-0.047980</td>\n",
       "      <td>-0.624921</td>\n",
       "      <td>-0.077048</td>\n",
       "      <td>-0.836037</td>\n",
       "      <td>-0.061121</td>\n",
       "      <td>-1.080234</td>\n",
       "      <td>-1.204642</td>\n",
       "      <td>-1.477192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>70579</th>\n",
       "      <td>-0.182166</td>\n",
       "      <td>-0.065463</td>\n",
       "      <td>-0.148792</td>\n",
       "      <td>-0.410621</td>\n",
       "      <td>-0.093707</td>\n",
       "      <td>5.178864</td>\n",
       "      <td>-0.105856</td>\n",
       "      <td>1.639140</td>\n",
       "      <td>-0.123185</td>\n",
       "      <td>-1.137013</td>\n",
       "      <td>-1.204642</td>\n",
       "      <td>-1.477192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>70580</th>\n",
       "      <td>-0.182166</td>\n",
       "      <td>-0.065463</td>\n",
       "      <td>-0.148792</td>\n",
       "      <td>-0.410637</td>\n",
       "      <td>-0.093696</td>\n",
       "      <td>5.178864</td>\n",
       "      <td>-0.105733</td>\n",
       "      <td>1.639140</td>\n",
       "      <td>-0.123045</td>\n",
       "      <td>-1.137013</td>\n",
       "      <td>-1.204642</td>\n",
       "      <td>-1.477192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>70581</th>\n",
       "      <td>-0.235009</td>\n",
       "      <td>-0.224132</td>\n",
       "      <td>-0.228371</td>\n",
       "      <td>-0.375802</td>\n",
       "      <td>-0.094151</td>\n",
       "      <td>-0.397497</td>\n",
       "      <td>-0.110662</td>\n",
       "      <td>0.109711</td>\n",
       "      <td>-0.128681</td>\n",
       "      <td>-1.137013</td>\n",
       "      <td>-1.203644</td>\n",
       "      <td>-1.477192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>70582</th>\n",
       "      <td>-0.234831</td>\n",
       "      <td>-0.219785</td>\n",
       "      <td>-0.221739</td>\n",
       "      <td>-0.409902</td>\n",
       "      <td>-0.094103</td>\n",
       "      <td>0.121771</td>\n",
       "      <td>-0.110563</td>\n",
       "      <td>-0.229110</td>\n",
       "      <td>-0.128482</td>\n",
       "      <td>-1.137013</td>\n",
       "      <td>-1.203644</td>\n",
       "      <td>-1.477192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>70583</th>\n",
       "      <td>-0.226609</td>\n",
       "      <td>-0.206003</td>\n",
       "      <td>-0.193887</td>\n",
       "      <td>-0.410843</td>\n",
       "      <td>-0.093796</td>\n",
       "      <td>0.483656</td>\n",
       "      <td>-0.109718</td>\n",
       "      <td>0.016602</td>\n",
       "      <td>-0.126828</td>\n",
       "      <td>-1.137013</td>\n",
       "      <td>-1.204642</td>\n",
       "      <td>-1.477192</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>70584 rows × 12 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       bytes_bothdir  bytes_onedir  total_packets  duration       pps  \\\n",
       "0          -0.239160     -0.226629      -0.241634 -0.411271  0.857007   \n",
       "1          -0.239160     -0.226629      -0.241634 -0.411271 -0.075575   \n",
       "2          -0.239160     -0.226629      -0.241634 -0.411271 -0.064796   \n",
       "3          -0.239160     -0.226629      -0.241634 -0.411271 -0.069122   \n",
       "4          -0.239160     -0.226629      -0.241634 -0.411271 -0.047980   \n",
       "...              ...           ...            ...       ...       ...   \n",
       "70579      -0.182166     -0.065463      -0.148792 -0.410621 -0.093707   \n",
       "70580      -0.182166     -0.065463      -0.148792 -0.410637 -0.093696   \n",
       "70581      -0.235009     -0.224132      -0.228371 -0.375802 -0.094151   \n",
       "70582      -0.234831     -0.219785      -0.221739 -0.409902 -0.094103   \n",
       "70583      -0.226609     -0.206003      -0.193887 -0.410843 -0.093796   \n",
       "\n",
       "       bps_onedir  bpp_onedir  bps_bothdir  bpp_bothdir  cnt_timed_src  \\\n",
       "0       -0.624921    0.581812    -0.836037     1.263197      -1.080234   \n",
       "1       -0.624921   -0.097139    -0.836037    -0.101503      -1.080234   \n",
       "2       -0.624921   -0.089291    -0.836037    -0.085729      -1.080234   \n",
       "3       -0.624921   -0.092441    -0.836037    -0.092060      -1.080234   \n",
       "4       -0.624921   -0.077048    -0.836037    -0.061121      -1.080234   \n",
       "...           ...         ...          ...          ...            ...   \n",
       "70579    5.178864   -0.105856     1.639140    -0.123185      -1.137013   \n",
       "70580    5.178864   -0.105733     1.639140    -0.123045      -1.137013   \n",
       "70581   -0.397497   -0.110662     0.109711    -0.128681      -1.137013   \n",
       "70582    0.121771   -0.110563    -0.229110    -0.128482      -1.137013   \n",
       "70583    0.483656   -0.109718     0.016602    -0.126828      -1.137013   \n",
       "\n",
       "       cnt_timed_dst  cnt_timed_srcdst  \n",
       "0          -1.204642         -1.477192  \n",
       "1          -1.204642         -1.477192  \n",
       "2          -1.204642         -1.477192  \n",
       "3          -1.204642         -1.477192  \n",
       "4          -1.204642         -1.477192  \n",
       "...              ...               ...  \n",
       "70579      -1.204642         -1.477192  \n",
       "70580      -1.204642         -1.477192  \n",
       "70581      -1.203644         -1.477192  \n",
       "70582      -1.203644         -1.477192  \n",
       "70583      -1.204642         -1.477192  \n",
       "\n",
       "[70584 rows x 12 columns]"
      ]
     },
     "execution_count": 107,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "adv_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train_original = df_train_f3\n",
    "X_train_mod = df_train_f3.copy()\n",
    "X_train_mod[selected_features] = adv_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "147.32.84.138     14775\n",
       "147.32.84.59      13401\n",
       "147.32.84.229      3480\n",
       "147.32.85.25       1973\n",
       "147.32.86.20       1307\n",
       "147.32.85.34       1214\n",
       "147.32.84.170       779\n",
       "147.32.84.94        774\n",
       "147.147.98.151      602\n",
       "147.124.93.156      599\n",
       "Name: src_ip, dtype: int64"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train_original.src_ip.value_counts()[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_anomaly = X_train_original[X_train_original['label']==-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 93,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "147.147.98.151     602\n",
       "147.124.93.156     599\n",
       "147.37.96.153      582\n",
       "147.236.100.149    581\n",
       "147.32.84.192      581\n",
       "147.231.96.180     579\n",
       "147.175.101.148    570\n",
       "147.177.96.192     560\n",
       "147.74.98.178      555\n",
       "147.94.93.195      552\n",
       "Name: src_ip, dtype: int64"
      ]
     },
     "execution_count": 93,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_anomaly.src_ip.value_counts()[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "metadata": {},
   "outputs": [],
   "source": [
    "bot_df = X_train_original[X_train_original['src_ip']=='147.147.98.151'] "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "14775"
      ]
     },
     "execution_count": 102,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bot_df = X_train_original[X_train_original['src_ip']=='147.32.84.138']\n",
    "len(bot_df)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 87,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "98908"
      ]
     },
     "execution_count": 87,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "anomaly = df_train_original['label']==-1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_bot_df = bot_df[selected_features]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "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>bytes_bothdir</th>\n",
       "      <th>bytes_onedir</th>\n",
       "      <th>total_packets</th>\n",
       "      <th>duration</th>\n",
       "      <th>pps</th>\n",
       "      <th>bps_onedir</th>\n",
       "      <th>bpp_onedir</th>\n",
       "      <th>bps_bothdir</th>\n",
       "      <th>bpp_bothdir</th>\n",
       "      <th>cnt_timed_src</th>\n",
       "      <th>cnt_timed_dst</th>\n",
       "      <th>cnt_timed_srcdst</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>15428</th>\n",
       "      <td>-0.037733</td>\n",
       "      <td>-0.023699</td>\n",
       "      <td>-0.031023</td>\n",
       "      <td>-0.122413</td>\n",
       "      <td>-0.294152</td>\n",
       "      <td>0.052320</td>\n",
       "      <td>-0.310663</td>\n",
       "      <td>-0.397636</td>\n",
       "      <td>-0.328687</td>\n",
       "      <td>-0.277976</td>\n",
       "      <td>-1.004642</td>\n",
       "      <td>-0.386971</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15429</th>\n",
       "      <td>-0.038419</td>\n",
       "      <td>-0.025559</td>\n",
       "      <td>-0.033676</td>\n",
       "      <td>-0.211261</td>\n",
       "      <td>-0.290957</td>\n",
       "      <td>-0.368939</td>\n",
       "      <td>-0.308027</td>\n",
       "      <td>-0.557725</td>\n",
       "      <td>-0.322911</td>\n",
       "      <td>-0.277976</td>\n",
       "      <td>-0.922823</td>\n",
       "      <td>-0.386971</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15430</th>\n",
       "      <td>0.725562</td>\n",
       "      <td>0.038679</td>\n",
       "      <td>0.609581</td>\n",
       "      <td>-0.211178</td>\n",
       "      <td>-0.272881</td>\n",
       "      <td>-0.475452</td>\n",
       "      <td>-0.297041</td>\n",
       "      <td>4.516604</td>\n",
       "      <td>0.184591</td>\n",
       "      <td>-0.277976</td>\n",
       "      <td>-0.922823</td>\n",
       "      <td>-0.386971</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15431</th>\n",
       "      <td>0.174108</td>\n",
       "      <td>-0.007072</td>\n",
       "      <td>0.150680</td>\n",
       "      <td>-0.211251</td>\n",
       "      <td>-0.265529</td>\n",
       "      <td>-0.469858</td>\n",
       "      <td>-0.292055</td>\n",
       "      <td>4.164975</td>\n",
       "      <td>0.317727</td>\n",
       "      <td>-0.277976</td>\n",
       "      <td>-0.922823</td>\n",
       "      <td>-0.386971</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15432</th>\n",
       "      <td>-0.038444</td>\n",
       "      <td>-0.025559</td>\n",
       "      <td>-0.033676</td>\n",
       "      <td>-0.211259</td>\n",
       "      <td>-0.291577</td>\n",
       "      <td>-0.368939</td>\n",
       "      <td>-0.308538</td>\n",
       "      <td>-0.568814</td>\n",
       "      <td>-0.324157</td>\n",
       "      <td>-0.277976</td>\n",
       "      <td>-0.922823</td>\n",
       "      <td>-0.386971</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       bytes_bothdir  bytes_onedir  total_packets  duration       pps  \\\n",
       "15428      -0.037733     -0.023699      -0.031023 -0.122413 -0.294152   \n",
       "15429      -0.038419     -0.025559      -0.033676 -0.211261 -0.290957   \n",
       "15430       0.725562      0.038679       0.609581 -0.211178 -0.272881   \n",
       "15431       0.174108     -0.007072       0.150680 -0.211251 -0.265529   \n",
       "15432      -0.038444     -0.025559      -0.033676 -0.211259 -0.291577   \n",
       "\n",
       "       bps_onedir  bpp_onedir  bps_bothdir  bpp_bothdir  cnt_timed_src  \\\n",
       "15428    0.052320   -0.310663    -0.397636    -0.328687      -0.277976   \n",
       "15429   -0.368939   -0.308027    -0.557725    -0.322911      -0.277976   \n",
       "15430   -0.475452   -0.297041     4.516604     0.184591      -0.277976   \n",
       "15431   -0.469858   -0.292055     4.164975     0.317727      -0.277976   \n",
       "15432   -0.368939   -0.308538    -0.568814    -0.324157      -0.277976   \n",
       "\n",
       "       cnt_timed_dst  cnt_timed_srcdst  \n",
       "15428      -1.004642         -0.386971  \n",
       "15429      -0.922823         -0.386971  \n",
       "15430      -0.922823         -0.386971  \n",
       "15431      -0.922823         -0.386971  \n",
       "15432      -0.922823         -0.386971  "
      ]
     },
     "execution_count": 106,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_bot_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[1, 1, 0, ..., 1, 1, 1]])"
      ]
     },
     "execution_count": 104,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bot_pred = model.predict_class(np.array(train_bot_df))\n",
    "bot_pred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[    0    37]\n",
      " [    1 14738]]\n"
     ]
    }
   ],
   "source": [
    "unique, counts = np.unique(bot_pred, return_counts=True)\n",
    "print (np.asarray((unique, counts)).T)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch:  10\n",
      "Error Rate without adversarial examples: 0.4140315085571801\n",
      "\n",
      "Error Rate with adversarial examples, epsilon = 0.2: 0.5039810721976652\n",
      "Error Rate with adversarial examples, epsilon = 0.5: 0.5039810721976652\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\kebin\\Anaconda3\\lib\\site-packages\\ipykernel_launcher.py:41: RuntimeWarning: overflow encountered in exp\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "(array([[-0.23916046, -0.22662853, -0.24163367, ..., -1.08023427,\n",
       "         -1.20464185, -1.47719155],\n",
       "        [-0.23916046, -0.22662853, -0.24163367, ..., -1.08023427,\n",
       "         -1.20464185, -1.47719155],\n",
       "        [-0.23916046, -0.22662853, -0.24163367, ..., -1.08023427,\n",
       "         -1.20464185, -1.47719155],\n",
       "        ...,\n",
       "        [-0.23500942, -0.22413222, -0.22837064, ..., -1.13701282,\n",
       "         -1.20364406, -1.47719155],\n",
       "        [-0.23483064, -0.21978513, -0.22173912, ..., -1.13701282,\n",
       "         -1.20364406, -1.47719155],\n",
       "        [-0.22660855, -0.20600252, -0.19388676, ..., -1.13701282,\n",
       "         -1.20464185, -1.47719155]]),\n",
       " array([[-0.53916046, -0.52662853, -0.54163367, ..., -1.38023427,\n",
       "         -1.50464185, -1.17719155],\n",
       "        [-0.53916046, -0.52662853, -0.54163367, ..., -1.38023427,\n",
       "         -1.50464185, -1.17719155],\n",
       "        [-0.53916046, -0.52662853, -0.54163367, ..., -1.38023427,\n",
       "         -1.50464185, -1.17719155],\n",
       "        ...,\n",
       "        [-0.53500942, -0.52413222, -0.52837064, ..., -1.43701282,\n",
       "         -1.50364406, -1.17719155],\n",
       "        [-0.53483064, -0.51978513, -0.52173912, ..., -1.43701282,\n",
       "         -1.50364406, -1.17719155],\n",
       "        [-0.52660855, -0.50600252, -0.49388676, ..., -1.43701282,\n",
       "         -1.50464185, -1.17719155]]))"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "show_adversarials(10, X, Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "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": "3.7.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}