V2Ray Deep Packet Inspection.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Step 1: Collect Packets Data\n",
    "\n",
    "- Follow build instruction in [PacketSorter repository](https://github.com/rickyzhang82/PacketSorter).\n",
    "- Run **PacketSorter** application in your home router for a month.\n",
    "- Deploy V2Ray server in AWS Lightsail.\n",
    "- Deploy V2Ray client through vmess protcol at home.\n",
    "- Use V2Ray client to generate V2Ray traffic at home."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Step 2: Offline Data Cleaning\n",
    "\n",
    "- Follow build instruction in [PcapProcessor repository](https://github.com/rickyzhang82/PcapProcessor).\n",
    "- Retain pcap files which are not in missing caputing log from **PacketSorter**.\n",
    "- Retain pcap files which contains extactly **16** packets.\n",
    "- Drop the link layer header in each packet, i.e. the Ethernet frame header.\n",
    "- Mask source's and destination's IP address and port with zero value. Recompute the checksum in TCP and IP header cheksum.\n",
    "- Truncate or pad zero to each packet to **1500** bytes."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Step 3: Load Modified Packets\n",
    "\n",
    "## Data Exploration"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import os.path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Packet data type\n",
    "FIXED_PACKET_SIZE = 1500\n",
    "NUM_OF_PACKETS = 16\n",
    "\n",
    "# v2ray server IP\n",
    "V2RAY_HOST_IP = # Fill in your V2Ray server IP\n",
    "# v2ray server port\n",
    "V2RAY_PORT = # Fill in your V2Ray server port\n",
    "# v2ray traffic tag\n",
    "V2RAY_HOST_TAG = '%s.%s' % (V2RAY_HOST_IP, V2RAY_PORT)\n",
    "\n",
    "# Load constant\n",
    "DATA_ROOT = '/src/data/tcpsorter/mod-tcpsorter'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(16, 1500)\n",
      "    0     1     2     3     4     5     6     7     8     9     ...  1490  \\\n",
      "0     69     0     2    57     0     0    64     0    64     6  ...     0   \n",
      "1     69     0     5   220     0     0    64     0    56     6  ...   104   \n",
      "2     69     0     5   220     0     1    64     0    56     6  ...   228   \n",
      "3     69     0     1   135     0     2    64     0    56     6  ...     0   \n",
      "4     69     0     0   145     0     0    64     0    64     6  ...     0   \n",
      "5     69     0     0    58     0     0    64     0    56     6  ...     0   \n",
      "6     69     0     0    97     0     0    64     0    56     6  ...     0   \n",
      "7     69     0     2   102     0     0    64     0    64     6  ...     0   \n",
      "8     69     0     2   122     0     0    64     0    56     6  ...     0   \n",
      "9     69     0     5   220     0     0    64     0    56     6  ...   118   \n",
      "10    69     0     2   188     0     1    64     0    56     6  ...     0   \n",
      "11    69     0     2   111     0     0    64     0    64     6  ...     0   \n",
      "12    69     0     2   133     0     0    64     0    56     6  ...     0   \n",
      "13    69     0     5   220     0     0    64     0    56     6  ...   143   \n",
      "14    69     0     5   220     0     1    64     0    56     6  ...    37   \n",
      "15    69     0     5   220     0     2    64     0    56     6  ...   232   \n",
      "\n",
      "    1491  1492  1493  1494  1495  1496  1497  1498  1499  \n",
      "0      0     0     0     0     0     0     0     0     0  \n",
      "1    207    89   211    88    55   169   157   225    89  \n",
      "2    184   119   173    45   110   245    82   221   180  \n",
      "3      0     0     0     0     0     0     0     0     0  \n",
      "4      0     0     0     0     0     0     0     0     0  \n",
      "5      0     0     0     0     0     0     0     0     0  \n",
      "6      0     0     0     0     0     0     0     0     0  \n",
      "7      0     0     0     0     0     0     0     0     0  \n",
      "8      0     0     0     0     0     0     0     0     0  \n",
      "9    151    13     8   252    71   201   173   105    71  \n",
      "10     0     0     0     0     0     0     0     0     0  \n",
      "11     0     0     0     0     0     0     0     0     0  \n",
      "12     0     0     0     0     0     0     0     0     0  \n",
      "13    45    49   183    17   143   192   237    55   180  \n",
      "14   248   245   218   143    46   142   248   198    15  \n",
      "15    93    99   234    53    39    91   184   239    34  \n",
      "\n",
      "[16 rows x 1500 columns]\n"
     ]
    }
   ],
   "source": [
    "MOD_FILE_BASE = '192.168.3.37.65516-45.57.24.141.443.bin'\n",
    "MOD_FILE_PATH = os.path.join(DATA_ROOT, '26DEC2019', MOD_FILE_BASE)\n",
    "\n",
    "packet_dt = np.dtype((np.uint8, FIXED_PACKET_SIZE))\n",
    "df = pd.DataFrame(np.fromfile(MOD_FILE_PATH, dtype=packet_dt, count=NUM_OF_PACKETS))\n",
    "print(df.shape)\n",
    "print(df)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Create Data Generator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import math\n",
    "import os.path\n",
    "from pathlib import Path\n",
    "import glob\n",
    "from tensorflow.keras.utils import Sequence\n",
    "\n",
    "FIXED_PACKET_SIZE = 1500\n",
    "NUM_OF_PACKETS_PER_FILE = 16\n",
    "RESCALE_FACTOR = 1./255\n",
    "\n",
    "# v2ray traffic tag\n",
    "TRAINING_DATA_PERCENTAGE = 0.8\n",
    "PACKET_FILE_EXT = '*.bin'\n",
    "\n",
    "\n",
    "def rglob(data_root, file_ext):\n",
    "    files = list()\n",
    "    for filePath in Path(data_root).rglob(file_ext):\n",
    "        files.append(str(filePath))\n",
    "    return files\n",
    "\n",
    "\n",
    "def binary_classification(packet_path, match_string=V2RAY_HOST_TAG):\n",
    "    \"\"\"Binary network traffic classification function\n",
    "\n",
    "    :param packet_path: file path to packet\n",
    "    :param match_string:\n",
    "    :return: 1, if it is v2ray traffice. 0, otherwise.\n",
    "    \"\"\"\n",
    "    if packet_path.find(match_string) != -1:\n",
    "        return 1\n",
    "    else:\n",
    "        return 0\n",
    "\n",
    "\n",
    "def generate_train_validation_packet_path_list(data_root, training_pct=TRAINING_DATA_PERCENTAGE, eqaul_size=True):\n",
    "    file_list = rglob(data_root, PACKET_FILE_EXT)\n",
    "    v2ray_file_list = [file_path for file_path in file_list if binary_classification(file_path) == 1]\n",
    "    non_v2ray_file_list = [file_path for file_path in file_list if binary_classification(file_path) == 0]\n",
    "\n",
    "    if eqaul_size:\n",
    "        cut_off_count = min(len(v2ray_file_list), len(non_v2ray_file_list))\n",
    "        v2ray_file_size = cut_off_count\n",
    "        non_v2ray_file_size = cut_off_count\n",
    "    else:\n",
    "        v2ray_file_size = len(v2ray_file_list)\n",
    "        non_v2ray_file_size = len(non_v2ray_file_list)\n",
    "\n",
    "    v2ray_indexes = np.arange(len(v2ray_file_list))\n",
    "    np.random.shuffle(v2ray_indexes)\n",
    "    non_v2ray_indexes = np.arange(len(non_v2ray_file_list))\n",
    "    np.random.shuffle(non_v2ray_indexes)\n",
    "\n",
    "    training_file_list = [v2ray_file_list[index]\n",
    "                          for index in v2ray_indexes[:math.ceil(v2ray_file_size * training_pct)]] + \\\n",
    "                         [non_v2ray_file_list[index]\n",
    "                          for index in non_v2ray_indexes[:math.ceil(non_v2ray_file_size * training_pct)]]\n",
    "\n",
    "    validation_file_list = [v2ray_file_list[index]\n",
    "                            for index in v2ray_indexes[math.ceil(v2ray_file_size * training_pct): v2ray_file_size]] + \\\n",
    "                           [non_v2ray_file_list[index]\n",
    "                            for index in non_v2ray_indexes[math.ceil(non_v2ray_file_size * training_pct): non_v2ray_file_size]]\n",
    "\n",
    "    print(\"Statistics: \")\n",
    "    print(\"Total V2ray traffic %d, Total non-V2ray traffic %d\" % (len(v2ray_file_list), len(non_v2ray_file_list)))\n",
    "    print(\"Output train traffic %d, Total validation traffic %d\" % (len(training_file_list), len(validation_file_list)))\n",
    "\n",
    "    return training_file_list, validation_file_list\n",
    "\n",
    "\n",
    "class PacketDataGenerator(Sequence):\n",
    "    \"\"\"Generates data for Keras\n",
    "    Sequence based data generator. Suitable for building data generator for training and prediction.\n",
    "    \"\"\"\n",
    "    def __init__(self, data_path_list, classification_func=binary_classification, rescale=RESCALE_FACTOR,\n",
    "                 batch_size=32, dim=(NUM_OF_PACKETS_PER_FILE, FIXED_PACKET_SIZE), shuffle=True, predict=False):\n",
    "        \"\"\"Initialization\n",
    "\n",
    "        :param data_path_list: list of data path\n",
    "        :param classification_func: binary classification function\n",
    "        :param batch_size: batch size at each iteration\n",
    "        :param dim: tuple indicating data dimension\n",
    "        :param shuffle: True to shuffle label indexes after every epoch\n",
    "        \"\"\"\n",
    "        self.data_path_list = data_path_list\n",
    "        self.classification_func = classification_func\n",
    "        self.rescale = rescale\n",
    "        self.batch_size = batch_size\n",
    "        self.dim = dim\n",
    "        self.shuffle = shuffle\n",
    "        self.predict = predict\n",
    "        self.indexes = np.arange(len(self.data_path_list))\n",
    "        self.on_epoch_end()\n",
    "\n",
    "    def __len__(self):\n",
    "        \"\"\"Denotes the number samples of batches per epoch\n",
    "\n",
    "        :return: number of batches per epoch\n",
    "        \"\"\"\n",
    "        return int(np.floor(len(self.data_path_list) / self.batch_size))\n",
    "\n",
    "    def __getitem__(self, index):\n",
    "        \"\"\"Generate one batch of data\n",
    "\n",
    "        :param index: index of the batch\n",
    "        :return: X and y when fitting. X only when predicting\n",
    "        \"\"\"\n",
    "        # Generate indexes of the batch\n",
    "        indexes = self.indexes[index * self.batch_size:(index + 1) * self.batch_size]\n",
    "\n",
    "        # Generate sub packet path list\n",
    "        data_path_sub_list = [self.data_path_list[k] for k in indexes]\n",
    "\n",
    "        # Generate data\n",
    "        x = self.standardize(self._generate_x(data_path_sub_list))\n",
    "        if self.predict:\n",
    "            return x\n",
    "        else:\n",
    "            y = self._generate_y(data_path_sub_list)\n",
    "            return x, y\n",
    "\n",
    "    def standardize(self, x):\n",
    "        if self.rescale:\n",
    "            x *= self.rescale\n",
    "        return x\n",
    "\n",
    "    def on_epoch_end(self):\n",
    "        \"\"\"Updates indexes after each epoch\n",
    "\n",
    "        \"\"\"\n",
    "        if self.shuffle:\n",
    "            np.random.shuffle(self.indexes)\n",
    "\n",
    "    def _generate_x(self, file_path_list):\n",
    "        \"\"\"Generates data containing batch_size data\n",
    "\n",
    "        :param file_path_list: file path list\n",
    "        :return: batch of data\n",
    "        \"\"\"\n",
    "        # Initialization\n",
    "        x = np.empty((self.batch_size, *self.dim))\n",
    "\n",
    "        # Generate data\n",
    "        for i, file_path in enumerate(file_path_list):\n",
    "            # Store sample\n",
    "            x[i, ] = self._load_packet(file_path)\n",
    "\n",
    "        return x\n",
    "\n",
    "    def _generate_y(self, file_path_list):\n",
    "        \"\"\"Generates data containing batch_size\n",
    "\n",
    "        :param file_path_list: file path list\n",
    "        :return: batch if masks\n",
    "        \"\"\"\n",
    "        y = np.empty((self.batch_size, ), dtype=int)\n",
    "\n",
    "        # Generate data\n",
    "        for i, file_path in enumerate(file_path_list):\n",
    "            y[i,] = self.classification_func(file_path)\n",
    "\n",
    "        return y\n",
    "\n",
    "    def _load_packet(self, packet_path):\n",
    "        \"\"\"Load packet\n",
    "\n",
    "        :param packet_path: path to packet\n",
    "        :return: packet in numpy array\n",
    "        \"\"\"\n",
    "        packet_dt = np.dtype((np.uint8, FIXED_PACKET_SIZE))\n",
    "        packet_data = np.fromfile(packet_path, dtype=packet_dt, count=NUM_OF_PACKETS_PER_FILE)\n",
    "        return packet_data.astype(np.float32)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Statistics: \n",
      "Total V2ray traffic 5070, Total non-V2ray traffic 190531\n",
      "Output train traffic 8112, Total validation traffic 2028\n"
     ]
    }
   ],
   "source": [
    "# Generate training data and validation file list\n",
    "train_file_list, val_file_list = generate_train_validation_packet_path_list(data_root=DATA_ROOT, eqaul_size=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create train data generator and validation data generator\n",
    "train_data_generator = PacketDataGenerator(train_file_list)\n",
    "val_data_generator = PacketDataGenerator(val_file_list)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "253"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(train_data_generator)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(32, 16, 1500)\n",
      "[[[ 0.27058824  0.          0.00784314 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.00392157 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]\n",
      "  ..., \n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.00392157 ...,  0.          0.          0.        ]]\n",
      "\n",
      " [[ 0.27058824  0.          0.00392157 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.01960784 ...,  0.31764706  0.19607843\n",
      "    0.9254902 ]\n",
      "  [ 0.27058824  0.          0.00784314 ...,  0.          0.          0.        ]\n",
      "  ..., \n",
      "  [ 0.27058824  0.          0.01960784 ...,  0.70196078  0.89803922\n",
      "    0.03921569]\n",
      "  [ 0.27058824  0.          0.01960784 ...,  0.85490196  0.80784314\n",
      "    0.31764706]\n",
      "  [ 0.27058824  0.          0.01960784 ...,  0.89803922  0.8745098\n",
      "    0.42745098]]\n",
      "\n",
      " [[ 0.27058824  0.          0.00784314 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.01960784 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.01960784 ...,  0.          0.          0.        ]\n",
      "  ..., \n",
      "  [ 0.27058824  0.          0.00392157 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.01960784 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.00392157 ...,  0.          0.          0.        ]]\n",
      "\n",
      " ..., \n",
      " [[ 0.27058824  0.          0.00784314 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.01960784 ...,  0.19215686  0.36862745\n",
      "    0.44313725]\n",
      "  [ 0.27058824  0.          0.01960784 ...,  0.86666667  0.70588235\n",
      "    0.45490196]\n",
      "  ..., \n",
      "  [ 0.27058824  0.          0.00784314 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.00784314 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]]\n",
      "\n",
      " [[ 0.27058824  0.          0.00784314 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.00392157 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]\n",
      "  ..., \n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]]\n",
      "\n",
      " [[ 0.27058824  0.          0.00784314 ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]\n",
      "  ..., \n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]\n",
      "  [ 0.27058824  0.          0.         ...,  0.          0.          0.        ]]]\n",
      "(32,)\n",
      "[1 1 0 1 1 1 0 1 1 1 1 0 1 1 0 1 0 0 1 0 1 0 0 0 1 1 0 1 0 0 1 0]\n"
     ]
    }
   ],
   "source": [
    "x, y = train_data_generator.__getitem__(0)\n",
    "# 32 samples in each batch, 16 packets in each connection, 1500 byte in each packet\n",
    "print(np.shape(x))\n",
    "print(x)\n",
    "print(np.shape(y))\n",
    "print(y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Step 4: Classification\n",
    "\n",
    "## Model Construction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv1d_1 (Conv1D)            (None, 12, 200)           1500200   \n",
      "_________________________________________________________________\n",
      "conv1d_2 (Conv1D)            (None, 9, 100)            80100     \n",
      "_________________________________________________________________\n",
      "max_pooling1d_1 (MaxPooling1 (None, 4, 100)            0         \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 4, 100)            0         \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 400)               0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 256)               102656    \n",
      "_________________________________________________________________\n",
      "dropout_2 (Dropout)          (None, 256)               0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 128)               32896     \n",
      "_________________________________________________________________\n",
      "dropout_3 (Dropout)          (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 32)                4128      \n",
      "_________________________________________________________________\n",
      "dropout_4 (Dropout)          (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 1)                 33        \n",
      "=================================================================\n",
      "Total params: 1,720,013\n",
      "Trainable params: 1,720,013\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "from keras import layers\n",
    "from keras import models\n",
    "\n",
    "model = models.Sequential()\n",
    "\n",
    "# Convnet\n",
    "model.add(layers.Conv1D(200, 5, activation='relu',\n",
    "                        input_shape=(16, 1500), data_format=\"channels_last\"))\n",
    "model.add(layers.Conv1D(100, 4, activation='relu'))\n",
    "model.add(layers.MaxPooling1D(2))\n",
    "model.add(layers.Dropout(0.25))\n",
    "\n",
    "# Fully connected network\n",
    "model.add(layers.Flatten())\n",
    "model.add(layers.Dense(256))\n",
    "model.add(layers.Dropout(0.05))\n",
    "model.add(layers.Dense(128))\n",
    "model.add(layers.Dropout(0.05))\n",
    "model.add(layers.Dense(32))\n",
    "model.add(layers.Dropout(0.05))\n",
    "model.add(layers.Dense(1, activation='sigmoid'))\n",
    "\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras import optimizers\n",
    "\n",
    "model.compile(loss='binary_crossentropy',\n",
    "              optimizer=optimizers.RMSprop(lr=1e-4),\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/10\n",
      "253/253 [==============================] - 111s 440ms/step - loss: 0.1287 - acc: 0.9502 - val_loss: 0.0139 - val_acc: 0.9970\n",
      "Epoch 2/10\n",
      "253/253 [==============================] - 2s 9ms/step - loss: 0.0244 - acc: 0.9909 - val_loss: 0.0055 - val_acc: 0.9975\n",
      "Epoch 3/10\n",
      "253/253 [==============================] - 2s 9ms/step - loss: 0.0142 - acc: 0.9951 - val_loss: 0.0051 - val_acc: 0.9980\n",
      "Epoch 4/10\n",
      "253/253 [==============================] - 2s 8ms/step - loss: 0.0087 - acc: 0.9973 - val_loss: 0.0068 - val_acc: 0.9975\n",
      "Epoch 5/10\n",
      "253/253 [==============================] - 2s 9ms/step - loss: 0.0070 - acc: 0.9979 - val_loss: 0.0134 - val_acc: 0.9960\n",
      "Epoch 6/10\n",
      "253/253 [==============================] - 2s 9ms/step - loss: 0.0065 - acc: 0.9984 - val_loss: 0.0020 - val_acc: 0.9990\n",
      "Epoch 7/10\n",
      "253/253 [==============================] - 2s 8ms/step - loss: 0.0025 - acc: 0.9989 - val_loss: 0.0068 - val_acc: 0.9970\n",
      "Epoch 8/10\n",
      "253/253 [==============================] - 2s 8ms/step - loss: 0.0034 - acc: 0.9991 - val_loss: 0.0012 - val_acc: 0.9990\n",
      "Epoch 9/10\n",
      "253/253 [==============================] - 2s 9ms/step - loss: 0.0069 - acc: 0.9988 - val_loss: 0.0038 - val_acc: 0.9985\n",
      "Epoch 10/10\n",
      "253/253 [==============================] - 2s 8ms/step - loss: 0.0021 - acc: 0.9993 - val_loss: 0.0052 - val_acc: 0.9980\n"
     ]
    }
   ],
   "source": [
    "from keras.callbacks import EarlyStopping\n",
    "# Create train data generator and validation data generator\n",
    "train_generator = PacketDataGenerator(train_file_list)\n",
    "validation_generator = PacketDataGenerator(val_file_list)\n",
    "\n",
    "#es = EarlyStopping(monitor='val_loss', mode='auto', verbose=1)\n",
    "\n",
    "history = model.fit_generator(\n",
    "      train_generator,\n",
    "      epochs=10,\n",
    "      #callbacks=[es],\n",
    "      validation_data=validation_generator,\n",
    "      workers=3,\n",
    "      use_multiprocessing=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "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": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "acc = history.history['acc']\n",
    "val_acc = history.history['val_acc']\n",
    "loss = history.history['loss']\n",
    "val_loss = history.history['val_loss']\n",
    "\n",
    "epochs = range(len(acc))\n",
    "\n",
    "plt.plot(epochs, acc, 'bo', label='Training acc')\n",
    "plt.plot(epochs, val_acc, 'b', label='Validation acc')\n",
    "plt.title('Training and validation accuracy')\n",
    "plt.legend()\n",
    "\n",
    "plt.figure()\n",
    "\n",
    "plt.plot(epochs, loss, 'bo', label='Training loss')\n",
    "plt.plot(epochs, val_loss, 'b', label='Validation loss')\n",
    "plt.title('Training and validation loss')\n",
    "plt.legend()\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Ricky comments:\n",
    "\n",
    "- We have equal size of V2Ray traffic and non-V2Ray traffic. Both have **5070** samples.\n",
    "- We set aside 20% of data for validation. Both training data and validation data have the same 1:1 ratio of V2Ray traffic and non-V2Ray traffic.\n",
    "- The training accuracy rate can reach 99% after 1st epoch. So is validation accuracy is above 99%."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Early Stopping\n",
    "\n",
    "- To keep the best model, we apply early stopping techinque."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/1\n",
      "253/253 [==============================] - 2s 9ms/step - loss: 0.0016 - acc: 0.9994 - val_loss: 0.0028 - val_acc: 0.9985\n"
     ]
    }
   ],
   "source": [
    "from keras.callbacks import EarlyStopping\n",
    "# Create train data generator and validation data generator\n",
    "train_generator = PacketDataGenerator(train_file_list)\n",
    "validation_generator = PacketDataGenerator(val_file_list)\n",
    "\n",
    "es = EarlyStopping(monitor='val_loss', mode='auto', verbose=1)\n",
    "\n",
    "history = model.fit_generator(\n",
    "      train_generator,\n",
    "      callbacks=[es],\n",
    "      validation_data=validation_generator,\n",
    "      workers=3,\n",
    "      use_multiprocessing=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Save this model for later use\n",
    "from keras.models import load_model\n",
    "model.save('/src/data/tcpsorter/model/v2ray_classifier_equal_sample.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Evaluate Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "63/63 [==============================] - 0s 7ms/step\n",
      "shape of predict: (2016,), shape of real: (2016,)\n"
     ]
    }
   ],
   "source": [
    "def evaulate(file_list):\n",
    "    # predict y\n",
    "    predict_generator = PacketDataGenerator(file_list, predict=True, shuffle=False)\n",
    "    y_predict = model.predict_generator(predict_generator, workers=3, use_multiprocessing=True, verbose=1)\n",
    "    y_predict = y_predict.ravel()\n",
    "    # get real y\n",
    "    data_generator = PacketDataGenerator(file_list, predict=False, shuffle=False)\n",
    "    y_stat = None\n",
    "    x = 1\n",
    "    for _, y in data_generator:\n",
    "        if x > len(data_generator):\n",
    "            break\n",
    "        if y_stat is None:\n",
    "            y_stat = y\n",
    "        else:\n",
    "            y_stat = np.concatenate((y_stat, y), axis=0)\n",
    "        x+=1\n",
    "    return y_predict, y_stat\n",
    "\n",
    "from sklearn.metrics import roc_curve\n",
    "\n",
    "y_predict, y_stat = evaulate(val_file_list)\n",
    "print('shape of predict: %s, shape of real: %s' % (y_predict.shape, y_stat.shape))\n",
    "fpr_keras, tpr_keras, thresholds_keras = roc_curve(y_stat, y_predict)\n",
    "\n",
    "from sklearn.metrics import auc\n",
    "auc_keras = auc(fpr_keras, tpr_keras)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "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": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(1)\n",
    "plt.plot([0, 1], [0, 1], 'k--')\n",
    "plt.plot(fpr_keras, tpr_keras, label='Keras (area = {:.3f})'.format(auc_keras))\n",
    "\n",
    "plt.xlabel('False positive rate')\n",
    "plt.ylabel('True positive rate')\n",
    "plt.title('ROC curve')\n",
    "plt.legend(loc='best')\n",
    "plt.show()\n",
    "\n",
    "# Zoom in view of the upper left corner.\n",
    "plt.figure(2)\n",
    "plt.xlim(0, 0.2)\n",
    "plt.ylim(0.8, 1)\n",
    "plt.plot([0, 1], [0, 1], 'k--')\n",
    "plt.plot(fpr_keras, tpr_keras, label='Keras (area = {:.3f})'.format(auc_keras))\n",
    "\n",
    "\n",
    "plt.xlabel('False positive rate')\n",
    "plt.ylabel('True positive rate')\n",
    "plt.title('ROC curve (zoomed in at top left)')\n",
    "plt.legend(loc='best')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Ricky comments:\n",
    "\n",
    "- Perfect ROC curve."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Step 5: More Evaluation with unseen V2Ray traffic\n",
    "\n",
    "- Change vmess protocol's UUID in both server side and client side.\n",
    "- Use previously trained model to classfiy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Load Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.models import load_model\n",
    "\n",
    "model = load_model('/src/data/tcpsorter/model/v2ray_classifier_equal_sample.h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## unseen vmess Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load constant\n",
    "DATA_ROOT = '/src/data/tcpsorter/mod-tcpsorter.uuid.fb46221f-8f70-4f06-8b23-e2b61253c054'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Statistics: \n",
      "Total V2ray traffic 976, Total non-V2ray traffic 1522\n",
      "Output train traffic 2498, Total validation traffic 0\n"
     ]
    }
   ],
   "source": [
    "eval_file_list, _ = generate_train_validation_packet_path_list(data_root=DATA_ROOT, \n",
    "                                                               training_pct=1, eqaul_size=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "78/78 [==============================] - 7s 84ms/step\n"
     ]
    }
   ],
   "source": [
    "# Create evaluation generator\n",
    "eval_generator = PacketDataGenerator(eval_file_list, shuffle=False)\n",
    "\n",
    "eval_result = model.evaluate_generator(eval_generator,workers=3, use_multiprocessing=True, verbose=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0.013206126600994668, 0.99879807692307687]\n"
     ]
    }
   ],
   "source": [
    "print(eval_result)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "78/78 [==============================] - 1s 7ms/step\n",
      "shape of predict: (2496,), shape of real: (2496,)\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": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import roc_curve\n",
    "\n",
    "y_predict, y_stat = evaulate(eval_file_list)\n",
    "print('shape of predict: %s, shape of real: %s' % (y_predict.shape, y_stat.shape))\n",
    "fpr_keras, tpr_keras, thresholds_keras = roc_curve(y_stat, y_predict)\n",
    "\n",
    "from sklearn.metrics import auc\n",
    "auc_keras = auc(fpr_keras, tpr_keras)\n",
    "\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "plt.figure(1)\n",
    "plt.plot([0, 1], [0, 1], 'k--')\n",
    "plt.plot(fpr_keras, tpr_keras, label='Keras (area = {:.3f})'.format(auc_keras))\n",
    "\n",
    "plt.xlabel('False positive rate')\n",
    "plt.ylabel('True positive rate')\n",
    "plt.title('ROC curve')\n",
    "plt.legend(loc='best')\n",
    "plt.show()\n",
    "\n",
    "# Zoom in view of the upper left corner.\n",
    "plt.figure(2)\n",
    "plt.xlim(0, 0.2)\n",
    "plt.ylim(0.8, 1)\n",
    "plt.plot([0, 1], [0, 1], 'k--')\n",
    "plt.plot(fpr_keras, tpr_keras, label='Keras (area = {:.3f})'.format(auc_keras))\n",
    "\n",
    "\n",
    "plt.xlabel('False positive rate')\n",
    "plt.ylabel('True positive rate')\n",
    "plt.title('ROC curve (zoomed in at top left)')\n",
    "plt.legend(loc='best')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Ricky comments:\n",
    "\n",
    "- Even UUID of vmess protocol has been changed, the previously trained model can still reach **0.99879** accuracy!\n",
    "- Another perfect ROC curve."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Conclusion\n",
    "\n",
    "- Yes, [The patent claim](https://files.catbox.moe/vmzj04.pdf) is valid. \n",
    "- People who use V2Ray are fucked. Big borther knows you are using V2Ray without knowing V2Ray server IP or port. They can block it at any time at will\n",
    "- A new question arises: how can we bypass it?"
   ]
  },
  {
   "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}