some_looking_around.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "import torch\n",
    "from VAE1_mix import VAE3\n",
    "from datasets import load_dataset\n",
    "import torch.optim as optim\n",
    "import torch.utils.data\n",
    "from IPython.display import clear_output\n",
    "from sklearn.preprocessing import OneHotEncoder, StandardScaler, MinMaxScaler, LabelEncoder\n",
    "from sklearn.compose import ColumnTransformer\n",
    "from sklearn.model_selection import train_test_split\n",
    "import torch.nn.functional as F"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# full_data=pd.concat([X2,y], axis=1)\n",
    "# values_to_match=['ARP_poisioning', 'DDOS_Slowloris', 'DOS_SYN_Hping',\n",
    "#                  'Metasploit_Brute_Force_SSH', 'NMAP_FIN_SCAN',\n",
    "#                  'NMAP_OS_DETECTION', 'NMAP_TCP_scan', 'NMAP_UDP_SCAN',\n",
    "#                  'NMAP_XMAS_TREE_SCAN']\n",
    "# anomalous_data=full_data[full_data['Attack_type'].isin(values_to_match)]\n",
    "# anomalous_data=anomalous_data.iloc[:, :-1]\n",
    "# # anomalous_data\n",
    "# print(anomalous_data[anomalous_data.notna().any(axis=1)].shape, anomalous_data.shape)\n",
    "# anomalous_data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = load_dataset(\"aamui/rt_iot_2022\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "df=dataset['train'].to_pandas().filter(regex='^((?!max|min|avg|std).)*$')\n",
    "\n",
    "df = df.drop(columns = ['id.orig_p', 'id.resp_p']) # drop IDs\n",
    "\n",
    "# one-hot encode categoricals and standardize numerics\n",
    "cat_cols = ['proto', 'service']\n",
    "quant_cols = df.select_dtypes(include=['int64', 'float64']).columns.tolist()\n",
    "\n",
    "preprocessor = ColumnTransformer(\n",
    "    transformers = [\n",
    "        ('quant', StandardScaler(), quant_cols),\n",
    "        ('cat', OneHotEncoder(), cat_cols)\n",
    "    ], \n",
    "    remainder = 'passthrough')\n",
    "\n",
    "df_processed = preprocessor.fit_transform(df)\n",
    "column_names = quant_cols + preprocessor.named_transformers_['cat'].get_feature_names_out(cat_cols).tolist()\n",
    "column_names += df.columns.drop(quant_cols + cat_cols).tolist()\n",
    "df = pd.DataFrame(df_processed, columns = column_names)\n",
    "\n",
    "# print('Final preprocessed data shape:', df.shape)\n",
    "\n",
    "def binary_label_convert(data):\n",
    "    normal_types = ['Wipro_bulb', 'Thing_Speak', 'MQTT_Publish', 'Alexa']\n",
    "    data['Binary_label'] = data['Attack_type'].apply(lambda x: 0 if x in normal_types else 1)\n",
    "    return data\n",
    "\n",
    "df = binary_label_convert(df)\n",
    "\n",
    "# performing splits for train, cv, test\n",
    "normal = df[df['Binary_label'] == 0]\n",
    "attack = df[df['Binary_label'] == 1]\n",
    "\n",
    "# normal data: 70% for training and 30% for testing/validation\n",
    "train_data, temp_normal = train_test_split(normal, test_size = 0.3, random_state = 21, stratify = normal['Attack_type'])\n",
    "# further split normal data: 15% for cv and 15% for test\n",
    "validate_normal, test_normal = train_test_split(temp_normal, test_size = 0.5, random_state = 21)\n",
    "# attack data: 50% for cv, 50% for test\n",
    "validate_attack, test_attack = train_test_split(attack, test_size = 0.5, random_state = 21)\n",
    "\n",
    "# combine\n",
    "cv_data = pd.concat([validate_normal, validate_attack])\n",
    "test_data = pd.concat([test_normal, test_attack])\n",
    "\n",
    "# final split variables\n",
    "X_train = train_data.drop(columns = ['Attack_type', 'Binary_label'])\n",
    "y_train = pd.DataFrame(train_data['Binary_label'])\n",
    "type_train = train_data[[\"Attack_type\"]]\n",
    "\n",
    "X_cv = cv_data.drop(columns = ['Attack_type', 'Binary_label'])\n",
    "y_cv = pd.DataFrame(cv_data['Binary_label'])\n",
    "type_cv = cv_data[[\"Attack_type\"]]\n",
    "\n",
    "X_test = test_data.drop(columns = ['Attack_type', 'Binary_label'])\n",
    "y_test = pd.DataFrame(test_data['Binary_label'])\n",
    "type_test = test_data[[\"Attack_type\"]]\n",
    "\n",
    "# convert to tensors\n",
    "X_train = X_train.astype('float32')\n",
    "X_train = torch.tensor(X_train.values, dtype = torch.float32)\n",
    "y_train = torch.tensor(y_train.values, dtype = torch.float32)\n",
    "\n",
    "X_cv = X_cv.astype('float32')\n",
    "X_cv = torch.tensor(X_cv.values, dtype = torch.float32)\n",
    "y_cv = torch.tensor(y_cv.values, dtype = torch.float32)\n",
    "\n",
    "X_test = X_test.astype('float32')\n",
    "X_test = torch.tensor(X_test.values, dtype = torch.float32)\n",
    "y_test = torch.tensor(y_test.values, dtype = torch.float32)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# https://arxiv.org/pdf/2006.08204\n",
    "# def beta_elbow(x_hat, x, beta, N, z_mu, z_sigma):\n",
    "#     # this loss assumes that all columns are continuous which i believe is true\n",
    "#     beta_cross_entropy=-((beta+1)/beta) * torch.mean(torch.exp((-0.5 * z_sigma * beta) * torch.sum((x_hat - x)**2))-1)\n",
    "#     kl_div=torch.mean(0.5 * torch.sum(1 + 2 * torch.log(z_sigma) - (z_mu**2)  - (z_sigma**2), -1))\n",
    "#     return beta_cross_entropy + kl_div"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_loss(loss):\n",
    "    epoch = len(loss)\n",
    "    plt.plot(1+np.arange(epoch), loss, marker='o', color='red')\n",
    "    plt.ylim(bottom=0)\n",
    "    plt.title('ELBO Loss during training')\n",
    "    plt.xlabel('Epoch')\n",
    "    plt.ylabel('ELBO Loss')\n",
    "    plt.grid()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 64\n",
    "cat_feature_idx=[i for i in range(43,56)]\n",
    "cat_feature_idx.extend([17,18])\n",
    "train_dataset = torch.utils.data.TensorDataset(X_train)\n",
    "train_loader = torch.utils.data.DataLoader(train_dataset, batch_size = batch_size, shuffle = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 1000x400 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch [40/40], ELBO Loss: 8.5897\n"
     ]
    }
   ],
   "source": [
    "vae3=VAE3(z_dim=40, input_dim=X_train.shape[1], h_dim1=400, h_dim2=300)\n",
    "optimizer=optim.Adam(vae3.parameters(), lr=1e-6)\n",
    "beta_losses=[]\n",
    "epochs=40\n",
    "\n",
    "for epoch in range(epochs):\n",
    "    vae3.train()\n",
    "    for data in train_loader:\n",
    "        data=data[0]\n",
    "        optimizer.zero_grad()\n",
    "        recon, mu, sigma=vae3(data)\n",
    "        loss, _ =vae3.combo_elbo(recon, data, mu, sigma, cat_feature_indicies=cat_feature_idx)\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "    beta_losses.append(loss.item())\n",
    "\n",
    "    clear_output(wait=True)\n",
    "    plt.figure(figsize=(10,4))\n",
    "    plot_loss(beta_losses)\n",
    "    plt.xlim(right=40)\n",
    "    plt.show()\n",
    "\n",
    "    print(f'Epoch [{epoch + 1}/{40}], ELBO Loss: {loss.item():.4f}')\n"
   ]
  }
 ],
 "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": "3.11.9"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}