DL-based FraudDetection in BAF dataset .ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "5744b134",
   "metadata": {},
   "source": [
    "<div align=\"center\">\n",
    "    <br><b><font size=\"6\">Unmasking Fraud in Bank Account Openings - Demo 2</font></b><br><br>\n",
    "</div>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "132aa7f5",
   "metadata": {},
   "source": [
    "In this notebook, we will dive into more **deep learning-based methods** to address fraud detection in BAF dataset."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "13667c93",
   "metadata": {},
   "source": [
    "# Outline:\n",
    "<br>\n",
    "<ul>\n",
    "    <li><b>Dataset Preparation</b></li><br>\n",
    "    <li><b>Methods</b></li>\n",
    "    <ul>\n",
    "        <li><b>TabNet</b></li>\n",
    "        <li><b>Autoencoder</b></li>\n",
    "        <li><b>ICL</b></li>\n",
    "    </ul>\n",
    "    <br>\n",
    "    <li><b>Results</b></li><br>\n",
    "</ul>"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c4e040f2",
   "metadata": {},
   "source": [
    "# Dataset Preparation\n",
    "\n",
    "<p>Here, we use BAFDataset class to prepare the BAF dataset for our experiments, handling loading, train-test splitting, and preprocessing tasks in a clear and organized manner.</p>\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "f44294a8",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "environ({'SLURM_MPI_TYPE': 'pmi2', 'CONDA_SHLVL': '0', 'LD_LIBRARY_PATH': '/scratch/ssd001/pkgs/cudnn-11.4-v8.2.4.15/lib64:/scratch/ssd001/pkgs/cuda-11.6/targets/x86_64-linux/lib', 'CONDA_EXE': '/h/vkhazaie/anaconda3/bin/conda', 'SRUN_DEBUG': '3', 'SLURM_STEP_ID': '0', 'SLURM_STEP_GPUS': '2', 'SLURM_NODEID': '0', 'SLURM_TASK_PID': '4505', 'SSH_CONNECTION': '97.108.101.79 51253 172.17.15.236 22', 'SLURM_PRIO_PROCESS': '0', 'LANG': 'en_US.UTF-8', 'SLURM_SUBMIT_DIR': '/fs01/home/vkhazaie', 'PMI_SIZE': '1', 'OLDPWD': '/h/vkhazaie', 'SLURM_STEPID': '0', 'SLURM_SRUN_COMM_HOST': '172.17.15.236', 'SLURM_DISTRIBUTION': 'cyclic,pack', 'ROCR_VISIBLE_DEVICES': '0', 'SLURM_PROCID': '0', 'SLURM_JOB_GID': '11114', 'SLURMD_NODENAME': 'gpu097', 'VIRTUAL_ENV': '/ssd003/projects/aieng/public/anomaly_detection_environment', 'SLURM_TASKS_PER_NODE': '1', 'S_COLORS': 'auto', '_CE_M': '', 'XDG_SESSION_ID': '45822', 'SLURM_NNODES': '1', 'USER': 'vkhazaie', 'SLURM_LAUNCH_NODE_IPADDR': '172.17.15.236', 'SLURM_STEP_TASKS_PER_NODE': '1', 'PWD': '/h/vkhazaie/anomaly-detection-project', 'SLURM_JOB_NODELIST': 'gpu097', 'HOME': '/h/vkhazaie', 'PMI_FD': '11', 'SLURM_CLUSTER_NAME': 'vaughan2', 'CONDA_PYTHON_EXE': '/h/vkhazaie/anaconda3/bin/python', 'SLURM_NODELIST': 'gpu097', 'SLURM_GPUS_ON_NODE': '1', 'SSH_CLIENT': '97.108.101.79 51253 22', 'LMOD_VERSION': '8.2.4', 'SLURM_NTASKS': '1', 'SLURM_UMASK': '0002', 'SLURM_JOB_CPUS_PER_NODE': '2', 'BASH_ENV': '/opt/lmod/lmod/init/bash', 'SLURM_TOPOLOGY_ADDR': 'gpu097', '_CE_CONDA': '', 'SLURM_WORKING_CLUSTER': 'vaughan2:vremote2:6817:9728:109', 'SLURM_STEP_NODELIST': 'gpu097', 'SLURM_JOB_NAME': 'bash', 'SLURM_SRUN_COMM_PORT': '36779', 'TMPDIR': '/tmp', 'LMOD_sys': 'Linux', 'SLURM_JOBID': '10715756', 'SLURM_CONF': '/opt/slurm-22.05.3/etc/slurm.conf', 'SLURM_JOB_QOS': 'high', 'LMOD_ROOT': '/opt/lmod', 'SLURM_TOPOLOGY_ADDR_PATTERN': 'node', 'SSH_TTY': '/dev/pts/316', 'MAIL': '/var/mail/vkhazaie', 'ZE_AFFINITY_MASK': '0', 'SLURM_CPUS_ON_NODE': '2', 'SLURM_JOB_NUM_NODES': '1', 'SHELL': '/bin/bash', 'TERM': 'xterm-color', 'SLURM_JOB_UID': '11070', 'SLURM_JOB_PARTITION': 't4v2', 'SLURM_SCRIPT_CONTEXT': 'prolog_task', 'SLURM_PTY_WIN_ROW': '54', 'SLURM_JOB_USER': 'vkhazaie', 'CUDA_VISIBLE_DEVICES': '0', 'SLURM_PTY_WIN_COL': '209', 'PMI_JOBID': '10715756.0', 'SLURM_NPROCS': '1', 'SHLVL': '2', 'SLURM_SUBMIT_HOST': 'v3', 'SLURM_JOB_ACCOUNT': 'affiliate', 'MANPATH': '/opt/lmod/lmod/share/man:', 'SLURM_STEP_LAUNCHER_PORT': '36779', 'MODULEPATH': '/opt/modulefiles:/pkgs/modulefiles:/pkgs/environment-modules', 'SLURM_PTY_PORT': '41001', 'SLURM_GTIDS': '0', 'LOGNAME': 'vkhazaie', 'DBUS_SESSION_BUS_ADDRESS': 'unix:path=/run/user/11070/bus', 'XDG_RUNTIME_DIR': '/run/user/11070', 'MODULEPATH_ROOT': '/opt/modulefiles', 'PATH': '/ssd003/projects/aieng/public/anomaly_detection_environment/bin:/ssd003/projects/aieng/public/anomaly_detection_environment/bin:/pkgs/cuda-11.6/bin:/h/vkhazaie/anaconda3/condabin:/opt/slurm/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin', 'SLURM_JOB_ID': '10715756', 'PS1': '(anomaly_detection_environment) ${debian_chroot:+($debian_chroot)}\\\\u@\\\\h:\\\\w\\\\$ ', 'SLURM_STEP_NUM_TASKS': '1', 'MODULESHOME': '/opt/lmod/lmod', 'LMOD_SETTARG_FULL_SUPPORT': 'no', 'LMOD_PKG': '/opt/lmod/lmod', 'SLURM_STEP_NUM_NODES': '1', 'PMI_RANK': '0', 'LMOD_CMD': '/opt/lmod/lmod/libexec/lmod', 'SLURM_LOCALID': '0', 'GPU_DEVICE_ORDINAL': '0', 'LMOD_DIR': '/opt/lmod/lmod/libexec', 'BASH_FUNC_module%%': '() {  eval $($LMOD_CMD bash \"$@\") && eval $(${LMOD_SETTARG_CMD:-:} -s sh)\\n}', 'BASH_FUNC_ml%%': '() {  eval $($LMOD_DIR/ml_cmd \"$@\")\\n}', '_': '/ssd003/projects/aieng/public/anomaly_detection_environment/bin/jupyter', 'PYDEVD_USE_FRAME_EVAL': 'NO', 'JPY_PARENT_PID': '10964', 'CLICOLOR': '1', 'PAGER': 'cat', 'GIT_PAGER': 'cat', 'MPLBACKEND': 'module://matplotlib_inline.backend_inline', 'KMP_DUPLICATE_LIB_OK': 'True', 'KMP_INIT_AT_FORK': 'FALSE'})\n"
     ]
    }
   ],
   "source": [
    "# Import the necessary packages and libraries\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "from baf import BAFDataset\n",
    "from metrics import calculate_tpr_at_fpr, calculate_fairness_metrics\n",
    "import os\n",
    "import joblib\n",
    "import time\n",
    "import numpy as np\n",
    "import random\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.optim as optim\n",
    "from torch.utils.data import DataLoader, TensorDataset\n",
    "from sklearn.metrics import roc_curve\n",
    "from pytorch_tabnet.metrics import Metric\n",
    "from pytorch_tabnet.tab_model import TabNetClassifier\n",
    "from pytorch_tabnet.pretraining import TabNetPretrainer\n",
    "\n",
    "print(os.environ)\n",
    "seed = 100\n",
    "np.random.seed(seed)\n",
    "random.seed(seed)\n",
    "torch.manual_seed(seed)\n",
    "if torch.cuda.is_available():\n",
    "    torch.cuda.manual_seed_all(seed)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "065efc54",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Instantiate the BAF class\n",
    "ds = BAFDataset()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "21f6d243",
   "metadata": {},
   "outputs": [],
   "source": [
    "# set the path to the dataset and models\n",
    "\n",
    "base_path = \"/ssd003/projects/aieng/public/anomaly_detection_datasets/BAF\"\n",
    "model_path = \"/ssd003/projects/aieng/public/anomaly_detection_models/BAF\"\n",
    "\n",
    "variant = \"Base.csv\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "99e9556c",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Add file to path\n",
    "file_path = os.path.join(base_path, variant)\n",
    "\n",
    "# Load the dataset\n",
    "df = ds.load_data(file_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "a6682e22",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Split the dataset into train and test sets\n",
    "(X_train, y_train), (X_test, y_test) = ds.train_test_split(df, month=6)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "77b38c51",
   "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>income</th>\n",
       "      <th>name_email_similarity</th>\n",
       "      <th>prev_address_months_count</th>\n",
       "      <th>current_address_months_count</th>\n",
       "      <th>customer_age</th>\n",
       "      <th>days_since_request</th>\n",
       "      <th>intended_balcon_amount</th>\n",
       "      <th>payment_type</th>\n",
       "      <th>zip_count_4w</th>\n",
       "      <th>velocity_6h</th>\n",
       "      <th>...</th>\n",
       "      <th>phone_mobile_valid</th>\n",
       "      <th>bank_months_count</th>\n",
       "      <th>has_other_cards</th>\n",
       "      <th>proposed_credit_limit</th>\n",
       "      <th>foreign_request</th>\n",
       "      <th>source</th>\n",
       "      <th>session_length_in_minutes</th>\n",
       "      <th>device_os</th>\n",
       "      <th>keep_alive_session</th>\n",
       "      <th>device_distinct_emails_8w</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>96843</th>\n",
       "      <td>0.8</td>\n",
       "      <td>0.153411</td>\n",
       "      <td>-1</td>\n",
       "      <td>105</td>\n",
       "      <td>30</td>\n",
       "      <td>0.003467</td>\n",
       "      <td>-0.906840</td>\n",
       "      <td>AC</td>\n",
       "      <td>1779</td>\n",
       "      <td>1147.615990</td>\n",
       "      <td>...</td>\n",
       "      <td>1</td>\n",
       "      <td>-1</td>\n",
       "      <td>0</td>\n",
       "      <td>200.0</td>\n",
       "      <td>0</td>\n",
       "      <td>INTERNET</td>\n",
       "      <td>3.911696</td>\n",
       "      <td>linux</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>96844</th>\n",
       "      <td>0.8</td>\n",
       "      <td>0.442439</td>\n",
       "      <td>-1</td>\n",
       "      <td>107</td>\n",
       "      <td>40</td>\n",
       "      <td>0.021131</td>\n",
       "      <td>-1.553115</td>\n",
       "      <td>AC</td>\n",
       "      <td>1195</td>\n",
       "      <td>10788.642570</td>\n",
       "      <td>...</td>\n",
       "      <td>1</td>\n",
       "      <td>-1</td>\n",
       "      <td>0</td>\n",
       "      <td>200.0</td>\n",
       "      <td>0</td>\n",
       "      <td>INTERNET</td>\n",
       "      <td>23.696737</td>\n",
       "      <td>other</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>96845</th>\n",
       "      <td>0.9</td>\n",
       "      <td>0.328328</td>\n",
       "      <td>-1</td>\n",
       "      <td>79</td>\n",
       "      <td>70</td>\n",
       "      <td>0.009110</td>\n",
       "      <td>-0.801146</td>\n",
       "      <td>AC</td>\n",
       "      <td>845</td>\n",
       "      <td>8504.451753</td>\n",
       "      <td>...</td>\n",
       "      <td>1</td>\n",
       "      <td>-1</td>\n",
       "      <td>0</td>\n",
       "      <td>2000.0</td>\n",
       "      <td>0</td>\n",
       "      <td>INTERNET</td>\n",
       "      <td>2.803927</td>\n",
       "      <td>windows</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>96846</th>\n",
       "      <td>0.7</td>\n",
       "      <td>0.033834</td>\n",
       "      <td>-1</td>\n",
       "      <td>28</td>\n",
       "      <td>30</td>\n",
       "      <td>0.004707</td>\n",
       "      <td>-1.171706</td>\n",
       "      <td>AB</td>\n",
       "      <td>1697</td>\n",
       "      <td>6011.217859</td>\n",
       "      <td>...</td>\n",
       "      <td>1</td>\n",
       "      <td>28</td>\n",
       "      <td>0</td>\n",
       "      <td>1000.0</td>\n",
       "      <td>0</td>\n",
       "      <td>INTERNET</td>\n",
       "      <td>5.262441</td>\n",
       "      <td>windows</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>96847</th>\n",
       "      <td>0.9</td>\n",
       "      <td>0.150643</td>\n",
       "      <td>-1</td>\n",
       "      <td>164</td>\n",
       "      <td>40</td>\n",
       "      <td>0.012221</td>\n",
       "      <td>-0.227823</td>\n",
       "      <td>AB</td>\n",
       "      <td>1474</td>\n",
       "      <td>8378.324044</td>\n",
       "      <td>...</td>\n",
       "      <td>1</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>1500.0</td>\n",
       "      <td>0</td>\n",
       "      <td>INTERNET</td>\n",
       "      <td>5.728984</td>\n",
       "      <td>windows</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 29 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       income  name_email_similarity  prev_address_months_count  \\\n",
       "96843     0.8               0.153411                         -1   \n",
       "96844     0.8               0.442439                         -1   \n",
       "96845     0.9               0.328328                         -1   \n",
       "96846     0.7               0.033834                         -1   \n",
       "96847     0.9               0.150643                         -1   \n",
       "\n",
       "       current_address_months_count  customer_age  days_since_request  \\\n",
       "96843                           105            30            0.003467   \n",
       "96844                           107            40            0.021131   \n",
       "96845                            79            70            0.009110   \n",
       "96846                            28            30            0.004707   \n",
       "96847                           164            40            0.012221   \n",
       "\n",
       "       intended_balcon_amount payment_type  zip_count_4w   velocity_6h  ...  \\\n",
       "96843               -0.906840           AC          1779   1147.615990  ...   \n",
       "96844               -1.553115           AC          1195  10788.642570  ...   \n",
       "96845               -0.801146           AC           845   8504.451753  ...   \n",
       "96846               -1.171706           AB          1697   6011.217859  ...   \n",
       "96847               -0.227823           AB          1474   8378.324044  ...   \n",
       "\n",
       "       phone_mobile_valid  bank_months_count  has_other_cards  \\\n",
       "96843                   1                 -1                0   \n",
       "96844                   1                 -1                0   \n",
       "96845                   1                 -1                0   \n",
       "96846                   1                 28                0   \n",
       "96847                   1                  5                0   \n",
       "\n",
       "       proposed_credit_limit foreign_request    source  \\\n",
       "96843                  200.0               0  INTERNET   \n",
       "96844                  200.0               0  INTERNET   \n",
       "96845                 2000.0               0  INTERNET   \n",
       "96846                 1000.0               0  INTERNET   \n",
       "96847                 1500.0               0  INTERNET   \n",
       "\n",
       "       session_length_in_minutes device_os  keep_alive_session  \\\n",
       "96843                   3.911696     linux                   0   \n",
       "96844                  23.696737     other                   0   \n",
       "96845                   2.803927   windows                   0   \n",
       "96846                   5.262441   windows                   1   \n",
       "96847                   5.728984   windows                   0   \n",
       "\n",
       "       device_distinct_emails_8w  \n",
       "96843                          1  \n",
       "96844                          2  \n",
       "96845                          1  \n",
       "96846                          2  \n",
       "96847                          1  \n",
       "\n",
       "[5 rows x 29 columns]"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "e9f5f1db",
   "metadata": {},
   "outputs": [],
   "source": [
    "# One-hot encode the categorical features in the dataset\n",
    "X_train, X_test = ds.one_hot_encode_categorical(X_train, X_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "4116c3a2",
   "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>income</th>\n",
       "      <th>name_email_similarity</th>\n",
       "      <th>prev_address_months_count</th>\n",
       "      <th>current_address_months_count</th>\n",
       "      <th>customer_age</th>\n",
       "      <th>days_since_request</th>\n",
       "      <th>intended_balcon_amount</th>\n",
       "      <th>zip_count_4w</th>\n",
       "      <th>velocity_6h</th>\n",
       "      <th>velocity_24h</th>\n",
       "      <th>...</th>\n",
       "      <th>16</th>\n",
       "      <th>17</th>\n",
       "      <th>18</th>\n",
       "      <th>19</th>\n",
       "      <th>20</th>\n",
       "      <th>21</th>\n",
       "      <th>22</th>\n",
       "      <th>23</th>\n",
       "      <th>24</th>\n",
       "      <th>25</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>96843</th>\n",
       "      <td>0.8</td>\n",
       "      <td>0.153411</td>\n",
       "      <td>-1</td>\n",
       "      <td>105</td>\n",
       "      <td>30</td>\n",
       "      <td>0.003467</td>\n",
       "      <td>-0.906840</td>\n",
       "      <td>1779</td>\n",
       "      <td>1147.615990</td>\n",
       "      <td>3076.384487</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>96844</th>\n",
       "      <td>0.8</td>\n",
       "      <td>0.442439</td>\n",
       "      <td>-1</td>\n",
       "      <td>107</td>\n",
       "      <td>40</td>\n",
       "      <td>0.021131</td>\n",
       "      <td>-1.553115</td>\n",
       "      <td>1195</td>\n",
       "      <td>10788.642570</td>\n",
       "      <td>5790.128646</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>96845</th>\n",
       "      <td>0.9</td>\n",
       "      <td>0.328328</td>\n",
       "      <td>-1</td>\n",
       "      <td>79</td>\n",
       "      <td>70</td>\n",
       "      <td>0.009110</td>\n",
       "      <td>-0.801146</td>\n",
       "      <td>845</td>\n",
       "      <td>8504.451753</td>\n",
       "      <td>5912.163054</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>96846</th>\n",
       "      <td>0.7</td>\n",
       "      <td>0.033834</td>\n",
       "      <td>-1</td>\n",
       "      <td>28</td>\n",
       "      <td>30</td>\n",
       "      <td>0.004707</td>\n",
       "      <td>-1.171706</td>\n",
       "      <td>1697</td>\n",
       "      <td>6011.217859</td>\n",
       "      <td>5505.127811</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>96847</th>\n",
       "      <td>0.9</td>\n",
       "      <td>0.150643</td>\n",
       "      <td>-1</td>\n",
       "      <td>164</td>\n",
       "      <td>40</td>\n",
       "      <td>0.012221</td>\n",
       "      <td>-0.227823</td>\n",
       "      <td>1474</td>\n",
       "      <td>8378.324044</td>\n",
       "      <td>4930.672316</td>\n",
       "      <td>...</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 50 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       income  name_email_similarity  prev_address_months_count  \\\n",
       "96843     0.8               0.153411                         -1   \n",
       "96844     0.8               0.442439                         -1   \n",
       "96845     0.9               0.328328                         -1   \n",
       "96846     0.7               0.033834                         -1   \n",
       "96847     0.9               0.150643                         -1   \n",
       "\n",
       "       current_address_months_count  customer_age  days_since_request  \\\n",
       "96843                           105            30            0.003467   \n",
       "96844                           107            40            0.021131   \n",
       "96845                            79            70            0.009110   \n",
       "96846                            28            30            0.004707   \n",
       "96847                           164            40            0.012221   \n",
       "\n",
       "       intended_balcon_amount  zip_count_4w   velocity_6h  velocity_24h  ...  \\\n",
       "96843               -0.906840          1779   1147.615990   3076.384487  ...   \n",
       "96844               -1.553115          1195  10788.642570   5790.128646  ...   \n",
       "96845               -0.801146           845   8504.451753   5912.163054  ...   \n",
       "96846               -1.171706          1697   6011.217859   5505.127811  ...   \n",
       "96847               -0.227823          1474   8378.324044   4930.672316  ...   \n",
       "\n",
       "        16   17   18   19   20   21   22   23   24   25  \n",
       "96843  0.0  0.0  0.0  1.0  0.0  1.0  0.0  0.0  0.0  0.0  \n",
       "96844  0.0  0.0  0.0  1.0  0.0  0.0  0.0  1.0  0.0  0.0  \n",
       "96845  0.0  0.0  0.0  1.0  0.0  0.0  0.0  0.0  1.0  0.0  \n",
       "96846  0.0  0.0  0.0  1.0  0.0  0.0  0.0  0.0  1.0  0.0  \n",
       "96847  0.0  0.0  0.0  1.0  0.0  0.0  0.0  0.0  1.0  0.0  \n",
       "\n",
       "[5 rows x 50 columns]"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "b3586836",
   "metadata": {},
   "outputs": [],
   "source": [
    "# define these dictionaries for storing the results\n",
    "performance_results = {}\n",
    "fairness_results = {}\n",
    "\n",
    "# Dictionaries to store training and inference times\n",
    "training_times_dict = {}\n",
    "inference_times_dict = {}"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5eda7b1b",
   "metadata": {},
   "source": [
    "# Methods\n",
    "\n",
    "Now, let's delve into the methods employed to analyze the BAF dataset. The following section will outline the approaches and techniques used to address fraud detection in this high-stakes banking context."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f708c70b",
   "metadata": {},
   "source": [
    "## TabNet: Attentive Interpretable Tabular Learning\n",
    "\n",
    "**TabNet** presents an innovative approach that bridges the gap between two prominent machine learning paradigms: **Decision Trees (DTs)** and **Deep Neural Networks (DNNs)**. The methodology introduced in this work combines the strengths of these two approaches to create a powerful learning framework for real-world **tabular datasets**. While **DNNs lack the transparency and interpretability of DTs**, DTs often struggle to capture the intricacies of complex patterns present in diverse tabular data. TabNet's primary contribution revolves around its meticulously designed architecture, which effectively integrates the output structure of DTs within the flexible framework of DNNs.\n",
    "\n",
    "The central concept driving TabNet is the emulation of **decision boundaries** akin to those established by DTs. However, TabNet introduces a novel mechanism by leveraging a sequence of **learnable masks** to guide feature selection at each step. These masks, derived through the innovative **sparsemax normalization technique**, offer a dual advantage: they optimize parameter usage efficiently and promote interpretability by **highlighting the most pertinent features**. Moreover, the incorporation of **sparsity regularization** empowers the model with the ability to control the extent of feature selection, a crucial capability when dealing with datasets containing redundant or irrelevant features.\n",
    "\n",
    "TabNet's architecture excels due to its **sequential multi-step approach**, wherein decisions are iteratively refined by progressively considering the most influential features. Inspired by applications in visual processing and reinforcement learning, the **top-down attention mechanism** equips the model to concentrate on relevant information, effectively tackling the challenges posed by high-dimensional inputs. This results in a dynamic decision process that effectively harnesses the strengths of both local and global features, thereby enhancing TabNet's proficiency in capturing intricate relationships within tabular data.\n",
    "\n",
    "In essence, TabNet strikes a remarkable balance between **predictive power and interpretability**. Its capability to not only **outperform traditional DTs** but also provide **invaluable insights into feature importance** positions it as a potent solution for learning from tabular datasets. This becomes particularly relevant in domains where comprehending the decision-making process is just as pivotal as achieving accurate predictions.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "bf8e2c66",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/ssd003/projects/aieng/public/anomaly_detection_environment/lib/python3.7/site-packages/pytorch_tabnet/abstract_model.py:82: UserWarning: Device used : cuda\n",
      "  warnings.warn(f\"Device used : {self.device}\")\n"
     ]
    }
   ],
   "source": [
    "# Create an instance of TabNet model\n",
    "clf = TabNetClassifier()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "a9e1fb90",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create a custom metric class to pass to TabNet\n",
    "class TPR_at_05FPR(Metric):\n",
    "    def __init__(self):\n",
    "        self._name = \"tpr_at_05fpr\"\n",
    "        self._maximize = True\n",
    "\n",
    "    def __call__(self, y_true, y_score):\n",
    "        fpr, tpr, threshold = roc_curve(y_true, y_score[:, 1])\n",
    "        tpr_at_05fpr = tpr[fpr < 0.05][-1]\n",
    "        return tpr_at_05fpr"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "929c98b0",
   "metadata": {},
   "source": [
    "Set 1 for automatic weight balancing or try different values and see how the performance changes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "07d244a5",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/ssd003/projects/aieng/public/anomaly_detection_environment/lib/python3.7/site-packages/pytorch_tabnet/abstract_model.py:687: UserWarning: No early stopping will be performed, last training weights will be used.\n",
      "  warnings.warn(wrn_msg)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "epoch 0  | loss: 0.05198 |  0:00:59s\n",
      "epoch 1  | loss: 0.04631 |  0:01:58s\n",
      "epoch 2  | loss: 0.04549 |  0:02:56s\n",
      "epoch 3  | loss: 0.04523 |  0:03:55s\n",
      "epoch 4  | loss: 0.04475 |  0:04:54s\n",
      "Training Time: 629.1139430999756 seconds\n"
     ]
    }
   ],
   "source": [
    "# Start measuring training time\n",
    "start_time = time.time()\n",
    "\n",
    "# Train TabNet\n",
    "clf.fit(\n",
    "    X_train.values, \n",
    "    y_train.values,\n",
    "    eval_metric=[TPR_at_05FPR, \"auc\"],\n",
    "    max_epochs=5,\n",
    "    batch_size=512,\n",
    "    weights=0\n",
    ")\n",
    "\n",
    "# End measuring training time\n",
    "end_time = time.time()\n",
    "\n",
    "# Storing and printing the time\n",
    "tabnet_training_time = end_time - start_time\n",
    "print(f\"Training Time: {tabnet_training_time} seconds\")\n",
    "training_times_dict[(variant.split('.')[0], \"TabNet\")] = {\"Training Time\":tabnet_training_time}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "c7188f46",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Inference Time: 6.193806409835815 seconds\n"
     ]
    }
   ],
   "source": [
    "# Start measuring inference time\n",
    "start_time = time.time()\n",
    "\n",
    "# Predict on the test set\n",
    "tabnet_scored_test = clf.predict_proba(X_test.values)[:,1]\n",
    "\n",
    "# End measuring inference time\n",
    "end_time = time.time()\n",
    "\n",
    "# Storing and printing the time\n",
    "tabnet_inference_time = end_time - start_time\n",
    "print(f\"Inference Time: {tabnet_inference_time} seconds\")\n",
    "inference_times_dict[(variant.split('.')[0], \"TabNet\")] = {\"Inference Time\":tabnet_inference_time}"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6c210279",
   "metadata": {},
   "source": [
    "If you want to save the model, you can use:\n",
    "\n",
    "```python\n",
    "clf.save_model(\"PATH_TO_SAVE_MODEL\")\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "31d44222",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model TPR: 0.5069, Model FPR: 0.05, Model Threshold: 0.028200000524520874, Model AUROC: 0.8822\n"
     ]
    }
   ],
   "source": [
    "# Calculate the TPR@5%FPR on the test set\n",
    "metrics_dict = calculate_tpr_at_fpr(y_test, tabnet_scored_test, fpr_lim=0.05)\n",
    "performance_results[(variant.split('.')[0], 'TabNet')] = metrics_dict\n",
    "print(f\"Model TPR: {metrics_dict['TPR']}, Model FPR: {metrics_dict['FPR']}, Model Threshold: {metrics_dict['Threshold']}, Model AUROC: {metrics_dict['AUROC']}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "bd07e25c",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fairness Ratio: 0.3459 \n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Calculate the fairness metrics on the test set\n",
    "fairness_ratio = calculate_fairness_metrics(y_test, tabnet_scored_test, X_test, fpr_lim=0.05)\n",
    "fairness_results[(variant.split('.')[0], 'TabNet')] = {\"Fairness Ratio\":fairness_ratio}\n",
    "print(f\"Fairness Ratio: {fairness_ratio}\", \"\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "73a36e98",
   "metadata": {},
   "source": [
    "## Autoencoder\n",
    "\n",
    "Now we want to apply some unsupervised methods for anomaly detection. In this cell, we begin by importing essential libraries required for building and training neural networks, as well as for data preprocessing and evaluation. We also import Scikit-learn libraries for metrics calculation and data scaling.\n",
    "\n",
    "Following the imports, we define the architecture of the autoencoder model. An autoencoder is a type of neural network consisting of an encoder and a decoder. The purpose of an autoencoder is to learn a compressed representation of input data and then reconstruct the original data from this representation.\n",
    "\n",
    "We create the `Autoencoder` class as a subclass of `nn.Module`, which is PyTorch's base class for defining neural network modules. The class includes methods for initializing the architecture and performing forward passes.\n",
    "\n",
    "The encoder is structured using `nn.Sequential`, which is a convenient way to define a sequence of layers. It starts with a linear layer that reduces the input data's dimensionality to the specified `hidden_size`, followed by a rectified linear unit (ReLU) activation function to introduce non-linearity.\n",
    "\n",
    "The decoder, resembling the encoder but in reverse order, takes the encoded representation and aims to reconstruct the original input data. The simplicity of this architecture allows for easy experimentation with various configurations and layer sizes."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "df5a0741",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Define the Autoencoder architecture\n",
    "class Autoencoder(nn.Module):\n",
    "    def __init__(self, input_size, hidden_size):\n",
    "        super(Autoencoder, self).__init__()\n",
    "        self.encoder = nn.Sequential(\n",
    "            nn.Linear(input_size, hidden_size),\n",
    "            nn.ReLU()\n",
    "        )\n",
    "        self.decoder = nn.Sequential(\n",
    "            nn.Linear(hidden_size, input_size),\n",
    "            nn.ReLU()\n",
    "        )\n",
    "        \n",
    "    def forward(self, x):\n",
    "        encoded = self.encoder(x)\n",
    "        decoded = self.decoder(encoded)\n",
    "        return decoded"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "fc458c9a",
   "metadata": {},
   "source": [
    "In this cell, we set the hyperparameters that will govern the training process of the autoencoder. Hyperparameters are configuration choices that influence the behavior of the neural network during training. It's important to choose appropriate hyperparameters to ensure effective learning and convergence.\n",
    "\n",
    "We define the `input_size` as the number of features in the input data and `hidden_size` as the number of neurons in the hidden layer of the autoencoder. These values can be adjusted based on the complexity of your data and the desired capacity of the model.\n",
    "\n",
    "Other hyperparameters include the `batch_size`, which determines the number of data samples processed in each training iteration, the `learning_rate`, which controls the step size during optimization, and the `num_epochs`, which specifies the number of times the entire training dataset is passed through the model.\n",
    "\n",
    "With the hyperparameters set, we create an instance of the `Autoencoder` class that we defined earlier. This instance represents our neural network model, which will be trained to learn a compressed representation of the input data.\n",
    "\n",
    "We also define the loss function and optimizer. The loss function (`nn.MSELoss()`) calculates the mean squared error between the reconstructed data and the original data. The optimizer (`optim.Adam()`) implements the Adam optimization algorithm, which updates the model's parameters to minimize the loss during training.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "3bfbf494",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Hyperparameters\n",
    "input_size = 50  # Number of input features\n",
    "hidden_size = 30  # Number of neurons in the hidden layer\n",
    "batch_size = 512\n",
    "learning_rate = 0.0001\n",
    "num_epochs = 5\n",
    "\n",
    "# Create an instance of the Autoencoder\n",
    "autoencoder = Autoencoder(input_size, hidden_size)\n",
    "\n",
    "# Define loss function and optimizer\n",
    "criterion = nn.MSELoss()\n",
    "optimizer = optim.Adam(autoencoder.parameters(), lr=learning_rate)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "99ce53f1",
   "metadata": {},
   "source": [
    "In this cell, we convert the DataFrames to PyTorch tensors using the `torch.tensor` function. This conversion prepares the data for feeding into the neural network model.\n",
    "\n",
    "With the data in the form of PyTorch tensors, we are now ready to proceed with training the autoencoder model in subsequent cells."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "b9d94bb3",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Convert DataFrame to PyTorch tensors\n",
    "X_train_tensor = torch.tensor(X_train.values, dtype=torch.float32)\n",
    "X_test_tensor = torch.tensor(X_test.values, dtype=torch.float32)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "02b7bf10",
   "metadata": {},
   "source": [
    "In this cell, we begin the training process of the autoencoder neural network. Training a neural network involves iteratively adjusting its parameters to minimize a chosen loss function and improve its performance on the given data.\n",
    "\n",
    "We create a `train_dataset` using the scaled training data, and a `train_dataloader` using `TensorDataset` and `DataLoader` from PyTorch. The dataloader allows us to efficiently iterate over batches of data during training.\n",
    "\n",
    "We initiate a loop that runs for a specified number of epochs. In each epoch, the autoencoder processes batches of training data through the network and updates its parameters to reduce the loss. The loss is calculated using the mean squared error (`MSELoss`) between the input data and the reconstructed output.\n",
    "\n",
    "The `optimizer` then applies gradient descent to adjust the neural network's weights and biases, moving in the direction that minimizes the loss. This process iterates over multiple epochs, gradually improving the autoencoder's ability to capture useful features and create a compressed representation of the data.\n",
    "\n",
    "At the end of each epoch, we print the loss to monitor the training progress and observe how it decreases over time. This cell establishes the core training loop, and the subsequent cells will focus on evaluating the trained autoencoder's performance and analyzing its results.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "773262ad",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch [1/5], Loss: 109621.1641\n",
      "Epoch [2/5], Loss: 49789.3008\n",
      "Epoch [3/5], Loss: 15754.1221\n",
      "Epoch [4/5], Loss: 4542.7529\n",
      "Epoch [5/5], Loss: 708.6297\n",
      "Training Time: 43.945231914520264 seconds\n"
     ]
    }
   ],
   "source": [
    "# Create DataLoader for batching\n",
    "train_dataset = TensorDataset(X_train_tensor)\n",
    "train_dataloader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)\n",
    "\n",
    "\n",
    "# Start measuring training time\n",
    "start_time = time.time()\n",
    "\n",
    "# Training loop\n",
    "for epoch in range(num_epochs):\n",
    "    for batch in train_dataloader:\n",
    "        optimizer.zero_grad()\n",
    "        outputs = autoencoder(batch[0])\n",
    "        loss = criterion(outputs, batch[0])\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "    print(f\"Epoch [{epoch+1}/{num_epochs}], Loss: {loss.item():.4f}\")\n",
    "\n",
    "# End measuring training time\n",
    "end_time = time.time()\n",
    "\n",
    "# Storing and printing the time\n",
    "ae_training_time = end_time - start_time\n",
    "print(f\"Training Time: {ae_training_time} seconds\")\n",
    "training_times_dict[(variant.split('.')[0], \"AE\")] = {\"Training Time\":ae_training_time}"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "0f9cd662",
   "metadata": {},
   "source": [
    "In this cell, we evaluate the performance of the trained autoencoder for anomaly detection on the test data. Anomaly detection involves identifying instances that significantly deviate from the norm in the dataset.\n",
    "\n",
    "We pass the scaled test data through the trained autoencoder's decoder to obtain reconstructed data. The difference between the original test data and the reconstructed data is used to calculate the reconstruction errors. These errors represent how well the autoencoder is able to recreate the input data.\n",
    "\n",
    "To calculate the reconstruction errors, we use the mean squared error (`MSELoss`) with the reduction set to `'none'`. This means that the reconstruction error is computed for each individual data point without aggregating."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "ead12a5f",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Inference Time: 0.13280272483825684 seconds\n",
      "Model TPR: 0.0653, Model FPR: 0.0498, Model Threshold: 2304.104736328125, Model AUROC: 0.6174\n",
      "Fairness Ratio: 0.3826 \n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Start measuring inference time\n",
    "start_time = time.time()\n",
    "\n",
    "# Pass the test data through the autoencoder\n",
    "with torch.no_grad():\n",
    "    reconstructed_data = autoencoder(X_test_tensor)\n",
    "\n",
    "# End measuring inference time\n",
    "end_time = time.time()\n",
    "\n",
    "# Storing and printing the time\n",
    "ae_inference_time = end_time - start_time\n",
    "print(f\"Inference Time: {ae_inference_time} seconds\")\n",
    "inference_times_dict[(variant.split('.')[0], \"AE\")] = {\"Inference Time\":ae_inference_time}\n",
    "\n",
    "# Calculate reconstruction errors\n",
    "reconstruction_errors = ((reconstructed_data - X_test_tensor)**2).mean(dim=1)\n",
    "\n",
    "# Calculate the TPR@5%FPR on the test set\n",
    "metrics_dict = calculate_tpr_at_fpr(y_test.values, reconstruction_errors.numpy(), fpr_lim=0.05)\n",
    "performance_results[(variant.split('.')[0], 'AE')] = metrics_dict\n",
    "print(f\"Model TPR: {metrics_dict['TPR']}, Model FPR: {metrics_dict['FPR']}, Model Threshold: {metrics_dict['Threshold']}, Model AUROC: {metrics_dict['AUROC']}\")\n",
    "\n",
    "# Calculate the fairness metrics on the test set\n",
    "fairness_ratio = calculate_fairness_metrics(y_test, reconstruction_errors.numpy(), X_test, fpr_lim=0.05)\n",
    "fairness_results[(variant.split('.')[0], 'AE')] = {\"Fairness Ratio\":fairness_ratio}\n",
    "print(f\"Fairness Ratio: {fairness_ratio}\", \"\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "df967e0e",
   "metadata": {},
   "source": [
    "## DeepOD: Deep Learning-Based Anomaly Detection Framework\n",
    "\n",
    "DeepOD is an open-source Python framework designed for deep learning-based anomaly detection. It offers a unified implementation of various detection models based on PyTorch.\n",
    "\n",
    "DeepOD currently encompasses 13 deep anomaly detection algorithms in the unsupervised and weakly-supervised paradigms. The framework provides a platform for easy experimentation with these models, and additional baseline algorithms will be incorporated in the future.\n",
    "\n",
    "**Detection Models:**\n",
    "\n",
    "The following detection models are included in DeepOD:\n",
    "\n",
    "| Model       | Venue | Year | Type            | Title\n",
    "|-------------|-------|------|-----------------|----------------------------------------------------\n",
    "| Deep SVDD   | ICML  | 2018 | Unsupervised    | Deep One-Class Classification\n",
    "| REPEN       | KDD   | 2018 | Unsupervised    | Learning Representations of Ultrahigh-dimensional Data for Random Distance-based Outlier Detection\n",
    "| RDP         | IJCAI | 2020 | Unsupervised    | Unsupervised Representation Learning by Predicting Random Distances\n",
    "| RCA         | IJCAI | 2021 | Unsupervised    | RCA: A Deep Collaborative Autoencoder Approach for Anomaly Detection\n",
    "| GOAD        | ICLR  | 2020 | Unsupervised    | Classification-Based Anomaly Detection for General Data\n",
    "| NeuTraL     | ICML  | 2021 | Unsupervised    | Neural Transformation Learning for Deep Anomaly Detection Beyond Images\n",
    "| ICL         | ICLR  | 2022 | Unsupervised    | Anomaly Detection for Tabular Data with Internal Contrastive Learning\n",
    "| DIF         | TKDE  | 2023 | Unsupervised    | Deep Isolation Forest for Anomaly Detection\n",
    "| SLAD        | ICML  | 2023 | Unsupervised    | Fascinating Supervisory Signals and Where to Find Them: Deep Anomaly Detection with Scale Learning\n",
    "| DevNet      | KDD   | 2019 | Weakly-Supervised | Deep Anomaly Detection with Deviation Networks\n",
    "| PReNet      | KDD   | 2023 | Weakly-Supervised | Deep Weakly-supervised Anomaly Detection\n",
    "| Deep SAD    | ICLR  | 2020 | Weakly-Supervised | Deep Semi-Supervised Anomaly Detection\n",
    "| FeaWAD      | TNNLS | 2021 | Weakly-Supervised | Feature Encoding with AutoEncoders for Weakly-supervised Anomaly Detection\n",
    "\n",
    "**Usage Examples:**\n",
    "\n",
    "DeepOD is user-friendly and can be employed with only a few lines of code. The API style is similar to sklearn and [PyOD](https://github.com/yzhao062/pyod).\n",
    "\n",
    "```python\n",
    "# Unsupservised methods\n",
    "from deepod.models.dsvdd import DeepSVDD\n",
    "clf = DeepSVDD()\n",
    "clf.fit(X_train, y=None)\n",
    "scores = clf.decision_function(X_test)\n",
    "\n",
    "# Weakly-supervised methods\n",
    "from deepod.models.devnet import DevNet\n",
    "clf = DevNet()\n",
    "clf.fit(X_train, y=semi_y)  # semi_y uses 1 for known anomalies and 0 for unlabeled data\n",
    "scores = clf.decision_function(X_test)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "58a0ffb4",
   "metadata": {},
   "source": [
    "## ICL: Anomaly Detection for Tabular Data with Internal Contrastive Learning"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c2aef1a3",
   "metadata": {},
   "source": [
    "The **ICL method** focuses on **out-of-class sample detection** in **tabular data**, where the data structure is largely unknown. It aims to **capture the structure** of single training class samples by learning mappings that **maximize the mutual information** between each sample and the **masked-out portion**. This is accomplished through a **contrastive loss**, learning mappings individually for each sample. The score for a test sample is determined by measuring the **contrastive loss of the learned mappings applied to the masked parts** of the sample. ICL **outperforms existing methods** with a substantial accuracy advantage across benchmarks.\n",
    "\n",
    "\n",
    "**Contrastive learning**, rooted in metric learning and unsupervised representation learning, maximizes similarity between queries and positive samples while minimizing it with negative samples. Existing methods, often using image transformations, **don't apply to tabular data** due to the absence of invariant transformation groups. Qiu et al.'s NeuTraL AD work is a similar method for tabular data but **differs significantly**, utilizing specific masks and single feature extractors.\n",
    "\n",
    "\n",
    "ICL employs a training set of in-class samples and aims to design a **scoring function** that maps samples to low values if they belong to the same distribution and high values otherwise. Hyperparameters include **k for feature subset size, u for embedding size, and τ for loss temperature**. The method **constructs pairs from each sample and learns mappings to maximize mutual information**. Mutual information maximization is achieved through the **noise contrastive estimation framework**."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "d977a45b",
   "metadata": {},
   "outputs": [],
   "source": [
    "from deepod.models.icl import ICL"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "74afc862",
   "metadata": {},
   "outputs": [],
   "source": [
    "# unsupervised method\n",
    "clf = ICL(epochs=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "4cf132b2",
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Start Training...\n",
      "ensemble size: 2\n",
      "kernel size: 10\n",
      "ICLNet(\n",
      "  (enc_f_net): MLPnet(\n",
      "    (network): Sequential(\n",
      "      (0): LinearBlock(\n",
      "        (linear): Linear(in_features=40, out_features=100, bias=False)\n",
      "        (act_layer): Tanh()\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "      (1): LinearBlock(\n",
      "        (linear): Linear(in_features=100, out_features=50, bias=False)\n",
      "        (act_layer): LeakyReLU(negative_slope=0.01)\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "      (2): LinearBlock(\n",
      "        (linear): Linear(in_features=50, out_features=128, bias=False)\n",
      "        (act_layer): LeakyReLU(negative_slope=0.01)\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "  )\n",
      "  (enc_g_net): MLPnet(\n",
      "    (network): Sequential(\n",
      "      (0): LinearBlock(\n",
      "        (linear): Linear(in_features=10, out_features=50, bias=False)\n",
      "        (act_layer): LeakyReLU(negative_slope=0.01)\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "      (1): LinearBlock(\n",
      "        (linear): Linear(in_features=50, out_features=25, bias=False)\n",
      "        (act_layer): LeakyReLU(negative_slope=0.01)\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "      (2): LinearBlock(\n",
      "        (linear): Linear(in_features=25, out_features=128, bias=False)\n",
      "        (act_layer): LeakyReLU(negative_slope=0.01)\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "  )\n",
      ")\n",
      "epoch  1, training loss: 2.215789, time: 88.9s\n",
      "kernel size: 10\n",
      "ICLNet(\n",
      "  (enc_f_net): MLPnet(\n",
      "    (network): Sequential(\n",
      "      (0): LinearBlock(\n",
      "        (linear): Linear(in_features=40, out_features=100, bias=False)\n",
      "        (act_layer): Tanh()\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "      (1): LinearBlock(\n",
      "        (linear): Linear(in_features=100, out_features=50, bias=False)\n",
      "        (act_layer): LeakyReLU(negative_slope=0.01)\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "      (2): LinearBlock(\n",
      "        (linear): Linear(in_features=50, out_features=128, bias=False)\n",
      "        (act_layer): LeakyReLU(negative_slope=0.01)\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "  )\n",
      "  (enc_g_net): MLPnet(\n",
      "    (network): Sequential(\n",
      "      (0): LinearBlock(\n",
      "        (linear): Linear(in_features=10, out_features=50, bias=False)\n",
      "        (act_layer): LeakyReLU(negative_slope=0.01)\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "      (1): LinearBlock(\n",
      "        (linear): Linear(in_features=50, out_features=25, bias=False)\n",
      "        (act_layer): LeakyReLU(negative_slope=0.01)\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "      (2): LinearBlock(\n",
      "        (linear): Linear(in_features=25, out_features=128, bias=False)\n",
      "        (act_layer): LeakyReLU(negative_slope=0.01)\n",
      "        (bn_layer): BatchNorm1d(41, eps=1e-05, momentum=0.1, affine=False, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "  )\n",
      ")\n",
      "epoch  1, training loss: 2.135577, time: 87.1s\n",
      "Start Inference on the training data...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "testing: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 12422/12422 [00:38<00:00, 324.59it/s]\n",
      "testing: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 12422/12422 [00:37<00:00, 334.43it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training Time: 965.2103323936462 seconds\n"
     ]
    }
   ],
   "source": [
    "# Start measuring training time\n",
    "start_time = time.time()\n",
    "\n",
    "# Train ICL\n",
    "clf.fit(X_train.values, y=None)\n",
    "\n",
    "# End measuring training time\n",
    "end_time = time.time()\n",
    "\n",
    "# Storing and printing the time\n",
    "icl_training_time = end_time - start_time\n",
    "print(f\"Training Time: {icl_training_time} seconds\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "e305b153",
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "testing: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3204/3204 [00:09<00:00, 340.95it/s]\n",
      "testing: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3204/3204 [00:09<00:00, 342.52it/s]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Inference Time: 19.250097274780273 seconds\n"
     ]
    }
   ],
   "source": [
    "# Start measuring inference time\n",
    "start_time = time.time()\n",
    "\n",
    "# Predict on the test set\n",
    "icl_scored_test = clf.decision_function(X_test.values)\n",
    "\n",
    "# End measuring inference time\n",
    "end_time = time.time()\n",
    "\n",
    "# Storing and printing the time\n",
    "icl_inference_time = end_time - start_time\n",
    "print(f\"Inference Time: {icl_inference_time} seconds\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d5544169",
   "metadata": {},
   "source": [
    "If you want to save the model, you can use:\n",
    "\n",
    "```python\n",
    "torch.save(clf.net.state_dict(), \"PATH_TO_SAVE_MODEL\")\n",
    "```"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "2c8bff50",
   "metadata": {},
   "outputs": [],
   "source": [
    "training_times_dict[(variant.split('.')[0], \"ICL\")] = {\"Training Time\":icl_training_time}\n",
    "inference_times_dict[(variant.split('.')[0], \"ICL\")] = {\"Inference Time\":icl_inference_time}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "b02dbc2d",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model TPR: 0.0865, Model FPR: 0.0499, Model Threshold: 3.1268, Model AUROC: 0.6172\n"
     ]
    }
   ],
   "source": [
    "# Calculate the TPR@5%FPR on the test set\n",
    "metrics_dict = calculate_tpr_at_fpr(y_test, icl_scored_test, fpr_lim=0.05)\n",
    "performance_results[(variant.split('.')[0], 'ICL')] = metrics_dict\n",
    "print(f\"Model TPR: {metrics_dict['TPR']}, Model FPR: {metrics_dict['FPR']}, Model Threshold: {metrics_dict['Threshold']}, Model AUROC: {metrics_dict['AUROC']}\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "5352bc16",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Fairness Ratio: 0.9953 \n",
      "\n"
     ]
    }
   ],
   "source": [
    "# Calculate the fairness metrics on the test set\n",
    "fairness_ratio = calculate_fairness_metrics(y_test, icl_scored_test, X_test, fpr_lim=0.05)\n",
    "fairness_results[(variant.split('.')[0], 'ICL')] = {\"Fairness Ratio\":fairness_ratio}\n",
    "print(f\"Fairness Ratio: {fairness_ratio}\", \"\\n\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "3cf3796e",
   "metadata": {},
   "source": [
    "# Results\n",
    "\n",
    "In the forthcoming section, we will present the results obtained from our comprehensive analysis. We will showcase the performance metrics of the evaluated models on the BAF dataset, delving into their effectiveness in detecting fraudulent online bank account opening applications. Moreover, we will explore the fairness metrics, ensuring unbiased decision-making in the context of different user groups. To assess the real-world applicability of the models, we will provide insights into their training and inference times, offering valuable insights into their efficiency and suitability for practical use."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2458ac75",
   "metadata": {},
   "source": [
    "## Table\n",
    "\n",
    "Below you can see all the result in one table:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "08effeef",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create a DataFrame from all results\n",
    "df_performance_results = pd.DataFrame(performance_results).transpose()\n",
    "df_fairness_results = pd.DataFrame(fairness_results).transpose()\n",
    "df_training_times_dict = pd.DataFrame(training_times_dict).transpose()\n",
    "df_inference_times_dict = pd.DataFrame(inference_times_dict).transpose()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "1c5ad1a8",
   "metadata": {},
   "outputs": [],
   "source": [
    "# concatenate all results together\n",
    "all_results = pd.concat([df_performance_results, df_fairness_results, df_training_times_dict, df_inference_times_dict], axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "id": "96344431",
   "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></th>\n",
       "      <th>TPR</th>\n",
       "      <th>FPR</th>\n",
       "      <th>Threshold</th>\n",
       "      <th>AUROC</th>\n",
       "      <th>Fairness Ratio</th>\n",
       "      <th>Training Time</th>\n",
       "      <th>Inference Time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">Base</th>\n",
       "      <th>TabNet</th>\n",
       "      <td>0.5069</td>\n",
       "      <td>0.0500</td>\n",
       "      <td>0.028200</td>\n",
       "      <td>0.8822</td>\n",
       "      <td>0.3459</td>\n",
       "      <td>629.113943</td>\n",
       "      <td>6.193806</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>AE</th>\n",
       "      <td>0.0653</td>\n",
       "      <td>0.0498</td>\n",
       "      <td>2304.104736</td>\n",
       "      <td>0.6174</td>\n",
       "      <td>0.3826</td>\n",
       "      <td>43.945232</td>\n",
       "      <td>0.132803</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ICL</th>\n",
       "      <td>0.0865</td>\n",
       "      <td>0.0499</td>\n",
       "      <td>3.126800</td>\n",
       "      <td>0.6172</td>\n",
       "      <td>0.9953</td>\n",
       "      <td>965.210332</td>\n",
       "      <td>19.250097</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                TPR     FPR    Threshold   AUROC  Fairness Ratio  \\\n",
       "Base TabNet  0.5069  0.0500     0.028200  0.8822          0.3459   \n",
       "     AE      0.0653  0.0498  2304.104736  0.6174          0.3826   \n",
       "     ICL     0.0865  0.0499     3.126800  0.6172          0.9953   \n",
       "\n",
       "             Training Time  Inference Time  \n",
       "Base TabNet     629.113943        6.193806  \n",
       "     AE          43.945232        0.132803  \n",
       "     ICL        965.210332       19.250097  "
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_results"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "258e057a",
   "metadata": {},
   "source": [
    "## Plots\n",
    "\n",
    "The following is 4 plots demonstrating all results for all models:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "27fdb0db",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Extract TPR values\n",
    "tabnet_tpr = df_performance_results.loc[(variant.split('.')[0], 'TabNet'), 'TPR']\n",
    "ae_tpr = df_performance_results.loc[(variant.split('.')[0], 'AE'), 'TPR']\n",
    "icl_tpr = df_performance_results.loc[(variant.split('.')[0], 'ICL'), 'TPR']\n",
    "tpr_values = [tabnet_tpr, ae_tpr, icl_tpr]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "e3d13940",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Extract Fairness Ratio values\n",
    "tabnet_fn = df_fairness_results.loc[(variant.split('.')[0], 'TabNet'), 'Fairness Ratio']\n",
    "ae_fn = df_fairness_results.loc[(variant.split('.')[0], 'AE'), 'Fairness Ratio']\n",
    "icl_fn = df_fairness_results.loc[(variant.split('.')[0], 'ICL'), 'Fairness Ratio']\n",
    "fn_values = [tabnet_fn, ae_fn, icl_fn]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "id": "1a79ed11",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Extract Training Time values\n",
    "tabnet_tnt = df_training_times_dict.loc[(variant.split('.')[0], 'TabNet'), 'Training Time']\n",
    "ae_tnt = df_training_times_dict.loc[(variant.split('.')[0], 'AE'), 'Training Time']\n",
    "icl_tnt = df_training_times_dict.loc[(variant.split('.')[0], 'ICL'), 'Training Time']\n",
    "tnt_values = [tabnet_tnt, ae_tnt, icl_tnt]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "id": "29e6f720",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Extract Inference Time values\n",
    "tabnet_inf = df_inference_times_dict.loc[(variant.split('.')[0], 'TabNet'), 'Inference Time']\n",
    "ae_inf = df_inference_times_dict.loc[(variant.split('.')[0], 'AE'), 'Inference Time']\n",
    "icl_inf = df_inference_times_dict.loc[(variant.split('.')[0], 'ICL'), 'Inference Time']\n",
    "inf_values = [tabnet_inf, ae_inf, icl_inf]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "id": "9f69d9d2",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1200x1000 with 4 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Create a 2x2 grid of subplots\n",
    "fig, axs = plt.subplots(2, 2, figsize=(12, 10))\n",
    "\n",
    "models = ['TabNet', 'AE', 'ICL']\n",
    "\n",
    "# Plot TPR values\n",
    "axs[0, 0].bar(models, tpr_values)\n",
    "axs[0, 0].set_xlabel('Models')\n",
    "axs[0, 0].set_ylabel('TPR@5%FPR')\n",
    "axs[0, 0].set_title('Performance')\n",
    "\n",
    "# Plot Fairness Ratio values\n",
    "axs[0, 1].bar(models, fn_values)\n",
    "axs[0, 1].set_xlabel('Models')\n",
    "axs[0, 1].set_ylabel('Fairness Ratio')\n",
    "axs[0, 1].set_title('Fairness')\n",
    "\n",
    "# Plot Training Time values\n",
    "axs[1, 0].bar(models, tnt_values)\n",
    "axs[1, 0].set_xlabel('Models')\n",
    "axs[1, 0].set_ylabel('Training Time')\n",
    "axs[1, 0].set_title('Time')\n",
    "\n",
    "# Plot Inference Time values\n",
    "axs[1, 1].bar(models, inf_values)\n",
    "axs[1, 1].set_xlabel('Models')\n",
    "axs[1, 1].set_ylabel('Inference Time')\n",
    "axs[1, 1].set_title('Time')\n",
    "\n",
    "# Adjust layout\n",
    "plt.tight_layout()\n",
    "\n",
    "# Show the combined plot\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "id": "5d1ad3b6",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1000x600 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Define the list of model names and their corresponding y_scores\n",
    "model_names = ['TabNet', 'AE', 'ICL']\n",
    "y_scores_dict = {\n",
    "    'TabNet': tabnet_scored_test,\n",
    "    'AE': reconstruction_errors.numpy(),\n",
    "    'ICL': icl_scored_test,\n",
    "}\n",
    "\n",
    "# Generate FPR values from 5% to 40%\n",
    "fpr_values = np.logspace(np.log10(0.05), np.log10(0.4), num=100)\n",
    "\n",
    "# Initialize a dictionary to store interpolated TPR values for each model\n",
    "interpolated_tpr = {}\n",
    "\n",
    "# Calculate TPR values for each model at the desired FPR points\n",
    "for model_name in model_names:\n",
    "    scored_test = y_scores_dict[model_name]\n",
    "    tpr_list = [calculate_tpr_at_fpr(y_test, scored_test, fpr_lim=i)['TPR'] for i in fpr_values]\n",
    "    interpolated_tpr[model_name] = tpr_list\n",
    "\n",
    "# Plotting\n",
    "plt.figure(figsize=(10, 6))\n",
    "for model_name, model_tpr_interp in interpolated_tpr.items():\n",
    "    plt.plot(fpr_values, model_tpr_interp, label=model_name)\n",
    "\n",
    "plt.xscale('log')\n",
    "plt.xlabel('False Positive Rate (FPR)')\n",
    "plt.ylabel('True Positive Rate (TPR)')\n",
    "plt.title('TPR vs. FPR for Different Models')\n",
    "plt.legend()\n",
    "plt.grid(True)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "85ecfdc1",
   "metadata": {},
   "source": [
    "# References:\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "<ul>\n",
    "    <li>Jesus S, Pombal J, Alves D, Cruz A, Saleiro P, Ribeiro R, Gama J, Bizarro P. Turning the tables: Biased, imbalanced, dynamic tabular datasets for ml evaluation. Advances in Neural Information Processing Systems. 2022 Dec 6;35:33563-75.</li>\n",
    "\n",
    "</ul>"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "anomaly_detection_environment",
   "language": "python",
   "name": "anomaly_detection_environment"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.9"
  },
  "vscode": {
   "interpreter": {
    "hash": "10259be683146ba0cfb84d8d370984231ebebae69a5ceae05aa739bef317bea6"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}