10-imbalance2(focal-loss).ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## focal loss model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "dataset:\n",
    "    https://www.kaggle.com/ealaxi/paysim1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:39:12.706862Z",
     "start_time": "2019-10-06T04:39:06.018607Z"
    },
    "_cell_guid": "7b4a9f11-c3ee-3525-8c28-c32b1ac0c586"
   },
   "outputs": [],
   "source": [
    "import numpy as np # linear algebra\n",
    "import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n",
    "from tensorflow import keras\n",
    "from focal_loss import BinaryFocalLoss\n",
    "np.random.seed(42)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:39:23.348936Z",
     "start_time": "2019-10-06T04:39:12.708826Z"
    },
    "_cell_guid": "31bf5b85-e43e-2e03-7c86-a7ce42a978bf"
   },
   "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>step</th>\n",
       "      <th>amount</th>\n",
       "      <th>oldbalanceOrg</th>\n",
       "      <th>newbalanceOrig</th>\n",
       "      <th>oldbalanceDest</th>\n",
       "      <th>newbalanceDest</th>\n",
       "      <th>isFraud</th>\n",
       "      <th>isFlaggedFraud</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>9839.64</td>\n",
       "      <td>170136.0</td>\n",
       "      <td>160296.36</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>1864.28</td>\n",
       "      <td>21249.0</td>\n",
       "      <td>19384.72</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1</td>\n",
       "      <td>181.00</td>\n",
       "      <td>181.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1</td>\n",
       "      <td>181.00</td>\n",
       "      <td>181.0</td>\n",
       "      <td>0.00</td>\n",
       "      <td>21182.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1</td>\n",
       "      <td>11668.14</td>\n",
       "      <td>41554.0</td>\n",
       "      <td>29885.86</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   step    amount  oldbalanceOrg  newbalanceOrig  oldbalanceDest  \\\n",
       "0     1   9839.64       170136.0       160296.36             0.0   \n",
       "1     1   1864.28        21249.0        19384.72             0.0   \n",
       "2     1    181.00          181.0            0.00             0.0   \n",
       "3     1    181.00          181.0            0.00         21182.0   \n",
       "4     1  11668.14        41554.0        29885.86             0.0   \n",
       "\n",
       "   newbalanceDest  isFraud  isFlaggedFraud  \n",
       "0             0.0        0               0  \n",
       "1             0.0        0               0  \n",
       "2             0.0        1               0  \n",
       "3             0.0        1               0  \n",
       "4             0.0        0               0  "
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# create data frame containing your data, each column can be accessed # by df['column   name']\n",
    "dataset = pd.read_csv('PS_20174392719_1491204439457_log.csv')\n",
    "del dataset['nameDest']\n",
    "del dataset['nameOrig']\n",
    "del dataset['type']\n",
    "\n",
    "dataset.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:39:23.450547Z",
     "start_time": "2019-10-06T04:39:23.351116Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    6354407\n",
       "1       8213\n",
       "Name: isFraud, dtype: int64"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset['isFraud'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:39:23.458524Z",
     "start_time": "2019-10-06T04:39:23.452754Z"
    }
   },
   "outputs": [],
   "source": [
    "def feature_normalize(dataset):\n",
    "    mu = np.mean(dataset, axis=0)\n",
    "    sigma = np.std(dataset, axis=0)\n",
    "    return (dataset - mu) / sigma"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:43:38.878036Z",
     "start_time": "2019-10-06T04:43:35.849778Z"
    },
    "_cell_guid": "22768c2d-9bd9-3ad6-3cf7-cb124e755981"
   },
   "outputs": [],
   "source": [
    "#Splitting the Training/Test Data\n",
    "\n",
    "from sklearn.model_selection  import train_test_split\n",
    "X, y = dataset.iloc[:,:-2], dataset.iloc[:, -2]\n",
    "X = feature_normalize(X.values)\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:43:41.828035Z",
     "start_time": "2019-10-06T04:43:39.710942Z"
    },
    "_cell_guid": "c54a1358-8290-efad-7f3a-39df646850cf"
   },
   "outputs": [],
   "source": [
    "from tensorflow.keras.models import Sequential\n",
    "import tensorflow as tf\n",
    "\n",
    "model = Sequential()\n",
    "\n",
    "from tensorflow.keras.layers import Dense\n",
    "\n",
    "input_dim = X_train.shape[1]\n",
    "\n",
    "model.add(Dense(10, input_dim=input_dim, activation='relu', name='input'))\n",
    "model.add(Dense(20, activation='relu', name='fc1'))\n",
    "model.add(Dense(10, activation='relu', name='fc2'))\n",
    "model.add(Dense(1, activation='sigmoid', name='output'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:45:33.922514Z",
     "start_time": "2019-10-06T04:45:33.866979Z"
    }
   },
   "outputs": [],
   "source": [
    "model.compile(loss=BinaryFocalLoss(pos_weight=2, gamma=2),\n",
    "              optimizer='nadam',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:45:37.999387Z",
     "start_time": "2019-10-06T04:45:37.991663Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input (Dense)                (None, 10)                70        \n",
      "_________________________________________________________________\n",
      "fc1 (Dense)                  (None, 20)                220       \n",
      "_________________________________________________________________\n",
      "fc2 (Dense)                  (None, 10)                210       \n",
      "_________________________________________________________________\n",
      "output (Dense)               (None, 1)                 11        \n",
      "=================================================================\n",
      "Total params: 511\n",
      "Trainable params: 511\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:45:40.956649Z",
     "start_time": "2019-10-06T04:45:40.952562Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(5090096,)"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:47:01.148384Z",
     "start_time": "2019-10-06T04:45:41.678151Z"
    },
    "_cell_guid": "9698ed89-e4e9-33a3-abdc-cb44fa032f8d"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/3\n",
      "5091/5091 [==============================] - 10s 2ms/step - loss: 0.0017 - accuracy: 0.9993\n",
      "Epoch 2/3\n",
      "5091/5091 [==============================] - 9s 2ms/step - loss: 0.0017 - accuracy: 0.9993\n",
      "Epoch 3/3\n",
      "5091/5091 [==============================] - 8s 2ms/step - loss: 0.0016 - accuracy: 0.9994\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.keras.callbacks.History at 0x1bf8335a6c8>"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "model.fit(X_train, y_train, epochs=3, batch_size=1000)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:47:11.391877Z",
     "start_time": "2019-10-06T04:47:08.397639Z"
    },
    "_cell_guid": "ef09528a-24db-d9b1-5d05-85cac75d5d27"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1273/1273 [==============================] - 2s 1ms/step - loss: 0.0017 - accuracy: 0.9994\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[0.0016733587253838778, 0.9993595480918884]"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "score = model.evaluate(X_test, y_test, batch_size=1000)\n",
    "score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:47:39.212218Z",
     "start_time": "2019-10-06T04:47:36.035747Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 360x360 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "from sklearn import metrics\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "sns.set(font_scale=2)\n",
    "predictions = model.predict(X_test, batch_size=1000)\n",
    "\n",
    "LABELS = ['Normal','Fraud'] \n",
    "\n",
    "confusion_matrix = metrics.confusion_matrix(y_test.values, predictions.flatten()>0.5)\n",
    "plt.figure(figsize=(5, 5))\n",
    "sns.heatmap(confusion_matrix, xticklabels=LABELS, yticklabels=LABELS, annot=True, fmt=\"d\", annot_kws={\"size\": 20});\n",
    "plt.title(\"Confusion matrix\", fontsize=20)\n",
    "plt.ylabel('True label', fontsize=20)\n",
    "plt.xlabel('Predicted label', fontsize=20)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Total miss-classified labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-10-06T04:47:46.838135Z",
     "start_time": "2019-10-06T04:47:46.833144Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "815"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "values = confusion_matrix.view()\n",
    "error_count = values.sum() - np.trace(values)\n",
    "error_count"
   ]
  }
 ],
 "metadata": {
  "_change_revision": 158,
  "_is_fork": false,
  "kernel_info": {
   "name": "python3"
  },
  "kernelspec": {
   "display_name": "tf23",
   "language": "python",
   "name": "tf23"
  },
  "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"
  },
  "nteract": {
   "version": "0.15.0"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}