schoolPaper_creditCard_Analysis.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 3,
   "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>ID</th>\n",
       "      <th>LIMIT_BAL</th>\n",
       "      <th>SEX</th>\n",
       "      <th>EDUCATION</th>\n",
       "      <th>MARRIAGE</th>\n",
       "      <th>AGE</th>\n",
       "      <th>PAY_0</th>\n",
       "      <th>PAY_2</th>\n",
       "      <th>PAY_3</th>\n",
       "      <th>PAY_4</th>\n",
       "      <th>PAY_5</th>\n",
       "      <th>PAY_6</th>\n",
       "      <th>BILL_AMT1</th>\n",
       "      <th>BILL_AMT2</th>\n",
       "      <th>BILL_AMT3</th>\n",
       "      <th>BILL_AMT4</th>\n",
       "      <th>BILL_AMT5</th>\n",
       "      <th>BILL_AMT6</th>\n",
       "      <th>PAY_AMT1</th>\n",
       "      <th>PAY_AMT2</th>\n",
       "      <th>PAY_AMT3</th>\n",
       "      <th>PAY_AMT4</th>\n",
       "      <th>PAY_AMT5</th>\n",
       "      <th>PAY_AMT6</th>\n",
       "      <th>default.payment.next.month</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>20000.0</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>24</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>-1</td>\n",
       "      <td>-1</td>\n",
       "      <td>-2</td>\n",
       "      <td>-2</td>\n",
       "      <td>3913.0</td>\n",
       "      <td>3102.0</td>\n",
       "      <td>689.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>689.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>2</td>\n",
       "      <td>120000.0</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>26</td>\n",
       "      <td>-1</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>2682.0</td>\n",
       "      <td>1725.0</td>\n",
       "      <td>2682.0</td>\n",
       "      <td>3272.0</td>\n",
       "      <td>3455.0</td>\n",
       "      <td>3261.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1000.0</td>\n",
       "      <td>1000.0</td>\n",
       "      <td>1000.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>2000.0</td>\n",
       "      <td>1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>3</td>\n",
       "      <td>90000.0</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>34</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>29239.0</td>\n",
       "      <td>14027.0</td>\n",
       "      <td>13559.0</td>\n",
       "      <td>14331.0</td>\n",
       "      <td>14948.0</td>\n",
       "      <td>15549.0</td>\n",
       "      <td>1518.0</td>\n",
       "      <td>1500.0</td>\n",
       "      <td>1000.0</td>\n",
       "      <td>1000.0</td>\n",
       "      <td>1000.0</td>\n",
       "      <td>5000.0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>4</td>\n",
       "      <td>50000.0</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>37</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>46990.0</td>\n",
       "      <td>48233.0</td>\n",
       "      <td>49291.0</td>\n",
       "      <td>28314.0</td>\n",
       "      <td>28959.0</td>\n",
       "      <td>29547.0</td>\n",
       "      <td>2000.0</td>\n",
       "      <td>2019.0</td>\n",
       "      <td>1200.0</td>\n",
       "      <td>1100.0</td>\n",
       "      <td>1069.0</td>\n",
       "      <td>1000.0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>5</td>\n",
       "      <td>50000.0</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>57</td>\n",
       "      <td>-1</td>\n",
       "      <td>0</td>\n",
       "      <td>-1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>8617.0</td>\n",
       "      <td>5670.0</td>\n",
       "      <td>35835.0</td>\n",
       "      <td>20940.0</td>\n",
       "      <td>19146.0</td>\n",
       "      <td>19131.0</td>\n",
       "      <td>2000.0</td>\n",
       "      <td>36681.0</td>\n",
       "      <td>10000.0</td>\n",
       "      <td>9000.0</td>\n",
       "      <td>689.0</td>\n",
       "      <td>679.0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   ID  LIMIT_BAL  SEX  EDUCATION  MARRIAGE  AGE  PAY_0  PAY_2  PAY_3  PAY_4  PAY_5  PAY_6  \\\n",
       "0   1    20000.0    2          2         1   24      2      2     -1     -1     -2     -2   \n",
       "1   2   120000.0    2          2         2   26     -1      2      0      0      0      2   \n",
       "2   3    90000.0    2          2         2   34      0      0      0      0      0      0   \n",
       "3   4    50000.0    2          2         1   37      0      0      0      0      0      0   \n",
       "4   5    50000.0    1          2         1   57     -1      0     -1      0      0      0   \n",
       "\n",
       "   BILL_AMT1  BILL_AMT2  BILL_AMT3  BILL_AMT4  BILL_AMT5  BILL_AMT6  PAY_AMT1  PAY_AMT2  PAY_AMT3  \\\n",
       "0     3913.0     3102.0      689.0        0.0        0.0        0.0       0.0     689.0       0.0   \n",
       "1     2682.0     1725.0     2682.0     3272.0     3455.0     3261.0       0.0    1000.0    1000.0   \n",
       "2    29239.0    14027.0    13559.0    14331.0    14948.0    15549.0    1518.0    1500.0    1000.0   \n",
       "3    46990.0    48233.0    49291.0    28314.0    28959.0    29547.0    2000.0    2019.0    1200.0   \n",
       "4     8617.0     5670.0    35835.0    20940.0    19146.0    19131.0    2000.0   36681.0   10000.0   \n",
       "\n",
       "   PAY_AMT4  PAY_AMT5  PAY_AMT6  default.payment.next.month  \n",
       "0       0.0       0.0       0.0                           1  \n",
       "1    1000.0       0.0    2000.0                           1  \n",
       "2    1000.0    1000.0    5000.0                           0  \n",
       "3    1100.0    1069.0    1000.0                           0  \n",
       "4    9000.0     689.0     679.0                           0  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt # this is used for the plot the graph \n",
    "import seaborn as sns # used for plot interactive graph. \n",
    "from pandas import set_option\n",
    "plt.style.use('ggplot') # nice plots\n",
    "pd.set_option('display.width', 100)\n",
    "pd.set_option('display.max_columns', None)\n",
    "\n",
    "dataset = pd.read_csv('/Users/nic/Documents/学校论文/代码/datasets/UCI_Credit_Card.csv')\n",
    "dataset.head(5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x1164ecf60>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "X = dataset.drop(['default.payment.next.month'],axis=1)\n",
    "y = dataset['default.payment.next.month']\n",
    "X.corrwith(dataset['default.payment.next.month']).plot.bar(\n",
    "        figsize = (20, 10), title = \"Correlation with Default\", fontsize = 20,\n",
    "        rot = 90, grid = True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "X = X[['PAY_0', 'BILL_AMT1', 'PAY_AMT2']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "min_train = X_train.min()\n",
    "range_train = (X_train - min_train).max()\n",
    "X_train_scaled = (X_train - min_train)/range_train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "min_test = X_test.min()\n",
    "range_test = (X_test - min_test).max()\n",
    "X_test_scaled = (X_test - min_test)/range_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.44793272018432617"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.ensemble  import AdaBoostClassifier\n",
    "from time import time\n",
    "adaboost =AdaBoostClassifier()\n",
    "\n",
    "start = time()\n",
    "adaboost.fit(X_train_scaled, y_train)\n",
    "end = time()\n",
    "train_time_ada=end-start\n",
    "train_time_ada"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "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>Model</th>\n",
       "      <th>Accuracy</th>\n",
       "      <th>Precision</th>\n",
       "      <th>Recall</th>\n",
       "      <th>F1 Score</th>\n",
       "      <th>ROC</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>Adaboost</td>\n",
       "      <td>0.786333</td>\n",
       "      <td>0.741935</td>\n",
       "      <td>0.017733</td>\n",
       "      <td>0.034639</td>\n",
       "      <td>0.508016</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      Model  Accuracy  Precision    Recall  F1 Score       ROC\n",
       "0  Adaboost  0.786333   0.741935  0.017733  0.034639  0.508016"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from sklearn.metrics import  accuracy_score, f1_score, precision_score, recall_score, roc_auc_score\n",
    "y_pred = adaboost.predict(X_test_scaled)\n",
    "\n",
    "roc=roc_auc_score(y_test, y_pred)\n",
    "acc = accuracy_score(y_test, y_pred)\n",
    "prec = precision_score(y_test, y_pred)\n",
    "rec = recall_score(y_test, y_pred)\n",
    "f1 = f1_score(y_test, y_pred)\n",
    "\n",
    "results = pd.DataFrame([['Adaboost', acc,prec,rec, f1,roc]],\n",
    "               columns = ['Model', 'Accuracy', 'Precision', 'Recall', 'F1 Score','ROC'])\n",
    "results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.7838333333333334\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.78      1.00      0.88      4703\n",
      "           1       0.00      0.00      0.00      1297\n",
      "\n",
      "    accuracy                           0.78      6000\n",
      "   macro avg       0.39      0.50      0.44      6000\n",
      "weighted avg       0.61      0.78      0.69      6000\n",
      "\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Library/Python/3.7/site-packages/sklearn/metrics/_classification.py:1272: UndefinedMetricWarning: Precision and F-score are ill-defined and being set to 0.0 in labels with no predicted samples. Use `zero_division` parameter to control this behavior.\n",
      "  _warn_prf(average, modifier, msg_start, len(result))\n"
     ]
    }
   ],
   "source": [
    "from sklearn.linear_model import LogisticRegression\n",
    "from sklearn.model_selection import RandomizedSearchCV,cross_val_score\n",
    "from sklearn.metrics import classification_report\n",
    "\n",
    "param_grid = {'C': np.logspace(-5, 8, 15)}\n",
    "\n",
    "# Instantiate a logistic regression classifier\n",
    "logreg = LogisticRegression()\n",
    "\n",
    "# Instantiate the RandomizedSearchCV object\n",
    "logreg_cv = RandomizedSearchCV(logreg,param_grid , cv=5, random_state=0)\n",
    "\n",
    "# Fit it to the data\n",
    "logreg_cv.fit(X_train, y_train)\n",
    "\n",
    "LR = LogisticRegression(C=0.00005, random_state=0)\n",
    "LR.fit(X_train, y_train)\n",
    "y_pred = LR.predict(X_test)\n",
    "print('Accuracy:', accuracy_score(y_pred,y_test))\n",
    "\n",
    "## 5-fold cross-validation \n",
    "cv_scores =cross_val_score(LR, X, y, cv=5)\n",
    "\n",
    "# Print the 5-fold cross-validation scores\n",
    "print(classification_report(y_test, y_pred))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tuned Random Forest Parameters: {'n_estimators': 50, 'max_features': 3, 'max_depth': 2, 'criterion': 'gini'}\n"
     ]
    }
   ],
   "source": [
    "# Create the random grid\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "param_dist = {'n_estimators': [50,100,150,200,250],\n",
    "               \"max_features\": [1,2,3,4,5,6,7,8,9],\n",
    "               'max_depth': [1,2,3,4,5,6,7,8,9],\n",
    "               \"criterion\": [\"gini\", \"entropy\"]}\n",
    "\n",
    "rf = RandomForestClassifier()\n",
    "\n",
    "rf_cv = RandomizedSearchCV(rf, param_distributions = param_dist, \n",
    "                           cv = 5, random_state=0, n_jobs = -1)\n",
    "\n",
    "rf_cv.fit(X, y)\n",
    "\n",
    "print(\"Tuned Random Forest Parameters: %s\" % (rf_cv.best_params_))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 0.8286666666666667\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.84      0.97      0.90      4703\n",
      "           1       0.73      0.33      0.45      1297\n",
      "\n",
      "    accuracy                           0.83      6000\n",
      "   macro avg       0.79      0.65      0.67      6000\n",
      "weighted avg       0.82      0.83      0.80      6000\n",
      "\n",
      "Average 5-Fold CV Score: 0.8201 , Standard deviation: 0.0084\n"
     ]
    }
   ],
   "source": [
    "Ran = RandomForestClassifier(criterion= 'gini', max_depth= 2, \n",
    "                                     max_features= 3, n_estimators= 50, \n",
    "                                     random_state=0)\n",
    "Ran.fit(X_train, y_train)\n",
    "y_pred = Ran.predict(X_test)\n",
    "print('Accuracy:', accuracy_score(y_pred,y_test))\n",
    "\n",
    "## 5-fold cross-validation \n",
    "cv_scores =cross_val_score(Ran, X, y, cv=5)\n",
    "\n",
    "# Print the 5-fold cross-validation scores\n",
    "print(classification_report(y_test, y_pred))\n",
    "print(\"Average 5-Fold CV Score: {}\".format(round(np.mean(cv_scores),4)),\n",
    "      \", Standard deviation: {}\".format(round(np.std(cv_scores),4)))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(24000, 3)"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(30000, 25)"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dataset.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "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>PAY_0</th>\n",
       "      <th>BILL_AMT1</th>\n",
       "      <th>PAY_AMT2</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>3225</th>\n",
       "      <td>0.2</td>\n",
       "      <td>0.161646</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11815</th>\n",
       "      <td>0.1</td>\n",
       "      <td>0.146665</td>\n",
       "      <td>0.000163</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7338</th>\n",
       "      <td>0.4</td>\n",
       "      <td>0.164450</td>\n",
       "      <td>0.000891</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14980</th>\n",
       "      <td>0.4</td>\n",
       "      <td>0.171494</td>\n",
       "      <td>0.000089</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27167</th>\n",
       "      <td>0.2</td>\n",
       "      <td>0.153842</td>\n",
       "      <td>0.001187</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       PAY_0  BILL_AMT1  PAY_AMT2\n",
       "3225     0.2   0.161646  0.000000\n",
       "11815    0.1   0.146665  0.000163\n",
       "7338     0.4   0.164450  0.000891\n",
       "14980    0.4   0.171494  0.000089\n",
       "27167    0.2   0.153842  0.001187"
      ]
     },
     "execution_count": 61,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(24000, 3) (24000,) (6000, 3) (6000,)\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "%matplotlib inline\n",
    "from keras.layers import Input, Dense, LSTM\n",
    "from keras.models import Model\n",
    "\n",
    "from keras.layers import *\n",
    "from keras.models import *\n",
    "from keras.optimizers import Adam\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(24000, 6, 3) (24000,) (6000, 6, 3) (6000,)\n"
     ]
    }
   ],
   "source": [
    "from sklearn.preprocessing import MinMaxScaler\n",
    "scaler = MinMaxScaler(feature_range=(-1, 1))\n",
    "\n",
    "output_dim = 1\n",
    "batch_size = 256 #每轮训练模型时，样本的数量\n",
    "epochs = 60 #训练60轮次\n",
    "seq_len = 6 # 每3个为一组时间窗口\n",
    "hidden_size = 128\n",
    "\n",
    "TIME_STEPS = 6\n",
    "INPUT_DIM = 3\n",
    "\n",
    "lstm_units = 64\n",
    "dataset_flow = dataset[['PAY_0','PAY_2','PAY_3','PAY_4','PAY_5','PAY_6',\n",
    "                        'BILL_AMT1','BILL_AMT2','BILL_AMT3','BILL_AMT4','BILL_AMT5','BILL_AMT6',\n",
    "                        'PAY_AMT1','PAY_AMT2','PAY_AMT3','PAY_AMT4','PAY_AMT5','PAY_AMT6']]\n",
    "scaler.fit(dataset_flow)\n",
    "X_train, X_test, y_train, y_test = train_test_split(dataset_flow, y, test_size = 0.2, random_state = 1)\n",
    "# 归一化\n",
    "X_train = scaler.transform(X_train)\n",
    "X_test = scaler.transform(X_test)\n",
    "train_num = X_train.shape[0]\n",
    "test_num = X_test.shape[0]\n",
    "X_train = X_train.reshape((train_num, seq_len, 3))\n",
    "y_train = y_train.values\n",
    "X_test = X_test.reshape((test_num, seq_len, 3))\n",
    "y_test = y_test.values\n",
    "# X_train = np.array([data_train[i : i + seq_len, :] for i in range(data_train.shape[0] - seq_len)])\n",
    "# y_train = np.array([data_train[i + seq_len, 0] for i in range(data_train.shape[0]- seq_len)])\n",
    "# X_test = np.array([data_test[i : i + seq_len, :] for i in range(data_test.shape[0]- seq_len)])\n",
    "# y_test = np.array([data_test[i + seq_len, 0] for i in range(data_test.shape[0] - seq_len)])\n",
    "\n",
    "print(X_train.shape, y_train.shape, X_test.shape, y_test.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(None, 1, 64)\n"
     ]
    }
   ],
   "source": [
    "inputs = Input(shape=(TIME_STEPS, INPUT_DIM))\n",
    "#drop1 = Dropout(0.3)(inputs)\n",
    "\n",
    "x = Conv1D(filters = 64, kernel_size = 1, activation = 'relu')(inputs)  #, padding = 'same'\n",
    "#x = Conv1D(filters=128, kernel_size=5, activation='relu')(output1)#embedded_sequences\n",
    "x = MaxPooling1D(pool_size = 5)(x)\n",
    "x = Dropout(0.2)(x)\n",
    "print(x.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(None, 128)\n"
     ]
    }
   ],
   "source": [
    "lstm_out = Bidirectional(LSTM(lstm_units, activation='relu'), name='bilstm')(x)\n",
    "#lstm_out = LSTM(lstm_units,activation='relu')(x)\n",
    "print(lstm_out.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 116,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras import backend as K\n",
    "from keras.engine.topology import Layer\n",
    "import numpy as np\n",
    "from keras import initializers\n",
    "# Attention GRU network  未用     \n",
    "class AttLayer(Layer):\n",
    "    def __init__(self, **kwargs):\n",
    "        self.init = initializers.get('normal')\n",
    "        #self.input_spec = [InputSpec(ndim=3)]\n",
    "        super(AttLayer, self).__init__(**kwargs)\n",
    "\n",
    "    def build(self, input_shape):\n",
    "        assert len(input_shape)==128\n",
    "        #self.W = self.init((input_shape[-1],1))\n",
    "        self.W = self.init((input_shape[-1],))\n",
    "        #self.input_spec = [InputSpec(shape=input_shape)]\n",
    "        self.trainable_weights = [self.W]\n",
    "        super(AttLayer, self).build(input_shape)  # be sure you call this somewhere!\n",
    "\n",
    "    def call(self, x, mask=None):\n",
    "        eij = K.tanh(K.dot(x, self.W))\n",
    "        \n",
    "        ai = K.exp(eij)\n",
    "        weights = ai/K.sum(ai, axis=1).dimshuffle(0,'x')\n",
    "        \n",
    "        weighted_input = x*weights.dimshuffle(0,1,'x')\n",
    "        return weighted_input.sum(axis=1)\n",
    "\n",
    "    def get_output_shape_for(self, input_shape):\n",
    "        return (input_shape[0], input_shape[-1])\n",
    "    \n",
    "# l_att = AttLayer()(lstm_out)\n",
    "# output = Dense(1, activation='sigmoid')(l_att)\n",
    "# print(output.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.layers import Input, Dense, merge\n",
    "from keras import layers\n",
    "# ATTENTION PART STARTS HERE\n",
    "attention_probs = Dense(128, activation='sigmoid', name='attention_vec')(lstm_out)\n",
    "#attention_mul=layers.merge([stm_out,attention_probs], output_shape],mode='concat',concat_axis=1))\n",
    "attention_mul =Multiply()([lstm_out, attention_probs])\n",
    "#attention_mul = merge([lstm_out, attention_probs],output_shape=32, name='attention_mul', mode='mul')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_4\"\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_2 (InputLayer)            (None, 6, 3)         0                                            \n",
      "__________________________________________________________________________________________________\n",
      "conv1d_2 (Conv1D)               (None, 6, 64)        256         input_2[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "max_pooling1d_2 (MaxPooling1D)  (None, 1, 64)        0           conv1d_2[0][0]                   \n",
      "__________________________________________________________________________________________________\n",
      "dropout_2 (Dropout)             (None, 1, 64)        0           max_pooling1d_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "bilstm (Bidirectional)          (None, 128)          66048       dropout_2[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "attention_vec (Dense)           (None, 128)          16512       bilstm[0][0]                     \n",
      "__________________________________________________________________________________________________\n",
      "multiply_2 (Multiply)           (None, 128)          0           bilstm[0][0]                     \n",
      "                                                                 attention_vec[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dense_4 (Dense)                 (None, 1)            129         multiply_2[0][0]                 \n",
      "==================================================================================================\n",
      "Total params: 82,945\n",
      "Trainable params: 82,945\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "output = Dense(1, activation='relu')(attention_mul)\n",
    "#output = Dense(10, activation='sigmoid')(drop2)\n",
    "\n",
    "model = Model(inputs=inputs, outputs=output)\n",
    "print(model.summary())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 145,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/60\n",
      "24000/24000 [==============================] - 1s 55us/step - loss: 0.4425\n",
      "Epoch 2/60\n",
      "24000/24000 [==============================] - 0s 16us/step - loss: 0.4423\n",
      "Epoch 3/60\n",
      "24000/24000 [==============================] - 0s 15us/step - loss: 0.4425\n",
      "Epoch 4/60\n",
      "24000/24000 [==============================] - 0s 15us/step - loss: 0.4432\n",
      "Epoch 5/60\n",
      "24000/24000 [==============================] - 0s 16us/step - loss: 0.4420\n",
      "Epoch 6/60\n",
      "24000/24000 [==============================] - 0s 15us/step - loss: 0.4425\n",
      "Epoch 7/60\n",
      "24000/24000 [==============================] - 0s 15us/step - loss: 0.4428\n",
      "Epoch 8/60\n",
      "24000/24000 [==============================] - 0s 15us/step - loss: 0.4427\n",
      "Epoch 9/60\n",
      "24000/24000 [==============================] - 0s 17us/step - loss: 0.4425\n",
      "Epoch 10/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4423\n",
      "Epoch 11/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4422\n",
      "Epoch 12/60\n",
      "24000/24000 [==============================] - 0s 16us/step - loss: 0.4415\n",
      "Epoch 13/60\n",
      "24000/24000 [==============================] - 0s 15us/step - loss: 0.4422\n",
      "Epoch 14/60\n",
      "24000/24000 [==============================] - 0s 16us/step - loss: 0.4431\n",
      "Epoch 15/60\n",
      "24000/24000 [==============================] - 0s 16us/step - loss: 0.4420\n",
      "Epoch 16/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4426\n",
      "Epoch 17/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4478\n",
      "Epoch 18/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4444\n",
      "Epoch 19/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4430\n",
      "Epoch 20/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4434\n",
      "Epoch 21/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4432\n",
      "Epoch 22/60\n",
      "24000/24000 [==============================] - 1s 23us/step - loss: 0.4425\n",
      "Epoch 23/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4428\n",
      "Epoch 24/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4427\n",
      "Epoch 25/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4432\n",
      "Epoch 26/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4426\n",
      "Epoch 27/60\n",
      "24000/24000 [==============================] - 1s 21us/step - loss: 0.4421\n",
      "Epoch 28/60\n",
      "24000/24000 [==============================] - 1s 21us/step - loss: 0.4423\n",
      "Epoch 29/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4422\n",
      "Epoch 30/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4430\n",
      "Epoch 31/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4446\n",
      "Epoch 32/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4436\n",
      "Epoch 33/60\n",
      "24000/24000 [==============================] - 0s 21us/step - loss: 0.4440\n",
      "Epoch 34/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4425\n",
      "Epoch 35/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4436\n",
      "Epoch 36/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4428\n",
      "Epoch 37/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4424\n",
      "Epoch 38/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4425\n",
      "Epoch 39/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4433\n",
      "Epoch 40/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4419\n",
      "Epoch 41/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4425\n",
      "Epoch 42/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4422\n",
      "Epoch 43/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4456\n",
      "Epoch 44/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4452\n",
      "Epoch 45/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4441\n",
      "Epoch 46/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4427\n",
      "Epoch 47/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4440\n",
      "Epoch 48/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4446\n",
      "Epoch 49/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4432\n",
      "Epoch 50/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4425\n",
      "Epoch 51/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4430\n",
      "Epoch 52/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4427\n",
      "Epoch 53/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4429\n",
      "Epoch 54/60\n",
      "24000/24000 [==============================] - 0s 18us/step - loss: 0.4422\n",
      "Epoch 55/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4422\n",
      "Epoch 56/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4421\n",
      "Epoch 57/60\n",
      "24000/24000 [==============================] - 0s 20us/step - loss: 0.4419\n",
      "Epoch 58/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4430\n",
      "Epoch 59/60\n",
      "24000/24000 [==============================] - 1s 22us/step - loss: 0.4426\n",
      "Epoch 60/60\n",
      "24000/24000 [==============================] - 0s 19us/step - loss: 0.4420\n",
      "24000/24000 [==============================] - 0s 12us/step\n",
      "Train loss: 0.43942011992136637\n",
      "6000/6000 [==============================] - 0s 5us/step\n",
      "Test loss: 0.46350643269220987\n",
      "ROC-AUC: 0.7382044969300325\n"
     ]
    }
   ],
   "source": [
    "\n",
    "from keras import optimizers\n",
    "sgd = optimizers.SGD(lr=0.005, decay=1e-6, momentum=0.9, nesterov=True)\n",
    "model.compile(loss='binary_crossentropy', optimizer='adam')\n",
    "model.fit(X_train, y_train, epochs=60, batch_size=512, shuffle=False)\n",
    "y_pred = model.predict(X_test)\n",
    "print('Train loss:', model.evaluate(X_train, y_train, batch_size=batch_size))\n",
    "print('Test loss:', model.evaluate(X_test, y_test, batch_size=batch_size))\n",
    "print('ROC-AUC:', roc_auc_score(y_test, y_pred))\n",
    "# plt.plot(y_test, label='test')\n",
    "# plt.plot(y_pred, label='pred')\n",
    "# plt.legend()\n",
    "# plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 154,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0 1 0 ... 0 0 0] \n",
      " [[0.08065483]\n",
      " [0.27119496]\n",
      " [0.1429239 ]\n",
      " ...\n",
      " [0.11683801]\n",
      " [0.1557403 ]\n",
      " [0.12776509]]\n",
      "5985\n",
      "              precision    recall  f1-score   support\n",
      "\n",
      "           0       0.84      0.92      0.88      4663\n",
      "           1       0.59      0.41      0.48      1337\n",
      "\n",
      "    accuracy                           0.80      6000\n",
      "   macro avg       0.72      0.66      0.68      6000\n",
      "weighted avg       0.79      0.80      0.79      6000\n",
      "\n"
     ]
    }
   ],
   "source": [
    "y_pred_flatten = y_pred.flatten()\n",
    "print(y_test, '\\n',y_pred)\n",
    "print(len([y_p for y_p in y_pred if y_p!=0]))\n",
    "y_pred_new = [0 if y_p<0.45 else 1 for y_p in y_pred_flatten]\n",
    "print(classification_report(y_test, y_pred_new))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 155,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.8041666666666667\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import accuracy_score\n",
    "accuracy = accuracy_score(y_test, y_pred_new)\n",
    "print(accuracy)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}