【金融风控系列】_[3]_贷款违约识别

本文围绕Kaggle的Home Credit Default Risk赛题展开，利用客户申请表等7张表数据构建模型预测客户还款能力。通过数据清洗、特征工程，融合多表信息生成衍生特征，经LightGBM模型训练，最终线上评分为0.78277，为信用记录不足人群的贷款评估提供参考。

Home Credit Default Risk（家庭信用违约风险）

该赛题来自 KAGGLE，仅用作学习交流

---

由于信用记录不足或不存在，许多人往往被划分为低信用借贷人而难以获得贷款。 为了确保这些人群获得贷款，Home Credit公司利用替代数据（包括电信和交易信息）预测客户的还款能力。

Home Credit一共提供了7张表，一共218个字段，其中训练集样本约31万(逾期8%)，测试集样本约5万。

---

信息表

application_train/test 客户申请表

包含了

目标变量(客户是否违约-0/1变量)客户申请贷款信息(贷款类型, 贷款总额, 年金)客户基本信息(性别, 年龄, 家庭, 学历, 职业, 行业, 居住地情况)客户财务信息(年收入, 房/车情况)申请时提供的资料等.

bureau/bureau_balance 由其他金融机构提供给征信中心的客户信用记录历史(月数据)

包含了客户在征信中心的

信用记录,违约金额,违约时间等.

以时间序列(按行)的形式进行记录.

POS_CASH_balance 客户在Home Credit数据库中POS(point of sales)和现金贷款历史(月数据)

包含了客户

已付款情况未付款情况

credit_card_balance 客户在Home Credit数据库中信用卡的snapshot历史(月数据)

包含了客户

消费次数消费金额

等情况.

previous_application 客户先前的申请记录

包含了客户所有历史申请记录(申请信息, 申请结果等).

installments_payments 客户先前信用卡的还款记录

包含了客户的还款情况

还款日期是否逾期还款金额是否欠款等

---

参考：

[1] https://zhuanlan.zhihu.com/p/43541825

[2] https://www.kaggle.com/xucheng/cv-7993-private-score-7996/

[3] https://zhuanlan.zhihu.com/p/40790434

[4] https://www.kaggle.com/tahmidnafi/cse499

[5] https://blog.csdn.net/zhangchen2449/article/details/83338978

主要字段表

In [20]

#!unzip -q -o data/data105246/home_credit_default_risk.zip -d /home/aistudio/data

登录后复制        

unzip:  cannot find or open data/data104475/IEEE_CIS_Fraud_Detection.zip, data/data104475/IEEE_CIS_Fraud_Detection.zip.zip or data/data104475/IEEE_CIS_Fraud_Detection.zip.ZIP.

登录后复制        In [22]

# 安装依赖包!pip install xgboost!pip install lightgbm

登录后复制        

Looking in indexes: https://mirror.baidu.com/pypi/simple/Requirement already satisfied: xgboost in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (1.3.3)Requirement already satisfied: scipy in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from xgboost) (1.6.3)Requirement already satisfied: numpy in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from xgboost) (1.20.3)Looking in indexes: https://mirror.baidu.com/pypi/simple/Requirement already satisfied: lightgbm in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (3.1.1)Requirement already satisfied: scikit-learn!=0.22.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from lightgbm) (0.24.2)Requirement already satisfied: numpy in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from lightgbm) (1.20.3)Requirement already satisfied: scipy in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from lightgbm) (1.6.3)Requirement already satisfied: wheel in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from lightgbm) (0.36.2)Requirement already satisfied: threadpoolctl>=2.0.0 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from scikit-learn!=0.22.0->lightgbm) (2.1.0)Requirement already satisfied: joblib>=0.11 in /opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages (from scikit-learn!=0.22.0->lightgbm) (0.14.1)

登录后复制        In [23]

import osimport gcimport numpy as npimport pandas as pdfrom scipy.stats import kurtosisfrom sklearn.metrics import roc_auc_scorefrom sklearn.preprocessing import MinMaxScalerfrom sklearn.impute import SimpleImputerfrom sklearn.linear_model import LogisticRegressionimport matplotlib.pyplot as pltimport seaborn as snsimport warningsfrom sklearn.model_selection import train_test_split, cross_val_score, StratifiedKFoldimport xgboost as xgbfrom xgboost import XGBClassifierwarnings.simplefilter(action='ignore', category=FutureWarning)from lightgbm import LGBMClassifier

登录后复制    In [24]

DATA_DIRECTORY = "./data"df_train = pd.read_csv(os.path.join(DATA_DIRECTORY, 'application_train.csv'))df_test = pd.read_csv(os.path.join(DATA_DIRECTORY, 'application_test.csv'))df = df_train.append(df_test)del df_train, df_test; gc.collect()

登录后复制        

39

登录后复制登录后复制登录后复制                In [25]

df = df[df['AMT_INCOME_TOTAL'] < 20000000]df = df[df['CODE_GENDER'] != 'XNA']df['DAYS_EMPLOYED'].replace(365243, np.nan, inplace=True)df['DAYS_LAST_PHONE_CHANGE'].replace(0, np.nan, inplace=True)

登录后复制    In [26]

def get_age_group(days_birth):    age_years = -days_birth / 365    if age_years < 27: return 1    elif age_years < 40: return 2    elif age_years < 50: return 3    elif age_years < 65: return 4    elif age_years < 99: return 5    else: return 0

登录后复制    In [27]

docs = [f for f in df.columns if 'FLAG_DOC' in f]df['DOCUMENT_COUNT'] = df[docs].sum(axis=1)df['NEW_DOC_KURT'] = df[docs].kurtosis(axis=1)df['AGE_RANGE'] = df['DAYS_BIRTH'].apply(lambda x: get_age_group(x))

登录后复制    In [28]

df['EXT_SOURCES_PROD'] = df['EXT_SOURCE_1'] * df['EXT_SOURCE_2'] * df['EXT_SOURCE_3']df['EXT_SOURCES_WEIGHTED'] = df.EXT_SOURCE_1 * 2 + df.EXT_SOURCE_2 * 1 + df.EXT_SOURCE_3 * 3np.warnings.filterwarnings('ignore', r'All-NaN (slice|axis) encountered')for function_name in ['min', 'max', 'mean', 'nanmedian', 'var']:    feature_name = 'EXT_SOURCES_{}'.format(function_name.upper())    df[feature_name] = eval('np.{}'.format(function_name))(        df[['EXT_SOURCE_1', 'EXT_SOURCE_2', 'EXT_SOURCE_3']], axis=1)

登录后复制    In [29]

df['CREDIT_TO_ANNUITY_RATIO'] = df['AMT_CREDIT'] / df['AMT_ANNUITY']df['CREDIT_TO_GOODS_RATIO'] = df['AMT_CREDIT'] / df['AMT_GOODS_PRICE']df['ANNUITY_TO_INCOME_RATIO'] = df['AMT_ANNUITY'] / df['AMT_INCOME_TOTAL']df['CREDIT_TO_INCOME_RATIO'] = df['AMT_CREDIT'] / df['AMT_INCOME_TOTAL']df['INCOME_TO_EMPLOYED_RATIO'] = df['AMT_INCOME_TOTAL'] / df['DAYS_EMPLOYED']df['INCOME_TO_BIRTH_RATIO'] = df['AMT_INCOME_TOTAL'] / df['DAYS_BIRTH']    df['EMPLOYED_TO_BIRTH_RATIO'] = df['DAYS_EMPLOYED'] / df['DAYS_BIRTH']df['ID_TO_BIRTH_RATIO'] = df['DAYS_ID_PUBLISH'] / df['DAYS_BIRTH']df['CAR_TO_BIRTH_RATIO'] = df['OWN_CAR_AGE'] / df['DAYS_BIRTH']df['CAR_TO_EMPLOYED_RATIO'] = df['OWN_CAR_AGE'] / df['DAYS_EMPLOYED']df['PHONE_TO_BIRTH_RATIO'] = df['DAYS_LAST_PHONE_CHANGE'] / df['DAYS_BIRTH']

登录后复制    In [30]

def do_mean(df, group_cols, counted, agg_name):    gp = df[group_cols + [counted]].groupby(group_cols)[counted].mean().reset_index().rename(        columns={counted: agg_name})    df = df.merge(gp, on=group_cols, how='left')    del gp    gc.collect()    return df

登录后复制    In [31]

def do_median(df, group_cols, counted, agg_name):    gp = df[group_cols + [counted]].groupby(group_cols)[counted].median().reset_index().rename(        columns={counted: agg_name})    df = df.merge(gp, on=group_cols, how='left')    del gp    gc.collect()    return df

登录后复制    In [32]

def do_std(df, group_cols, counted, agg_name):    gp = df[group_cols + [counted]].groupby(group_cols)[counted].std().reset_index().rename(        columns={counted: agg_name})    df = df.merge(gp, on=group_cols, how='left')    del gp    gc.collect()    return df

登录后复制    In [33]

def do_sum(df, group_cols, counted, agg_name):    gp = df[group_cols + [counted]].groupby(group_cols)[counted].sum().reset_index().rename(        columns={counted: agg_name})    df = df.merge(gp, on=group_cols, how='left')    del gp    gc.collect()    return df

登录后复制    In [34]

group = ['ORGANIZATION_TYPE', 'NAME_EDUCATION_TYPE', 'OCCUPATION_TYPE', 'AGE_RANGE', 'CODE_GENDER']df = do_median(df, group, 'EXT_SOURCES_MEAN', 'GROUP_EXT_SOURCES_MEDIAN')df = do_std(df, group, 'EXT_SOURCES_MEAN', 'GROUP_EXT_SOURCES_STD')df = do_mean(df, group, 'AMT_INCOME_TOTAL', 'GROUP_INCOME_MEAN')df = do_std(df, group, 'AMT_INCOME_TOTAL', 'GROUP_INCOME_STD')df = do_mean(df, group, 'CREDIT_TO_ANNUITY_RATIO', 'GROUP_CREDIT_TO_ANNUITY_MEAN')df = do_std(df, group, 'CREDIT_TO_ANNUITY_RATIO', 'GROUP_CREDIT_TO_ANNUITY_STD')df = do_mean(df, group, 'AMT_CREDIT', 'GROUP_CREDIT_MEAN')df = do_mean(df, group, 'AMT_ANNUITY', 'GROUP_ANNUITY_MEAN')df = do_std(df, group, 'AMT_ANNUITY', 'GROUP_ANNUITY_STD')

登录后复制    In [35]

def label_encoder(df, categorical_columns=None):    if not categorical_columns:        categorical_columns = [col for col in df.columns if df[col].dtype == 'object']    for col in categorical_columns:        df[col], uniques = pd.factorize(df[col])    return df, categorical_columns

登录后复制    In [36]

def drop_application_columns(df):    drop_list = [        'CNT_CHILDREN', 'CNT_FAM_MEMBERS', 'HOUR_APPR_PROCESS_START',        'FLAG_EMP_PHONE', 'FLAG_MOBIL', 'FLAG_CONT_MOBILE', 'FLAG_EMAIL', 'FLAG_PHONE',        'FLAG_OWN_REALTY', 'REG_REGION_NOT_LIVE_REGION', 'REG_REGION_NOT_WORK_REGION',        'REG_CITY_NOT_WORK_CITY', 'OBS_30_CNT_SOCIAL_CIRCLE', 'OBS_60_CNT_SOCIAL_CIRCLE',        'AMT_REQ_CREDIT_BUREAU_DAY', 'AMT_REQ_CREDIT_BUREAU_MON', 'AMT_REQ_CREDIT_BUREAU_YEAR',         'COMMONAREA_MODE', 'NONLIVINGAREA_MODE', 'ELEVATORS_MODE', 'NONLIVINGAREA_AVG',        'FLOORSMIN_MEDI', 'LANDAREA_MODE', 'NONLIVINGAREA_MEDI', 'LIVINGAPARTMENTS_MODE',        'FLOORSMIN_AVG', 'LANDAREA_AVG', 'FLOORSMIN_MODE', 'LANDAREA_MEDI',        'COMMONAREA_MEDI', 'YEARS_BUILD_AVG', 'COMMONAREA_AVG', 'BASEMENTAREA_AVG',        'BASEMENTAREA_MODE', 'NONLIVINGAPARTMENTS_MEDI', 'BASEMENTAREA_MEDI',         'LIVINGAPARTMENTS_AVG', 'ELEVATORS_AVG', 'YEARS_BUILD_MEDI', 'ENTRANCES_MODE',        'NONLIVINGAPARTMENTS_MODE', 'LIVINGAREA_MODE', 'LIVINGAPARTMENTS_MEDI',        'YEARS_BUILD_MODE', 'YEARS_BEGINEXPLUATATION_AVG', 'ELEVATORS_MEDI', 'LIVINGAREA_MEDI',        'YEARS_BEGINEXPLUATATION_MODE', 'NONLIVINGAPARTMENTS_AVG', 'HOUSETYPE_MODE',        'FONDKAPREMONT_MODE', 'EMERGENCYSTATE_MODE'    ]    for doc_num in [2,4,5,6,7,9,10,11,12,13,14,15,16,17,19,20,21]:        drop_list.append('FLAG_DOCUMENT_{}'.format(doc_num))    df.drop(drop_list, axis=1, inplace=True)    return df

登录后复制    In [37]

df, le_encoded_cols = label_encoder(df, None)df = drop_application_columns(df)

登录后复制    In [38]

df = pd.get_dummies(df)

登录后复制    In [39]

bureau = pd.read_csv(os.path.join(DATA_DIRECTORY, 'bureau.csv'))

登录后复制    In [40]

bureau['CREDIT_DURATION'] = -bureau['DAYS_CREDIT'] + bureau['DAYS_CREDIT_ENDDATE']bureau['ENDDATE_DIF'] = bureau['DAYS_CREDIT_ENDDATE'] - bureau['DAYS_ENDDATE_FACT']bureau['DEBT_PERCENTAGE'] = bureau['AMT_CREDIT_SUM'] / bureau['AMT_CREDIT_SUM_DEBT']bureau['DEBT_CREDIT_DIFF'] = bureau['AMT_CREDIT_SUM'] - bureau['AMT_CREDIT_SUM_DEBT']bureau['CREDIT_TO_ANNUITY_RATIO'] = bureau['AMT_CREDIT_SUM'] / bureau['AMT_ANNUITY']

登录后复制    In [41]

def one_hot_encoder(df, categorical_columns=None, nan_as_category=True):    original_columns = list(df.columns)    if not categorical_columns:        categorical_columns = [col for col in df.columns if df[col].dtype == 'object']    df = pd.get_dummies(df, columns=categorical_columns, dummy_na=nan_as_category)    categorical_columns = [c for c in df.columns if c not in original_columns]    return df, categorical_columns

登录后复制    In [42]

def group(df_to_agg, prefix, aggregations, aggregate_by= 'SK_ID_CURR'):    agg_df = df_to_agg.groupby(aggregate_by).agg(aggregations)    agg_df.columns = pd.Index(['{}{}_{}'.format(prefix, e[0], e[1].upper())                               for e in agg_df.columns.tolist()])    return agg_df.reset_index()

登录后复制    In [43]

def group_and_merge(df_to_agg, df_to_merge, prefix, aggregations, aggregate_by= 'SK_ID_CURR'):    agg_df = group(df_to_agg, prefix, aggregations, aggregate_by= aggregate_by)    return df_to_merge.merge(agg_df, how='left', on= aggregate_by)

登录后复制    In [44]

def get_bureau_balance(path, num_rows= None):    bb = pd.read_csv(os.path.join(path, 'bureau_balance.csv'))    bb, categorical_cols = one_hot_encoder(bb, nan_as_category= False)    # Calculate rate for each category with decay    bb_processed = bb.groupby('SK_ID_BUREAU')[categorical_cols].mean().reset_index()    # Min, Max, Count and mean duration of payments (months)    agg = {'MONTHS_BALANCE': ['min', 'max', 'mean', 'size']}    bb_processed = group_and_merge(bb, bb_processed, '', agg, 'SK_ID_BUREAU')    del bb; gc.collect()    return bb_processed

登录后复制    In [45]

bureau, categorical_cols = one_hot_encoder(bureau, nan_as_category= False)bureau = bureau.merge(get_bureau_balance(DATA_DIRECTORY), how='left', on='SK_ID_BUREAU')bureau['STATUS_12345'] = 0for i in range(1,6):    bureau['STATUS_12345'] += bureau['STATUS_{}'.format(i)]

登录后复制    In [46]

features = ['AMT_CREDIT_MAX_OVERDUE', 'AMT_CREDIT_SUM_OVERDUE', 'AMT_CREDIT_SUM',    'AMT_CREDIT_SUM_DEBT', 'DEBT_PERCENTAGE', 'DEBT_CREDIT_DIFF', 'STATUS_0', 'STATUS_12345']agg_length = bureau.groupby('MONTHS_BALANCE_SIZE')[features].mean().reset_index()agg_length.rename({feat: 'LL_' + feat for feat in features}, axis=1, inplace=True)bureau = bureau.merge(agg_length, how='left', on='MONTHS_BALANCE_SIZE')del agg_length; gc.collect()

登录后复制        

39

登录后复制登录后复制登录后复制                In [47]

BUREAU_AGG = {    'SK_ID_BUREAU': ['nunique'],    'DAYS_CREDIT': ['min', 'max', 'mean'],    'DAYS_CREDIT_ENDDATE': ['min', 'max'],    'AMT_CREDIT_MAX_OVERDUE': ['max', 'mean'],    'AMT_CREDIT_SUM': ['max', 'mean', 'sum'],    'AMT_CREDIT_SUM_DEBT': ['max', 'mean', 'sum'],    'AMT_CREDIT_SUM_OVERDUE': ['max', 'mean', 'sum'],    'AMT_ANNUITY': ['mean'],    'DEBT_CREDIT_DIFF': ['mean', 'sum'],    'MONTHS_BALANCE_MEAN': ['mean', 'var'],    'MONTHS_BALANCE_SIZE': ['mean', 'sum'],    'STATUS_0': ['mean'],    'STATUS_1': ['mean'],    'STATUS_12345': ['mean'],    'STATUS_C': ['mean'],    'STATUS_X': ['mean'],    'CREDIT_ACTIVE_Active': ['mean'],    'CREDIT_ACTIVE_Closed': ['mean'],    'CREDIT_ACTIVE_Sold': ['mean'],    'CREDIT_TYPE_Consumer credit': ['mean'],    'CREDIT_TYPE_Credit card': ['mean'],    'CREDIT_TYPE_Car loan': ['mean'],    'CREDIT_TYPE_Mortgage': ['mean'],    'CREDIT_TYPE_Microloan': ['mean'],    'LL_AMT_CREDIT_SUM_OVERDUE': ['mean'],    'LL_DEBT_CREDIT_DIFF': ['mean'],    'LL_STATUS_12345': ['mean'],}BUREAU_ACTIVE_AGG = {    'DAYS_CREDIT': ['max', 'mean'],    'DAYS_CREDIT_ENDDATE': ['min', 'max'],    'AMT_CREDIT_MAX_OVERDUE': ['max', 'mean'],    'AMT_CREDIT_SUM': ['max', 'sum'],    'AMT_CREDIT_SUM_DEBT': ['mean', 'sum'],    'AMT_CREDIT_SUM_OVERDUE': ['max', 'mean'],    'DAYS_CREDIT_UPDATE': ['min', 'mean'],    'DEBT_PERCENTAGE': ['mean'],    'DEBT_CREDIT_DIFF': ['mean'],    'CREDIT_TO_ANNUITY_RATIO': ['mean'],    'MONTHS_BALANCE_MEAN': ['mean', 'var'],    'MONTHS_BALANCE_SIZE': ['mean', 'sum'],}BUREAU_CLOSED_AGG = {    'DAYS_CREDIT': ['max', 'var'],    'DAYS_CREDIT_ENDDATE': ['max'],    'AMT_CREDIT_MAX_OVERDUE': ['max', 'mean'],    'AMT_CREDIT_SUM_OVERDUE': ['mean'],    'AMT_CREDIT_SUM': ['max', 'mean', 'sum'],    'AMT_CREDIT_SUM_DEBT': ['max', 'sum'],    'DAYS_CREDIT_UPDATE': ['max'],    'ENDDATE_DIF': ['mean'],    'STATUS_12345': ['mean'],}BUREAU_LOAN_TYPE_AGG = {    'DAYS_CREDIT': ['mean', 'max'],    'AMT_CREDIT_MAX_OVERDUE': ['mean', 'max'],    'AMT_CREDIT_SUM': ['mean', 'max'],    'AMT_CREDIT_SUM_DEBT': ['mean', 'max'],    'DEBT_PERCENTAGE': ['mean'],    'DEBT_CREDIT_DIFF': ['mean'],    'DAYS_CREDIT_ENDDATE': ['max'],}BUREAU_TIME_AGG = {    'AMT_CREDIT_MAX_OVERDUE': ['max', 'mean'],    'AMT_CREDIT_SUM_OVERDUE': ['mean'],    'AMT_CREDIT_SUM': ['max', 'sum'],    'AMT_CREDIT_SUM_DEBT': ['mean', 'sum'],    'DEBT_PERCENTAGE': ['mean'],    'DEBT_CREDIT_DIFF': ['mean'],    'STATUS_0': ['mean'],    'STATUS_12345': ['mean'],}

登录后复制    In [48]

agg_bureau = group(bureau, 'BUREAU_', BUREAU_AGG)active = bureau[bureau['CREDIT_ACTIVE_Active'] == 1]agg_bureau = group_and_merge(active,agg_bureau,'BUREAU_ACTIVE_',BUREAU_ACTIVE_AGG)closed = bureau[bureau['CREDIT_ACTIVE_Closed'] == 1]agg_bureau = group_and_merge(closed,agg_bureau,'BUREAU_CLOSED_',BUREAU_CLOSED_AGG)del active, closed; gc.collect()for credit_type in ['Consumer credit', 'Credit card', 'Mortgage', 'Car loan', 'Microloan']:    type_df = bureau[bureau['CREDIT_TYPE_' + credit_type] == 1]    prefix = 'BUREAU_' + credit_type.split(' ')[0].upper() + '_'    agg_bureau = group_and_merge(type_df, agg_bureau, prefix, BUREAU_LOAN_TYPE_AGG)    del type_df; gc.collect()for time_frame in [6, 12]:    prefix = "BUREAU_LAST{}M_".format(time_frame)    time_frame_df = bureau[bureau['DAYS_CREDIT'] >= -30*time_frame]    agg_bureau = group_and_merge(time_frame_df, agg_bureau, prefix, BUREAU_TIME_AGG)    del time_frame_df; gc.collect()

登录后复制    In [49]

sort_bureau = bureau.sort_values(by=['DAYS_CREDIT'])gr = sort_bureau.groupby('SK_ID_CURR')['AMT_CREDIT_MAX_OVERDUE'].last().reset_index()gr.rename({'AMT_CREDIT_MAX_OVERDUE': 'BUREAU_LAST_LOAN_MAX_OVERDUE'}, inplace=True)agg_bureau = agg_bureau.merge(gr, on='SK_ID_CURR', how='left')agg_bureau['BUREAU_DEBT_OVER_CREDIT'] = \    agg_bureau['BUREAU_AMT_CREDIT_SUM_DEBT_SUM']/agg_bureau['BUREAU_AMT_CREDIT_SUM_SUM']agg_bureau['BUREAU_ACTIVE_DEBT_OVER_CREDIT'] = \    agg_bureau['BUREAU_ACTIVE_AMT_CREDIT_SUM_DEBT_SUM']/agg_bureau['BUREAU_ACTIVE_AMT_CREDIT_SUM_SUM']

登录后复制    In [50]

df = pd.merge(df, agg_bureau, on='SK_ID_CURR', how='left')del agg_bureau, bureaugc.collect()

登录后复制        

39

登录后复制登录后复制登录后复制                In [51]

prev = pd.read_csv(os.path.join(DATA_DIRECTORY, 'previous_application.csv'))pay = pd.read_csv(os.path.join(DATA_DIRECTORY, 'installments_payments.csv'))

登录后复制    In [52]

PREVIOUS_AGG = {    'SK_ID_PREV': ['nunique'],    'AMT_ANNUITY': ['min', 'max', 'mean'],    'AMT_DOWN_PAYMENT': ['max', 'mean'],    'HOUR_APPR_PROCESS_START': ['min', 'max', 'mean'],    'RATE_DOWN_PAYMENT': ['max', 'mean'],    'DAYS_DECISION': ['min', 'max', 'mean'],    'CNT_PAYMENT': ['max', 'mean'],    'DAYS_TERMINATION': ['max'],    # Engineered features    'CREDIT_TO_ANNUITY_RATIO': ['mean', 'max'],    'APPLICATION_CREDIT_DIFF': ['min', 'max', 'mean'],    'APPLICATION_CREDIT_RATIO': ['min', 'max', 'mean', 'var'],    'DOWN_PAYMENT_TO_CREDIT': ['mean'],}PREVIOUS_ACTIVE_AGG = {    'SK_ID_PREV': ['nunique'],    'SIMPLE_INTERESTS': ['mean'],    'AMT_ANNUITY': ['max', 'sum'],    'AMT_APPLICATION': ['max', 'mean'],    'AMT_CREDIT': ['sum'],    'AMT_DOWN_PAYMENT': ['max', 'mean'],    'DAYS_DECISION': ['min', 'mean'],    'CNT_PAYMENT': ['mean', 'sum'],    'DAYS_LAST_DUE_1ST_VERSION': ['min', 'max', 'mean'],    # Engineered features    'AMT_PAYMENT': ['sum'],    'INSTALMENT_PAYMENT_DIFF': ['mean', 'max'],    'REMAINING_DEBT': ['max', 'mean', 'sum'],    'REPAYMENT_RATIO': ['mean'],}PREVIOUS_LATE_PAYMENTS_AGG = {    'DAYS_DECISION': ['min', 'max', 'mean'],    'DAYS_LAST_DUE_1ST_VERSION': ['min', 'max', 'mean'],    # Engineered features    'APPLICATION_CREDIT_DIFF': ['min'],    'NAME_CONTRACT_TYPE_Consumer loans': ['mean'],    'NAME_CONTRACT_TYPE_Cash loans': ['mean'],    'NAME_CONTRACT_TYPE_Revolving loans': ['mean'],}PREVIOUS_LOAN_TYPE_AGG = {    'AMT_CREDIT': ['sum'],    'AMT_ANNUITY': ['mean', 'max'],    'SIMPLE_INTERESTS': ['min', 'mean', 'max', 'var'],    'APPLICATION_CREDIT_DIFF': ['min', 'var'],    'APPLICATION_CREDIT_RATIO': ['min', 'max', 'mean'],    'DAYS_DECISION': ['max'],    'DAYS_LAST_DUE_1ST_VERSION': ['max', 'mean'],    'CNT_PAYMENT': ['mean'],}PREVIOUS_TIME_AGG = {    'AMT_CREDIT': ['sum'],    'AMT_ANNUITY': ['mean', 'max'],    'SIMPLE_INTERESTS': ['mean', 'max'],    'DAYS_DECISION': ['min', 'mean'],    'DAYS_LAST_DUE_1ST_VERSION': ['min', 'max', 'mean'],    # Engineered features    'APPLICATION_CREDIT_DIFF': ['min'],    'APPLICATION_CREDIT_RATIO': ['min', 'max', 'mean'],    'NAME_CONTRACT_TYPE_Consumer loans': ['mean'],    'NAME_CONTRACT_TYPE_Cash loans': ['mean'],    'NAME_CONTRACT_TYPE_Revolving loans': ['mean'],}PREVIOUS_APPROVED_AGG = {    'SK_ID_PREV': ['nunique'],    'AMT_ANNUITY': ['min', 'max', 'mean'],    'AMT_CREDIT': ['min', 'max', 'mean'],    'AMT_DOWN_PAYMENT': ['max'],    'AMT_GOODS_PRICE': ['max'],    'HOUR_APPR_PROCESS_START': ['min', 'max'],    'DAYS_DECISION': ['min', 'mean'],    'CNT_PAYMENT': ['max', 'mean'],    'DAYS_TERMINATION': ['mean'],    # Engineered features    'CREDIT_TO_ANNUITY_RATIO': ['mean', 'max'],    'APPLICATION_CREDIT_DIFF': ['max'],    'APPLICATION_CREDIT_RATIO': ['min', 'max', 'mean'],    # The following features are only for approved applications    'DAYS_FIRST_DRAWING': ['max', 'mean'],    'DAYS_FIRST_DUE': ['min', 'mean'],    'DAYS_LAST_DUE_1ST_VERSION': ['min', 'max', 'mean'],    'DAYS_LAST_DUE': ['max', 'mean'],    'DAYS_LAST_DUE_DIFF': ['min', 'max', 'mean'],    'SIMPLE_INTERESTS': ['min', 'max', 'mean'],}PREVIOUS_REFUSED_AGG = {    'AMT_APPLICATION': ['max', 'mean'],    'AMT_CREDIT': ['min', 'max'],    'DAYS_DECISION': ['min', 'max', 'mean'],    'CNT_PAYMENT': ['max', 'mean'],    # Engineered features    'APPLICATION_CREDIT_DIFF': ['min', 'max', 'mean', 'var'],    'APPLICATION_CREDIT_RATIO': ['min', 'mean'],    'NAME_CONTRACT_TYPE_Consumer loans': ['mean'],    'NAME_CONTRACT_TYPE_Cash loans': ['mean'],    'NAME_CONTRACT_TYPE_Revolving loans': ['mean'],}

登录后复制    In [53]

ohe_columns = [    'NAME_CONTRACT_STATUS', 'NAME_CONTRACT_TYPE', 'CHANNEL_TYPE',    'NAME_TYPE_SUITE', 'NAME_YIELD_GROUP', 'PRODUCT_COMBINATION',    'NAME_PRODUCT_TYPE', 'NAME_CLIENT_TYPE']prev, categorical_cols = one_hot_encoder(prev, ohe_columns, nan_as_category= False)

登录后复制    In [54]

prev['APPLICATION_CREDIT_DIFF'] = prev['AMT_APPLICATION'] - prev['AMT_CREDIT']prev['APPLICATION_CREDIT_RATIO'] = prev['AMT_APPLICATION'] / prev['AMT_CREDIT']prev['CREDIT_TO_ANNUITY_RATIO'] = prev['AMT_CREDIT']/prev['AMT_ANNUITY']prev['DOWN_PAYMENT_TO_CREDIT'] = prev['AMT_DOWN_PAYMENT'] / prev['AMT_CREDIT']total_payment = prev['AMT_ANNUITY'] * prev['CNT_PAYMENT']prev['SIMPLE_INTERESTS'] = (total_payment/prev['AMT_CREDIT'] - 1)/prev['CNT_PAYMENT']

登录后复制    In [55]

approved = prev[prev['NAME_CONTRACT_STATUS_Approved'] == 1]active_df = approved[approved['DAYS_LAST_DUE'] == 365243]active_pay = pay[pay['SK_ID_PREV'].isin(active_df['SK_ID_PREV'])]active_pay_agg = active_pay.groupby('SK_ID_PREV')[['AMT_INSTALMENT', 'AMT_PAYMENT']].sum()active_pay_agg.reset_index(inplace= True)active_pay_agg['INSTALMENT_PAYMENT_DIFF'] = active_pay_agg['AMT_INSTALMENT'] - active_pay_agg['AMT_PAYMENT']active_df = active_df.merge(active_pay_agg, on= 'SK_ID_PREV', how= 'left')active_df['REMAINING_DEBT'] = active_df['AMT_CREDIT'] - active_df['AMT_PAYMENT']active_df['REPAYMENT_RATIO'] = active_df['AMT_PAYMENT'] / active_df['AMT_CREDIT']active_agg_df = group(active_df, 'PREV_ACTIVE_', PREVIOUS_ACTIVE_AGG)active_agg_df['TOTAL_REPAYMENT_RATIO'] = active_agg_df['PREV_ACTIVE_AMT_PAYMENT_SUM']/\                                            active_agg_df['PREV_ACTIVE_AMT_CREDIT_SUM']del active_pay, active_pay_agg, active_df; gc.collect()

登录后复制        

0

登录后复制登录后复制登录后复制登录后复制登录后复制                In [56]

prev['DAYS_FIRST_DRAWING'].replace(365243, np.nan, inplace= True)prev['DAYS_FIRST_DUE'].replace(365243, np.nan, inplace= True)prev['DAYS_LAST_DUE_1ST_VERSION'].replace(365243, np.nan, inplace= True)prev['DAYS_LAST_DUE'].replace(365243, np.nan, inplace= True)prev['DAYS_TERMINATION'].replace(365243, np.nan, inplace= True)

登录后复制    In [57]

prev['DAYS_LAST_DUE_DIFF'] = prev['DAYS_LAST_DUE_1ST_VERSION'] - prev['DAYS_LAST_DUE']approved['DAYS_LAST_DUE_DIFF'] = approved['DAYS_LAST_DUE_1ST_VERSION'] - approved['DAYS_LAST_DUE']

登录后复制    In [58]

categorical_agg = {key: ['mean'] for key in categorical_cols}

登录后复制    In [59]

agg_prev = group(prev, 'PREV_', {**PREVIOUS_AGG, **categorical_agg})agg_prev = agg_prev.merge(active_agg_df, how='left', on='SK_ID_CURR')del active_agg_df; gc.collect()

登录后复制        

0

登录后复制登录后复制登录后复制登录后复制登录后复制                In [60]

agg_prev = group_and_merge(approved, agg_prev, 'APPROVED_', PREVIOUS_APPROVED_AGG)refused = prev[prev['NAME_CONTRACT_STATUS_Refused'] == 1]agg_prev = group_and_merge(refused, agg_prev, 'REFUSED_', PREVIOUS_REFUSED_AGG)del approved, refused; gc.collect()

登录后复制        

0

登录后复制登录后复制登录后复制登录后复制登录后复制                In [61]

for loan_type in ['Consumer loans', 'Cash loans']:    type_df = prev[prev['NAME_CONTRACT_TYPE_{}'.format(loan_type)] == 1]    prefix = 'PREV_' + loan_type.split(" ")[0] + '_'    agg_prev = group_and_merge(type_df, agg_prev, prefix, PREVIOUS_LOAN_TYPE_AGG)    del type_df; gc.collect()

登录后复制    In [62]

pay['LATE_PAYMENT'] = pay['DAYS_ENTRY_PAYMENT'] - pay['DAYS_INSTALMENT']pay['LATE_PAYMENT'] = pay['LATE_PAYMENT'].apply(lambda x: 1 if x > 0 else 0)dpd_id = pay[pay['LATE_PAYMENT'] > 0]['SK_ID_PREV'].unique()

登录后复制    In [63]

agg_dpd = group_and_merge(prev[prev['SK_ID_PREV'].isin(dpd_id)], agg_prev,                                    'PREV_LATE_', PREVIOUS_LATE_PAYMENTS_AGG)del agg_dpd, dpd_id; gc.collect()

登录后复制        

0

登录后复制登录后复制登录后复制登录后复制登录后复制                In [64]

for time_frame in [12, 24]:    time_frame_df = prev[prev['DAYS_DECISION'] >= -30*time_frame]    prefix = 'PREV_LAST{}M_'.format(time_frame)    agg_prev = group_and_merge(time_frame_df, agg_prev, prefix, PREVIOUS_TIME_AGG)    del time_frame_df; gc.collect()del prev; gc.collect()

登录后复制        

0

登录后复制登录后复制登录后复制登录后复制登录后复制                In [65]

df = pd.merge(df, agg_prev, on='SK_ID_CURR', how='left')

登录后复制    In [66]

train = df[df['TARGET'].notnull()]test = df[df['TARGET'].isnull()]del dfgc.collect()

登录后复制        

98

登录后复制                In [67]

labels = train['TARGET']test_lebels=test['TARGET']train = train.drop(columns=['TARGET'])test = test.drop(columns=['TARGET'])

登录后复制    In [68]

feature = list(train.columns)train.replace([np.inf, -np.inf], np.nan, inplace=True)test.replace([np.inf, -np.inf], np.nan, inplace=True)test_df = test.copy()train_df = train.copy()train_df['TARGET'] = labelstest_df['TARGET'] = test_lebels

登录后复制    In [69]

imputer = SimpleImputer(strategy = 'median')imputer.fit(train)imputer.fit(test)train = imputer.transform(train)test = imputer.transform(test)

登录后复制    In [70]

scaler = MinMaxScaler(feature_range = (0, 1))scaler.fit(train)scaler.fit(test)train = scaler.transform(train)test = scaler.transform(test)

登录后复制    In [71]

from lightgbm import LGBMClassifierlgbmc = LGBMClassifier()lgbmc.fit(train, labels)

登录后复制        

LGBMClassifier()

登录后复制                In [72]

lgbm_pred = lgbmc.predict_proba(test)[:, 1]

登录后复制    In [74]

submit = test_df[['SK_ID_CURR']]submit['TARGET'] = lgbm_pred

登录后复制    In [75]

submit.to_csv('lgbm.csv', index = False)

登录后复制    

总结

数据的提交结果如下：（提交需要科学上网）

免责声明

游乐网为非赢利性网站，所展示的游戏/软件/文章内容均来自于互联网或第三方用户上传分享，版权归原作者所有，本站不承担相应法律责任。如您发现有涉嫌抄袭侵权的内容，请联系youleyoucom@outlook.com。

同类文章