各类特征的创建-业务统计特征创建

# 创建空字典
aggs = {}

# 连续/离散字段统计量提取范围
for col in numeric_cols:
    aggs[col] = ['nunique', 'mean', 'min', 'max','var','skew', 'sum']
for col in categorical_cols:
    aggs[col] = ['nunique']    
aggs['card_id'] = ['size', 'count']
cols = ['card_id']

# 借助groupby实现统计量计算
for key in aggs.keys():
    cols.extend([key+'_'+stat for stat in aggs[key]])

df = transaction[transaction['month_lag']<0].groupby('card_id').agg(aggs).reset_index()
df.columns = cols[:1] + [co+'_hist' for co in cols[1:]]

df2 = transaction[transaction['month_lag']>=0].groupby('card_id').agg(aggs).reset_index()
df2.columns = cols[:1] + [co+'_new' for co in cols[1:]]
df = pd.merge(df, df2, how='left',on='card_id')

df2 = transaction.groupby('card_id').agg(aggs).reset_index()
df2.columns = cols
df = pd.merge(df, df2, how='left',on='card_id')
del transaction
gc.collect()

# 生成训练集与测试集
train = pd.merge(train, df, how='left', on='card_id')
test =  pd.merge(test, df, how='left', on='card_id')
del df
train.to_csv("preprocess/train_groupby.csv", index=False)
test.to_csv("preprocess/test_groupby.csv", index=False)

gc.collect()