机器学习处理离散值方法之 95分位数盖帽法

def train_add_hat(x,features):
    import numpy as np
    import pandas as pd
    df=x.copy()
    q95_dict={}
    for col in features:
        q95=np.percentile(df[col],95)
        q95_dict[col]=q95
        b=np.array(df[col])
        c=list(map(lambda x:q95 if x>95 else x,b))
        df=df.drop(col,axis=1)
        df[col]=c
    return df,q95_dict

#用同一标准处理测试集
def add_hat(x,features,q95_dict):
    import numpy as np
    import pandas as pd
    df = x.copy()
    len_d=len(df.index)
    for col in features:
        q95=q95_dict[col]
        b=np.array(df[col])
        c=list(map(lambda x:q95 if x>q95 else x,b))
        df=df.drop(col,axis=1)
        df[col]=c
    return df