%matplotlib inline
import matplotlib.pyplot as plt
import pandas as pd
import numpy as np
df = pd.read_csv(‘data.csv‘)
df_raw = df.copy()
df.dropna(axis=0,inplace=True)
df.drop([‘custid‘,‘trade_no‘,‘bank_card_no‘,‘source‘,‘id_name‘],axis=1,inplace=True)
df.drop([‘consume_top_time_last_1_month‘,‘consume_top_time_last_6_month‘],axis=1,inplace=True)  #这两个特征重复了
new_df = pd.DataFrame(pd.to_datetime(df[‘loans_latest_time‘]) - pd.to_datetime(df[‘loans_latest_time‘].min()))
df[‘loans_latest_time‘] = new_df
df[‘loans_latest_time‘] = df[‘loans_latest_time‘].map(lambda x:x.days) 
new_df = pd.DataFrame(pd.to_datetime(df[‘latest_query_time‘]) - pd.to_datetime(df[‘latest_query_time‘].min()))
df[‘latest_query_time‘] = new_df
df[‘latest_query_time‘] = df[‘latest_query_time‘].map(lambda x:x.days)   #一样的操作
from sklearn import preprocessing
label = preprocessing.LabelEncoder()
reg = label.fit_transform(df[‘reg_preference_for_trad‘])
df[‘reg_preference_for_trad‘] = reg

#特征选择
#1根据相关系数
coor = df.corr()
corr_status = abs(coor[‘status‘])
corr_status_sorted = corr_status.sort_values(ascending=False)
corr_status_sorted = corr_status_sorted[corr_status_sorted.values>0.1]  #根据相关系数大于0.1，最终选择出12个特征
features_index = corr_status_sorted.index
corr_status_sorted

#2根据IV（信息价值）
#此函数借鉴于别人，因为自己也是才学习到IV这个概念
def calc_iv(df, feature, target, pr=False):
    """
    Set pr=True to enable printing of output.

    Output:
      * iv: float,
      * data: pandas.DataFrame
    """
    lst = []
    df[feature] = df[feature].fillna(‘NULL‘)

    for i in range(df[feature].nunique()):  # nuinque()是查看该序列(axis=0/1对应着列或行)的不同值的数量个数
        val = list(df[feature].unique())[i]  #这里是讲特征里的每个值作为分组，也可以分段分组
        lst.append([feature,
                    val,  # Value
                    df[df[feature] == val].count()[feature],  # 特征中等于分组值的计数
                    df[(df[feature] == val) & (df[target] == 0)].count()[feature],  # good rate
                    df[(df[feature] == val) & (df[target] == 1)].count()[feature]])  # bad rate
    data = pd.DataFrame(lst, columns=[‘Variable‘, ‘Value‘, ‘All‘, ‘Good‘, ‘Bad‘])

    data[‘Share‘] = data[‘All‘] / data[‘All‘].sum()
    data[‘Bad Rate‘] = data[‘Bad‘] / data[‘All‘]
    data[‘Distribution Good‘] = (data[‘All‘] - data[‘Bad‘]) / (data[‘All‘].sum() - data[‘Bad‘].sum())
    data[‘Distribution Bad‘] = data[‘Bad‘] / data[‘Bad‘].sum()
    data[‘WoE‘] = np.log(data[‘Distribution Good‘] / data[‘Distribution Bad‘]) #woe计算式为log（（根据分组值选出的好的/所有的好的）/
                                                                                        #（根据分组值选出的坏的的/所有的坏的）

    data = data.replace({‘WoE‘: {np.inf: 0, -np.inf: 0}})

    data[‘IV‘] = data[‘WoE‘] * (data[‘Distribution Good‘] - data[‘Distribution Bad‘]) #计算iv要乘后面那个是为了保证用比率来限制iv
                                                                                      #并且iv值范围为【0，+无穷】

    data = data.sort_values(by=[‘Variable‘, ‘Value‘], ascending=[True, True])
    data.index = range(len(data.index))

    if pr:
        print(data)
        print("IV = ", data[‘IV‘].sum())

    iv = data[‘IV‘].sum()   #所有分组值的iv求和就是该特征的iv值

    return iv, data

column_headers = list(df.columns.values)
# print(column_headers)

d=[]
for x in column_headers:   #输入每个特征，单独计算IV值
    IV_1, data = calc_iv(df, x, ‘status‘)
    d.append(IV_1)
#整理成Series类型并合并
column_headers=pd.Series(column_headers,name=‘feature‘)
d=pd.Series(d,name=‘iv_value‘)

# print(column_headers)
iv_result=pd.concat([column_headers,d],axis=1)
iv_result.sort_values(by=‘iv_value‘,ascending=False)

iv_result[iv_result[‘iv_value‘]>0.4]   #选择大于0.4值的作为特征

#3利用随机深林的算法算出特征的重要性
y = df[‘status‘]
X = df[df.loc[:,df.columns != ‘status‘].columns]
from sklearn.ensemble import RandomForestClassifier
forest_clf = RandomForestClassifier()
forest_clf.fit(X,y)
features_weights = dict(zip(df.columns,forest_clf.feature_importances_))
features_weights_sorted = sorted(features_weights.items(),key = lambda x:x[1],reverse = True)  #根据值排序，并选择前15个
for item in range(15):
    print(features_weights_sorted[item][0],‘:‘,features_weights_sorted[item][1])