pandas 实现等频分箱

#等频分箱
def frequencybox(demo, name, new_name, n):
    demo["tmp"] = pd.qcut(demo[name],n)
    
    group_by_age_bin = demo.groupby(["tmp"],as_index=True)
    df_min_max_bin = pd.DataFrame()#用来记录每个箱体的最大最小值
    df_min_max_bin["min_bin"] = group_by_age_bin[name].min()
    df_min_max_bin["max_bin"] = group_by_age_bin[name].max()

    df_min_max_bin.reset_index(inplace=True)
    q = df_min_max_bin["tmp"].values
    dicc = {}
    for i in range(len(q)):
        dicc[str(q[i])] = i+1
    demo["tmp"] = demo["tmp"].apply(lambda x: str(x))
    if(new_name==name):
        demo.drop(name, axis=1, inplace=True)
    demo[new_name] = demo["tmp"].map(dicc) 
    demo.drop("tmp", axis=1, inplace=True)
lm = all_feature.copy()
frequencybox(lm, "客户年龄","客户年龄", 5)

报错：
Bin edges must be unique: array([ 1., 1., 1., 1., 1., 1., 1., 1., 2., 3., 89.]).
You can drop duplicate edges by setting the ‘duplicates’ kwarg

解决方案；
http://www.cocoachina.com/articles/459499

https://stackoverflow.com/questions/20158597/how-to-qcut-with-non-unique-bin-edges

原来
pd.qcut(df, nbins)
变为
pd.qcut(df.rank(method=‘first’), nbins)

支持多列转换，并返回切分表格：

### 导出iv值 特征

iv_feature = tmp_feature.copy()
###等频分箱  NaN不进行分箱 还作为NaN出现
def frequencybox(demo, name_list, new_name_list, n):
	# demo : 输入原始df表
	# name_list :demo表中原始字段组成list
	# new_name_list :希望分箱后特征名称 所组成list	 name_list[i]可与new_name_list[i]重名
	# n: 分桶数
    if (len(name_list)!= len(new_name_list)):
        print("输入参数数量不一致！！！")
        return 
    result_df = pd.DataFrame()
    for i in range(len(name_list)):
        df_min_max_bin = pd.DataFrame()#用来记录每个箱体的最大最小值

        name = name_list[i]
        new_name = new_name_list[i]
        demo["tmp"] = pd.qcut(demo[name].rank(method='first'), n)
        group_by_age_bin = demo.groupby(["tmp"],as_index=True)
        df_min_max_bin["min_bin"] = group_by_age_bin[name].min()
        df_min_max_bin["max_bin"] = group_by_age_bin[name].max()
        df_min_max_bin["feature_name"] = name
        df_min_max_bin.reset_index(inplace=True)
        if(len(df_min_max_bin)==0):
            result_df = df_min_max_bin
        else:
            result_df = pd.concat([result_df, df_min_max_bin])
#             result_df = result_df.append(df_min_max_bin)
              
        q = df_min_max_bin["tmp"].values
        dicc = {}
        for i in range(len(q)):
            dicc[str(q[i])] = i+1
        demo["tmp"] = demo["tmp"].apply(lambda x: str(x))
        if(new_name==name):
            demo.drop(name, axis=1, inplace=True)
        demo[new_name] = demo["tmp"].map(dicc) 
        demo.drop("tmp", axis=1, inplace=True)
#     result_df.drop("index", inplace=True)
    result_df.reset_index(inplace=True)
    result_df.drop("index", axis =1 , inplace=True)
#     result_df = result_df.loc[:,~result_df.columns.isin(["index"])]
    return result_df
result_df = frequencybox(iv_feature, ["a","b","c"],["a","b","c"], 5)
result_df.head(20)

结果：

升级为增加WoE和IV属性

def frequencybox(demo, name_list, new_name_list, n, sum_label_1, sum_label_0, label_name):
	# demo : 输入原始df表
	# name_list :demo表中原始字段组成list
	# new_name_list :希望分箱后特征名称 所组成list	 name_list[i]可与new_name_list[i]重名
	# n: 分桶数
    
    if (len(name_list)!= len(new_name_list)):
        print("输入参数数量不一致！！！")
        return 
    result_df = pd.DataFrame()
    for i in range(len(name_list)):
        df_min_max_bin = pd.DataFrame()#用来记录每个箱体的最大最小值

        name = name_list[i]
        new_name = new_name_list[i]
        demo["tmp"] = pd.qcut(demo[name].rank(method='first'), n)
        group_by_age_bin = demo.groupby(["tmp"],as_index=True)
        df_min_max_bin["min"] = group_by_age_bin[name].min()
        df_min_max_bin["max"] = group_by_age_bin[name].max()
        df_min_max_bin["feature_name"] = name
        df_min_max_bin.reset_index(inplace=True)
        df_min_max_bin["valueFrom"] = df_min_max_bin["min"]
        df_min_max_bin["valueTo"] = df_min_max_bin["max"]
        df_min_max_bin["count"] = -100
        df_min_max_bin["nums_label_1"] = -10
        df_min_max_bin["nums_label_0"] = -10

        for i in range(0,len(df_min_max_bin)):
            if (i==len(df_min_max_bin)-1):
                df_min_max_bin.loc[i, "valueTo"] = df_min_max_bin.loc[i, "max"] + 0.01
            else:
                if(i!=0):
                    df_min_max_bin.loc[i, "valueFrom"] = df_min_max_bin.loc[i-1, "valueTo"]
            mmin = df_min_max_bin.loc[i, "valueFrom"]
            mmax = df_min_max_bin.loc[i, "valueTo"]
            tmp_de = demo[(demo[name]>=mmin) & (demo[name]<mmax)]
            print("第"+str(i)+"个分箱数量为："+ str(len(tmp_de)))
            df_min_max_bin.loc[i, "count"] = len(tmp_de)
            df_min_max_bin.loc[i, "nums_label_1"] = len(tmp_de[tmp_de[label_name]==1])
            df_min_max_bin.loc[i, "nums_label_0"] = len(tmp_de[tmp_de[label_name]==0])
            df_min_max_bin.loc[i, "min"] = tmp_de[name].min()
            df_min_max_bin.loc[i, "max"] = tmp_de[name].max()
            
            
        
        df_min_max_bin = df_min_max_bin[~(df_min_max_bin['valueFrom']>df_min_max_bin["valueTo"])]
        for x in ["tmp","min", "max", "valueFrom", "valueTo"]:
            df_min_max_bin[x] = df_min_max_bin[x].apply(lambda x: str(x))
            
            

        sum_nonull = df_min_max_bin["count"].sum()
        sum_count = sum_label_1 + sum_label_0
        if sum_nonull < sum_count:
            null_demo = demo[demo[name].isnull()]
#             null_label_1 = null_demo[label_name].value_counts()[1]
#             null_label_0 = null_demo[label_name].value_counts()[0]
            
            null_label_1 = len(null_demo[null_demo[label_name]==1])
            null_label_0 = len(null_demo[null_demo[label_name]==0])
        
            sum_null = sum_count-sum_nonull
            ha_dic = {"tmp":"null", "min":"null", "max":"null", "feature_name":name,
                     "valueFrom":"null", "valueTo":"null", "count":sum_null,
                     "nums_label_1":null_label_1, "nums_label_0":null_label_0}
            haha = pd.DataFrame(data = ha_dic,index = [0])
            haha["tmp"] = haha["tmp"].astype('category')
            df_min_max_bin= pd.concat([df_min_max_bin,haha])
            
            
        df_min_max_bin["ok1"] = ((df_min_max_bin["nums_label_1"]+1)/sum_label_1)/((df_min_max_bin["nums_label_0"]+1)/sum_label_0)
        
        df_min_max_bin["WoE"] = np.log(df_min_max_bin["ok1"])
		df_min_max_bin["ok2"] = ((df_min_max_bin["nums_label_1"]+1)/sum_label_1)-((df_min_max_bin["nums_label_0"]+1)/sum_label_0)
    
        df_min_max_bin["IV"] = df_min_max_bin["ok2"]*df_min_max_bin["WoE"]
        df_min_max_bin["IVs"] = df_min_max_bin["IV"].sum()
        if(len(df_min_max_bin)==0):
            result_df = df_min_max_bin
        else:
            result_df = pd.concat([result_df, df_min_max_bin])
              
        q = df_min_max_bin["tmp"].values
        dicc = {}
        for i in range(len(q)):
            dicc[str(q[i])] = i+1
        demo["tmp"] = demo["tmp"].apply(lambda x: str(x))
        if(new_name==name):
            demo.drop(name, axis=1, inplace=True)
        demo[new_name] = demo["tmp"].map(dicc) 
        demo.drop("tmp", axis=1, inplace=True)
    result_df.reset_index(inplace=True)
    result_df.drop("index", axis =1 , inplace=True)
    res_columns = ["feature_name", "valueFrom", "valueTo", "max", "min", "count", "WoE", "IV", "IVs"]
    result_df = result_df[res_columns]
    return result_df