建模样本分数分布matplotlib，结合psi判断模型拟合程度

建模样本分数分布matplotlib，结合psi判断模型拟合程度

画此图的目的是除了ks和auc指数，结合建模样本和OOT样本的psi，利用分数分布图进行比较，判断模型拟合程度。

psi计算

def multi_psi_for_continue_var(expected_frame, actual_frame, bins=10, bucket_type='bins', detail=False,
                               save_file_path=None):
    col_list = expected_frame.columns.tolist()
    if detail == True:
        psi_value = []
        for i in col_list:
            expected_array = expected_frame[i]
            actual_array = actual_frame[i]
            _psi = psi_for_continue_var(expected_array, actual_array, bins=bins, bucket_type=bucket_type, detail=detail,
                                        save_file_path=save_file_path)
            if not isinstance(_psi, int):
                _psi['col_name'] = str(i)
                print(i)
                psi_value.append(_psi)
            else:
                print(i)
        psi_all = pd.concat(psi_value, sort=False)
    elif detail == False:
        psi_value = {}
        for i in col_list:
            expected_array = expected_frame[i]
            actual_array = actual_frame[i]
            _psi = psi_for_continue_var(expected_array, actual_array, bins=bins, bucket_type=bucket_type, detail=detail,
                                        save_file_path=save_file_path)
            psi_value[i] = _psi
        psi_all = pd.DataFrame(psi_value, sort=False)
    return psi_all

分数分布图

def score_distribution_plot(data,title,label=None,perc=False):
    
    '''
    ----------------------------------------------------------------------
    功能: 画出单个项目分数分布图
    ----------------------------------------------------------------------
    :param data: numpy array of score by model，模型分数
    :param title: str of title，图标题
    :param perc: bool,取值为True时分数为百分制，False时为1分制
    ----------------------------------------------------------------------
    
    '''
    import pandas as pd
    import numpy as np
    from matplotlib import pyplot as plt
    import matplotlib
    
    def score(x):
        if x >=0 and x<10:
            return '[0-10)'
        elif x>=10 and x<20:
            return '[10-20)'
        elif x>=20 and x<30:
            return '[20-30)'
        elif x>=30 and x<40:
            return '[30-40)'
        elif x>=40 and x<50:
            return '[40-50)'
        elif x>=50 and x<60:
            return '[50-60)'
        elif x>=60 and x<70:
            return '[60-70)'
        elif x>=70 and x<80:
            return '[70-80)'
        elif x>=80 and x<90:
            return '[80-90)'
        elif x>=90 and x<100:
            return '[90-100]'
    
    x_laybel = ['[0-10)','[10-20)','[20-30)','[30-40)','[40-50)','[50-60)','[60-70)','[70-80)','[80-90)','[90-100]']
    data = pd.DataFrame(data)
    
    #判断分数是否为*100形式，默认否
    if perc == False:
        data = data*100
    dt1 = []
    for i in data.index.values.tolist():
        dt1.append(score(data.iloc[i][0]))
    dt1 = pd.DataFrame(dt1)
    dt2 = pd.DataFrame(columns=['[0-10)','[10-20)','[20-30)','[30-40)','[40-50)','[50-60)','[60-70)','[70-80)','[80-90)','[90-100]'],data=[[np.sum(dt1[0].values=='[0-10)'),np.sum(dt1[0].values=='[10-20)'),np.sum(dt1[0].values=='[20-30)'),np.sum(dt1[0].values=='[30-40)'),np.sum(dt1[0].values=='[40-50)'),np.sum(dt1[0].values=='[50-60)'),np.sum(dt1[0].values=='[60-70)'),np.sum(dt1[0].values=='[70-80)'),np.sum(dt1[0].values=='[80-90)'),np.sum(dt1[0].values=='[90-100)')]])
    #计算各分段占比
    dt3 = pd.DataFrame(dt2.apply(lambda x:x*100/len(dt1)))

    plt.figure(figsize=(10,8))#设置画布的尺寸
    plt.title(title,fontsize=20)#标题，并设定字号大小
    plt.xlabel(u'score',fontsize=14)#设置x轴，并设定字号大小
    plt.ylabel(u'%',fontsize=14)#设置y轴，并设定字号大小
 
    width_val = 0.7 #若显示 n 个柱状图，则width_val的值需小于1/n ，否则柱形图会有重合
 
    #alpha：透明度；width：柱子的宽度；facecolor：柱子填充色；edgecolor：柱子轮廓色；lw：柱子轮廓的宽度；label：图例；
    plt.bar(dt3.columns.values,dt3.iloc[0,:], alpha=1,width = width_val,  facecolor = 'blue', edgecolor = 'blue', lw=0.5, label=label)
    
    #标注数据
    for x,y in enumerate(dt3.loc[0,:]):
        plt.text(x, y+0.5, '%s' %round(y,1), ha='center')
 
    plt.legend(loc=2)#图例展示位置，数字代表第几象限