python psi指标

def Cal_Psi(score,pre_score,length=10):
    import math
    labels=['c'+str(i) for i in range(length)]
    True_out,bins=pd.qcut(score,q=length,retbins=True,labels=labels)
    bins[0] = bins[0]-0.001 #cut左开右闭，之前最小值再分组后组记号为空，这里减0.01划到最左侧区间
    
    Pre_out,bins_=pd.cut(pre_score,bins=bins,retbins=True,labels=labels)
    
    a=pd.DataFrame(pd.Series(True_out).value_counts()).rename(columns={0:'val1'})
    a=a.applymap(lambda y : y/len(score)) #计算实际值区间占比
    
    b=pd.DataFrame(pd.Series(Pre_out).value_counts()).rename(columns={0:'val2'})
    b=b.applymap(lambda y : y/len(score))  #计算预测值区间占比
    
    re=pd.merge(a,b,left_index=True,right_index=True)
    
    psi=0
    for i in range(len(re)):
        if re['val1'][i]==0:
            re['val1'][i]=0.000001
        if re['val2'][i]==0:
            re['val2'][i]=0.000001
        p=((re['val2'][i]-re['val1'][i])*(math.log((re['val2'][i]/re['val1'][i]))))
        
        psi=psi+p
    return psi