使用Pandas将SAS嵌套条件代码迁移至Python的正确性校验及优化咨询

现有实现正确性检查 #

你当前的嵌套np.where实现除了一处逻辑错误外，其余逻辑和原SAS代码对齐，错误点如下：

• 原SAS中A < 30000分支的第一层D值判断为D < 10000，但你的Python代码里误写为D < 40000，会导致该分支下的赋值逻辑完全错误。

此外嵌套np.where的写法可读性极差，后续维护或调整条件时很容易出错，虽本身兼容Pandas 0.24.2版本，但非常不推荐。

更优实现方案（兼容Pandas 0.24.2） #

推荐按优先级分条件掩码赋值的方案，逻辑清晰，易排查问题，完全适配你当前的版本要求：

import numpy as np
import pandas as pd

# 先初始化final列，未匹配到条件的保持NaN，可根据业务需求调整默认值
df['final'] = np.nan

# 第一分支：A<50且B<10000
cond1 = (df['A'] < 50) & (df['B'] < 10000)
df.loc[cond1 & (df['B'] >75) & (df['C']>52), 'final'] = 10
df.loc[cond1 & ~((df['B'] >75) & (df['C']>52)), 'final'] = 2

# 第二分支：前面不满足，且A<500且B<20000
cond2 = df['final'].isna() & (df['A'] < 500) & (df['B'] < 20000)
## 子分支D<150000
cond2_1 = cond2 & (df['D'] < 150000)
df.loc[cond2_1 & (df['B']>750) & (df['C']>52), 'final'] = 10
df.loc[cond2_1 & ~((df['B']>750) & (df['C']>52)), 'final'] = 2
## 子分支150000<=D<600000
cond2_2 = cond2 & df['final'].isna() & (df['D'] < 600000)
df.loc[cond2_2 & (df['B']>3000) & (df['C']>52), 'final'] =30
df.loc[cond2_2 & ~((df['B']>3000) & (df['C']>52)), 'final'] =10

# 第三分支：前面不满足，且A<7000
cond3 = df['final'].isna() & (df['A'] <7000)
## 子分支D<40000
cond3_1 = cond3 & (df['D'] <40000)
df.loc[cond3_1 & (df['B']>200) & (df['C']>52), 'final'] =10
df.loc[cond3_1 & ~((df['B']>200) & (df['C']>52)), 'final'] =2
## 子分支40000<=D<200000
cond3_2 = cond3 & df['final'].isna() & (df['D'] <200000)
df.loc[cond3_2 & (df['B']>1000) & (df['C']>52), 'final'] =30
df.loc[cond3_2 & ~((df['B']>1000) & (df['C']>52)), 'final'] =10

# 第四分支：前面不满足，且A<30000
cond4 = df['final'].isna() & (df['A'] <30000)
## 子分支D<10000
cond4_1 = cond4 & (df['D'] <10000)
df.loc[cond4_1 & (df['B']>50) & (df['C']>52), 'final'] =10
df.loc[cond4_1 & ~((df['B']>50) & (df['C']>52)), 'final'] =2
## 子分支10000<=D<100000
cond4_2 = cond4 & df['final'].isna() & (df['D'] <100000)
df.loc[cond4_2 & (df['B']>500) & (df['C']>52), 'final'] =30
df.loc[cond4_2 & ~((df['B']>500) & (df['C']>52)), 'final'] =10

# 最后分支：前面都不满足，且D<1000
cond5 = df['final'].isna() & (df['D'] <1000)
df.loc[cond5, 'final'] =10

方案优势 #

• 完全和原SAS的判断顺序、逻辑对齐，没有嵌套层级，每一步条件都清晰可见，排查错误成本极低
• 所有用到的loc赋值、布尔索引语法都完全兼容Pandas 0.24.2，不需要升级版本
• 后续需要调整条件阈值时，直接修改对应条件即可，不需要拆解多层嵌套结构

内容的提问来源于stack exchange，提问作者Dev