M6-011Mediumpandas 管道 · pandas pipelines🔥 大厂高频出现 3 次
如何用 pandas 做数据清洗的标准流程?
What is your standard data-cleaning pipeline in pandas?
💡 考场高分一句话答案:
五步:类型与时间解析 → 去重 → 缺失处理 → 异常值处理 → 一致性校验(守恒、量级)。
Five steps: parse types and timestamps, dedup, handle missing values, handle outliers, then run consistency checks on conservation and order of magnitude.
📌 核心考察要点 (Key Points)
- 类型:ts 转 datetime、id 转 str(避免前导零丢失)
- 去重:按业务键 + 时间,keep='last'
- 缺失:区分 MCAR/MAR/MNAR,计数类填 0、比率类保留 NaN
- 异常:winsorize 或按业务上限截断,不要直接删(会引入选择偏差)
- 校验:子集之和 = 总量、量级与上周同日对比
- Types: parse ts to datetime, cast ids to string so leading zeros survive
- Dedup: on business key plus time, keep the last
- Missing: distinguish MCAR/MAR/MNAR; fill 0 for counts, keep NaN for ratios
- Outliers: winsorise or cap at business limits rather than deleting, which introduces selection bias
🔍 大厂面试官连环追问 (Follow-ups)
- 为什么不要直接删除异常值?
- 如何检测时区错乱?
🎯 关联大厂考题实战训练 (3 题)
嫌字多看太累?进入苏格拉底式对话推演实验室
零打字负担,以大厂考官真实困境切入,通过点选直觉选项暴露思维盲区,手把手推导 Why。
⚡
Test muscle memory on this topic via 3-Layer Flashcards
Active Recall: Mentally formulate Prompt ➔ Math Derivation ➔ Benchmark Implementation in 3 minutes.