M6-010Hard模拟与校验 · Simulation & validation🔥 大厂高频出现 6 次
如何用模拟验证“实验结论是否可靠”?请给出思路和代码骨架。
How would you use simulation to verify that an experimental conclusion is reliable?
💡 考场高分一句话答案:
设定真实效应与噪声,模拟 1000 次实验,统计假阳性率与功效,用来校验方法或设计参数。
Set a true effect and noise level, simulate a thousand experiments, and measure the false-positive rate and power. Use it to validate a method or size a design.
📌 核心考察要点 (Key Points)
- A/A 模拟:真实效应为 0,看拒绝率是否接近 α(应 ≈ 5%)
- 功效模拟:设定真实效应 = MDE,看拒绝率是否 ≈ power(应 ≈ 80%)
- 用途:验证新指标、新方差估计、新序贯方法
- 注意:模拟要复现真实数据结构(相关性、重尾),否则结论不可迁移
- A/A simulation: true effect zero; the rejection rate should be about alpha (5%)
- Power simulation: set the true effect equal to the MDE; the rejection rate should be about 80%
- Use: validate new metrics, new variance estimators, sequential methods
- Caveat: the simulation must reproduce the real data structure (correlation, heavy tails) or the conclusion does not transfer
🔍 大厂面试官连环追问 (Follow-ups)
- 模拟结果偏离 5% 说明什么?
- 如何模拟重尾数据?
🎯 关联大厂考题实战训练 (6 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。
⚡
想检验本题的肌肉记忆?进入 3 阶闪卡主动回忆盲测
隐藏答案遮罩,自测:审题锚点 ➔ 数理公式机理 ➔ 工业标杆代码与连环追问。3 分钟攻破认知盲区!