Honey, I shrunk the target variable! Common pitfalls when transforming the targe...
summary
Claims that transforming the target (log) changes what the model actually minimizes and shows the math: minimizing RMSE on log targets yields the median when you exponentiate back unless you add sigma^2/2 correction to recover the mean. Uses scikit learn and random forest on a Kaggle used‑car dataset, provides Jupyter notebooks and a blog post, and reports concrete gains — e.g. improves RMSPE by ~28% in a Kaggle solution — so this is about actionable corrections not vague feature talk.
AI-written summary · about
chapters (7)
- 0:00Intro
- 1:39Metrics
- 3:47Use case
- 11:00I shrunk the target variable
- 15:33Can we use this
- 22:10Questions
- 30:39Summary
related talks
48mThe Missing Step in Moving from Data to Models
22mAccelerating ML impact
32mResource Utilization as a Metric for Machine Learning
27mPyTesting the Limits of Machine Learning
17mCodifying Data Science Intuition
35mWhat Failure Taught Me About Building High-Stakes Models
17mMaking sense of marginal effects
24mThe Forecast Whisperer: Secrets of Model Tuning Revealed