Regression and Ranking Metrics
A metric is a choice about which errors you will tolerate.
-
-
Classification Metrics and Decision Thresholds
Ranking, probability, and decision are three separate things to evaluate.
-
Working with Tabular Data at Scale
Unglamorous techniques, each worth a large constant factor.
-
Feature Engineering and Feature Selection
Both are called art. Both have measurable effects.
-
Data Preparation for Machine Learning
Most of a project’s decisions are made here, and almost none are recorded.
-
Data Leakage and Train-Validation-Test Splits
Leakage compromises evaluation when unavailable information enters fitting or selection; it need not raise every observed score.