Lessons
1What did the system know at prediction time?35 min read
Exclude features that were unavailable when a decision was made.
- →Construct point-in-time valid training rows
2Split by the unit you want to generalize to35 min read
Choose grouped and temporal validation from deployment requirements.
- →Choose validation units that match deployment
- →Prevent entity leakage across evaluation splits
3Labels can be missing for a reason35 min read
Explain delayed and selectively observed labels without treating them as negatives.
- →Define labels with delay and selection limits
4Make training and serving use the same feature meaning35 min read
Specify a model package that preserves preprocessing and missing-value semantics.
- →Package preprocessing with the model contract
- →Audit training-serving feature parity
Skills in this course
- 01Construct point-in-time valid training rowsConstruct point-in-time valid training rows.
- 02Choose validation units that match deploymentChoose validation units that match deployment.
- 03Define labels with delay and selection limitsDefine labels with delay and selection limits.
- 04Package preprocessing with the model contractPackage preprocessing with the model contract.
- 05Prevent entity leakage across evaluation splitsPrevent entity leakage across evaluation splits.
- 06Audit training-serving feature parityAudit training-serving feature parity.