Activity: Create a feature and inspection memo
Estimated active time: 45-75 minutes
Scenario
Use the synthetic learner-support dataset. The fictional platform wants to decide which model families deserve a place in an algorithm portfolio.
Task
Run feature selection, identify the columns that carry no signal, and report what removing them does to both score and model size.
Required sections
- Problem and decision.
- A ranking of features by importance, with the method named.
- Which features you removed and why.
- Validation score before and after removal.
- Keep, delay, or reject the reduced feature set.
- Limitation note.
Rules
- The dataset ships two columns with no relationship to the target. If your method does not find them, say so — that is a result about the method.
- Importance from a single fit is unstable. Repeat with a different seed before trusting a ranking.
- Removing a feature that improves the test score is selection on the test set.
Output
Use the relevant learner-resource template from learner-resources/ or write a one-page memo with the same sections.