BIZNIQUE / LINEAR ALGEBRA / 25
Keep relevant features without leaking the answer
Keep relevant features without leaking the answer
Selection is part of model development, so it belongs inside the evaluation procedure.
Embedded Files
THE IDEA
Read the example.
Read the example.
Choosing k of d features can reduce storage and simplify a model. It can also remove useful interactions or stable weak signals. This example isolates the storage effect; predictive quality requires its own evaluation.
Selected dense values = n × k
Use it well
Use it well
Selecting features using the full dataset before cross-validation exposes validation outcomes to the selection process. Repeat supervised selection within each training fold.
TRY IT YOURSELF
A quick check.
A quick check.
If a feature looks useful only after inspecting test labels, is the test set still an independent final check?
Check your reasoning
Check your reasoning
No. The selection decision has used information from the test set.
Page updated
Google Sites
Report abuse