--fold_column

Switch

--fold_column FIELD

Description

A column in --outcome_table giving pre-assigned fold labels to use for cross-validation in prediction/classification, instead of DLATK randomly splitting groups into folds.

Argument and Default Value

A column name. Default: None (random folds). Aliased as --fold_labels.

Details

Passed straight through to OutcomeGetter(..., fold_column=args.fold_column), so every group's fold membership comes from this column's value rather than a random assignment — useful for reproducible splits, or splits that must respect a grouping (e.g. no user split across folds if pre-aggregated by some other unit). Use --test_folds to restrict evaluation to a subset of the fold values found in this column.

Other Switches

Required Switches:

• One of --nfold_test_regression, --nfold_test_classifiers, or another cross-validated prediction switch

Optional Switches:

• --test_folds

Example Commands

dlatkInterface.py -d dla_tutorial -t msgs -c user_id -f 'feat$1gram$msgs$user_id$16to16' \
--outcome_table blog_outcomes --outcomes age --fold_column predefined_fold --combo_test_regression