--control_combo_sizes

Switch

--control_combo_sizes N [N2 ...]

Also aliased as --combo_sizes.

Description

Specifies the sizes of control combinations to try when running --nfold_test_regression or --nfold_test_classifiers. For each N given, every combination of N controls drawn from --outcome_controls is tried as its own model.

Argument and Default Value

One or more integers. Default: [] (no combo sizes; only the no-controls and all-controls models are run).

Details

For example, --outcome_controls age gender income --control_combo_sizes 1 2 runs every single-control model (age, gender, income) and every pair (age+gender, age+income, gender+income), in addition to the default no-controls and all-controls models. This is a way to explore how much of an outcome's variance is explained by each control (or pair of controls) without hand-writing every combination.

--all_controls_only restricts this to only the all-controls model, equivalent to passing every control's count as N.

Other Switches

Required Switches:

• --outcome_controls

• One of --nfold_test_classifiers or --nfold_test_regression

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 --outcome_controls gender is_student is_education \
--combo_test_regression --control_combo_sizes 1 2 --folds 10 --output_name ./output