--weighted_eval

Switch

--weighted_eval COLUMN

Description

Intended to weight the evaluation (test-set scoring) of a prediction run by an outcome-table column, using weighted least squares.

Argument and Default Value

A column name. Default: None.

Details

Known issue — always crashes when actually triggered. RegressionPredictor.testControlCombos() calls sm.WLS(zscore(ytest), zscore(ypred), weights).fit() (statsmodels' Weighted Least Squares), but regressionPredictor.py never imports anything named sm — no import statsmodels.api as sm anywhere in the file. Verified: running --combo_test_regression --weighted_eval <column> completes the first fold's normal (unweighted) prediction successfully, then crashes with NameError: name 'sm' is not defined as soon as it tries to compute the weighted evaluation. There is currently no way to get a weighted evaluation result from this switch.

Other Switches

Required Switches:

• --nfold_test_regression

Example Commands

Not usable as of this writing — see Known issue above.