Choose the same weight and number of iterations (argument maxitem) to compare the logit and probit models, so that the cverror result reflects the true strength of the models.
DROP TABLE IF EXISTS logitmodel; DROP TABLE IF EXISTS probitmodel; DROP TABLE IF EXISTS glmcvtable; SELECT * FROM CROSSVALIDATION ( ON (SELECT 1) PARTITION BY 1 Function ('glm') InputTable ('admissions_train') OutputTable ('logitmodel', 'probitmodel') InputColumns ('admitted', 'masters', 'gpa', 'stats', 'programming') CategoricalColumns ('masters', 'stats', 'programming') Link ('LOGIT', 'PROBIT') Weight ('1', '1') MaxIterNum ('25', '25') FoldNum ('5') CVParams('LINK', 'WEIGHT', 'MaxIterNum') CVTable ('glmcvtable') Metric ('AUROC') );