Available in: GBM, DRF, Deep Learning, GLM, GAM, PCA, GLRM, Naïve-Bayes, K-Means, XGBoost, Isolation Forest
This option allows you to specify to score during each iteration of model training. This option is useful when used with early stopping and attempting to make early stopping reproducible. When used with early stopping, the
stopping_rounds option applies to the number of scoring iterations that H2O has performed, so regular scoring iterations of small size help control early stopping the most (though there is a speed tradeoff to scoring more often). The default is to use H2O’s assessment of a reasonable ratio of training iterations to scoring time, which often results in inconsistent scoring gaps.
This option is disabled by default.