Cross-fitting
A sample splitting technique that divides data into multiple folds, trains models on the remaining folds, and evaluates on the held-out fold. It restores approximate independence among contributions and improves the statistical properties of the estimator.
Timeline (last 90 days)
Cross-fitting is a sample-splitting technique used to eliminate overfitting bias; the basic procedure is to randomly divide the data into K folds, and for each fold use the remaining K-1 folds of data to train the model, then compute the prediction or estimator on the held-out fold.
When random forests were used on the main data-generating mechanism, the coverage of estimators using in-sample fitted propensity scores dropped to 0.12–0.48, severely under-covering; whereas after cross-fitting both nuisances, coverage recovered to 0.91–1.00.
When the empirical process conditions cannot be rigorously justified, cross-fitting should be applied to both nuisances simultaneously.
All Facts (3)
Cross-fitting is a sample-splitting technique used to eliminate overfitting bias; the basic procedure is to randomly divide the data into K folds, and for each fold use the remaining K-1 folds of data to train the model, then compute the prediction or estimator on the held-out fold.
50%UnverifiedWhen random forests were used on the main data-generating mechanism, the coverage of estimators using in-sample fitted propensity scores dropped to 0.12–0.48, severely under-covering; whereas after cross-fitting both nuisances, coverage recovered to 0.91–1.00.
50%UnverifiedWhen the empirical process conditions cannot be rigorously justified, cross-fitting should be applied to both nuisances simultaneously.
50%