Figure 3From: Which Is Better: Holdout or Full-Sample Classifier Design?Expected 95% bounds for true error for experiments , , , , , and (a) to (f), resp.).Back to article page