Inner Loop Algorithm

Use this feature to specify the method to evaluate the learning curves within each inner loop.

The learning curve algorithm operates on successively larger portions of the original data. Full data is the portion on which the current point of the learning curve is being fit. The remaining part of the original data is the test set.

When full data is chosen, one can optionally perform cross-validation by partitioning it several times into training and test sets. The former are the subsets and the latter are the hold-out sets.

Available methods for evaluating the learning curves are described in the table below:

Method	This method is useful when you...
Models fit on full data (without test set), evaluate on test set	...want to evaluate results directly on a fixed test set.
Cross validation on subsets, evaluate on test set	...want to determine how well inner cross validationperforms on a test set.
Cross validation on subsets, evaluate on corresponding holdout sets	...do not have a pre-defined test set.
Cross validation on subsets, evaluate on both test and hold-out sets	...want to compare results for a test set versus those from the cross validation holdout sets.

To Specify the Method Used to Evaluate Learning Curves:

8

Select the desired method by clicking within the appropriate radio button.