Nested Cross Validation

Use Grid Search with K-Fold Cross Validation for two folds, then use the best model with K-Fold Cross Validation for five more folds.

So you use a nested cross validation loop to estimate the Hyperparameters and then evaluate the best model again. The resulting value can be used to compare between the best models with different model architectures.

For example to compare SVM with Decision Tree.