Adding Pairs to Subsets that Have Validation Pairs but No Training Pairs

An untrained subset is a subset that has not been trained with any training examples, because one or more record pairs for that subset are present only in the Validation dataset.

Some subsets that do not have any pairs in the training dataset might still be reasonably trained. This is because the model is being trained with some automatically generated examples that belong to these subsets. Nevertheless, it is recommended to have some training record pairs for all subsets that are present in the Validation dataset.