Evaluation: Support variable training size in evaluation

[ggianna]

The aim is to be able to generate a diagram showing if we learn better by adding more instances in the training set.