Train model on new dataset of increased size