Test vs validation split in siamese network

In notebook C1_W1_Lab_3_siamese, we are reporting Test Accuracy using ts_pairs split which is also used as Validation data during model training. Even though we are not using this data for any hyperparameter tuning, we could have done that. So, shouldn’t we list the accuracy as Validation Accuracy? If we want to report Test/Generalization accuracy, should we use a separate test split from the original dataset?

1 Like

In this case they have decided to have only 2 splits train and test, I think they want to convey the overall process and they are not focused on model performance on this lab. But yes if you have 3 splits then you can also use val split for improving performance.