research notes

If you want to build a solid model you have to follow that specific protocol of splitting your data into three sets: One for training, one for validation and one for final evaluation, which is the test set. The idea is that you train on your training data and tune your model with the results of metrics (accuracy, loss etc) that you get from your validation set. Your model doesn't "see" your vali..