Validation during fitting in Keras

Question

How does Keras' fit function calculate when validation set/validation split are NOT defined (I understand that the default values are None/0.0 respectively, so being not defined is practically default)? One always need a reference set to evaluate model performance...

Answer 1

If you don't define the validation set/ validation split for your model, then it has no way to check for it's performance because you have not provided anything to the model on which it can validate its performance. In this case, the model will run through training examples, will learn a way to minimize the cost function as per your code and that's it. You get a model which has learned a hypothesis from your data but how good the model is, it can only be checked by making predictions for the test set.