Print accuracy on retain and forget sets instead of the training set

[bkowshik]

In the context of the Unlearning Challenge, printing accuracy for the retain and forget sets is more appropriate instead of printing the accuracy for the overall Training set. For now, I have just modified the code part and have not run the Jupyter notebook which updates the output cells too.

[hyunjunekim]

How do you think about that printing orders are Retain -> Test -> Forget? Becuase Retain and Test should be kept and Forget should be decreased.