Hi everyone,
Just a few comments to (hopefully) make your life easier with mini-project 1:
1. We will not grade based on absolute performance of your networks. However, it is important to discuss your results; especially if they do not match the expected trends (e.g. with the number of hidden neutrons or number of hidden layers).
2. Do not spend too much time on fine-tuning hyper parameters (even though performance depends heavily on this). Again, it is important to get an intuition and to discuss the results.
Example: It might be the case that adding a hidden layer to a network will not necessarily improve performance, even though you tried optimising hyper parameters for quite a while. Do not spend more time on fine-tuning but instead discuss your observation and confront it with your expectations.
Best regards,
the TAs
