Is my implementation of calculating training and validation loss correct? After each iteration they are nearly same

[ptrblck]

EDIT: Let’s continue the discussion about the LeNet here.