Metric input shape

[Johanmkr]

We need to decide on the shape of y_pred and y_true that goes into the forward function of each metric class.

This is also discussed in #54

[hzavadil98]

y_pred - NxC, y_true - Nx1. The whole main is set up that way. #52

[Seilmast]

As mentioned by @hzavadil98 and in #52, probably use the raw vectors/matrices in the train validation loop
That is
y_true with the shape Nx1
y_pred with the shape NxC
Where N is the batchsize and C is the number of classes.
Y_pred is also the raw output of your model. No argmax or softmax before passing it to the metric classes.