Create a f_score loss function

autograd
2

Hi,
The result of (model(x)>.5).float() is a binary tensor. Gradients cannot exist for non-continuous values. You might want to use tanh() instead (see this thread for a similar issue Step Activation Function)