Obsolete "torch.autograd.Variable" in deeprank.learn.modelGenerator

[DarioMarzella]

Is your feature request related to a problem? Please describe. In deeprank.learn.modelGenerator (and, sequentially, in all generated models like deeprank.lean.model3d and deeprank.lean.model2d) in the _get_conv_output function, the method torch.autograd.Variable is used to keep track of the gradients. This method is outdated.

Describe alternatives you've considered torch.autograd.Variable could be replaced by simply setting 'requires_grad = True' for the output tensor