pytorch-discontinuous-function.ipynb

Your article is a great teaching moment - but I have run into issues running your code as is - specifically during training:

starting training

IndexError Traceback (most recent call last)
in ()
47 input_var = Variable(torch.from_numpy(this_inputs))
48
---> 49 output = model.forward(input_var)
50
51 # build a PyTorch variable with the target value

in forward(self, input)
40 input = input.data.numpy()
41
---> 42 return self.single_coord_func(input[:,0], 0) + self.single_coord_func(input[:,1], 1)

in single_coord_func(self, input, coord)
24 # fold in underflow/overflow bins
25 indices = np.clip(indices, 0, self.n_bins_xy[coord] - 1)
---> 26 return torch.cat([ self.coeffs[indices[row], coord] for row in range(len(input)) ])
27
28 #----------------------------------------

in (.0)
24 # fold in underflow/overflow bins
25 indices = np.clip(indices, 0, self.n_bins_xy[coord] - 1)
---> 26 return torch.cat([ self.coeffs[indices[row], coord] for row in range(len(input)) ])
27
28 #----------------------------------------

IndexError: only integers, slices (:), ellipsis (...), None and long or byte Variables are valid indices (got numpy.float32)