For j in range(16) >>> 'Tensor' object is not callable

1

I am a newbe and i am stumped … probably simply because i am missing something obvious.

Trying to show images of the kernels in a cnn layer.
The following works fine:

kernels = model.cnn1.weights.detach().clone().cpu().view(-1, 3, 3)

#max = torch.max(kernels)
#min = torch.min(kernels)
#range = max - min
#kernels = (kernels - min) / range

kernels = kernels.numpy()

fig, ax = plt.subplots(4, 4)


row = -1
for j in range(16):
  if j%4 == 0:
    row += 1
    col = 0
  ax[row][col].imshow(kernels[j]).set_cmap('Greys')
  ax[row][col].axis('off')
  col += 1

Now if i uncomment the “scaling” lines in the code above. My Jupiter notebook tosses the following:

TypeError                                 Traceback (most recent call last)
 in ()
     17 
     18 row = -1
---> 19 for j in range(16):
     20   if j%4 == 0:
     21     row += 1

TypeError: 'Tensor' object is not callable

Confusing as j there are no tensors on line 19.
printing kernels prior to the loop shows are “scaled” just fine.

Appreciate any assistance to shed light upon my ignorance.

2

Hi Marc!

This is really a python issue – not specific to pytorch.

range = max - min

This line (when uncommented) sets range to (refer to)
max - min, which, in this case, is a pytorch tensor.

Then when you try to execute

for j in range(16):

range no longer refers to the built-in python “range” function,
and the thing it does refer to (a pytorch tensor) is, indeed, not
callable.

Easy fix: Change the name of your max - min variable to
something else. Perhaps my_max_min_range_not_pythons?

The following pure python script shows this general behavior:

range = 'Marc'
print (range)
for  i in range (5):
    print (i)

(I consider this kind of thing to be a weakness in python. Who
has time to worry about tripping over such things?)

Best regards.

K. Frank

3

thanks Frank! … python is a new language to me … really appreciate the assist … not a typical behavior of all the other languages i have encountered :blush: