Raise ValueError if reducing axis dimension is 0

[old-school-kid]

Fixes the issue #16941

The solution notebook

[gbaned]

@old-school-kid Can you please check @fchollet's comments and keep us posted ? Thank you!

[fchollet]

@old-school-kid the PR appears empty at this time. Do you have any update?

[old-school-kid]

@fchollet can you take a look now? TIA

[gbaned]

@old-school-kid Can you please check @fchollet's comments and keep us posted ? Thank you!

[old-school-kid]

@fchollet Sorry for the extreme delay, had some academic commitments. Added two checks each for GlobalAveragePooling and GlobalMaxPooling for 1D and 3D (with keepdims=True). Please let me know if I should add more tests.

[old-school-kid]

Also keras-team/tf-keras#471 has a similar issue when 1 dimension is empty in case of Conv Transpose layers. I guess a similar solution could work for that too?!

[old-school-kid]

@gbaned Would you please remove the awaiting response from contributor label.? TIA

[qlzh727]

@fchollet, please take a look for this PR again.

[gbaned]

Hi @fchollet Can you please review this PR ? Thank you!

[old-school-kid]

Hi @fchollet Could you please give it a look? TIA

[old-school-kid]

Sorry to bug again @fchollet ! But would you please take a look at the PR? TIA

[jbischof]

cc @fchollet

[gbaned]

Hi @fchollet Can you please review this PR ? Thank you!