Check if 'cnn_output_size' is zero in CNNembedding.

[Ankush7890]

When configuring the CNN-based embedding layer using sbi.neural_nets.embedding_nets.CNNEmbedding, setting the parameters for num_cnn_layers, kernel_size, and pool_kernel_size incorrectly may result in a vague error message, such as:

RuntimeError: Given input size: (6x4x4). Calculated output size: (6x0x0). Output size is too small.

Proposed Solution:
Since the cnn_output_size is recalculated after each layer, a CNN setup with excessive depth results in an output size of zero. This could be leveraged to alert the user about appropriate layer size configurations or to adjust other parameters accordingly.

[michaeldeistler]

Great! Thank you!!

[michaeldeistler]

Unfortunately, ruff linting is not passing:

Please run ruff format sbi and ruff check sbi --fix.

[Ankush7890]

Thank you for letting me know. Pushed ruff linting related fixes 👍 .

[michaeldeistler]

Great! Ruff is passing now, I will merge once all other tests are also through. Thanks again!

[michaeldeistler]

Unfortunately, one of the tests has failed. To reproduce:
pytest tests/embedding_net_test.py -k test_1d_and_2d_cnn_embedding_net

This gives

            cnn_output_size = get_new_cnn_output_size(cnn_output_size, conv_layer, pool)
            assert (
>               cnn_output_size[0] > 0 and cnn_output_size[1] > 0
            ), f"""CNN output size is zero at layer {ii + 1}. Either reduce
                 num_cnn_layers to {ii} or adjust the kernel_size
                 and pool_kernel_size accordingly."""
E           IndexError: tuple index out of range

Could you have a look?

[Ankush7890]

Made the fix, and confirmed with the tests. This should be okay now 👍. Sorry, missed the variability with 1D convs before.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 78.39%. Comparing base (299854e) to head (9b0018c).
Report is 9 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #1281       +/-   ##
===========================================
- Coverage   89.24%   78.39%   -10.86%     
===========================================
  Files         119      119               
  Lines        8695     8849      +154     
===========================================
- Hits         7760     6937      -823     
- Misses        935     1912      +977     

Flag	Coverage Δ
unittests	78.39% <100.00%> (-10.86%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
sbi/neural_nets/embedding_nets/cnn.py	100.00% <100.00%> (ø)

... and 35 files with indirect coverage changes

[Ankush7890]

@michaeldeistler, Just curious, does this PR needs more modification? Or you think this addition would not be much useful?

Thanks.

[michaeldeistler]

Hi there, sorry I am currently travelling and therefore was a bit slow to have another look. I just ran tests, if they pass then I will merge. Thanks a lot for the contribution!

[Ankush7890]

Ah, I see. No worries. It seems all the tests pass now. Thanks, you can merge it whenever it is possible for you.

Also, It's an amazing opensource library. Thanks for creating it.