Fix import_optional_module function to return NoneType

[amrit110]

PR Type ([Feature | Fix | Documentation | Test])

Fix

Short Description

The import_optional_module currently returns None when a module cannot be imported and the error parameter is set to ignore. When we use isinstance this doesn't work, so introducing an option to return NoneType.

Tests Added

...

[fcogidi]

Instead of returning sometimes returning NoneType, would it be better to check if it's None before isinstance?

[amrit110]

Instead of returning sometimes returning NoneType, would it be better to check if it's None before isinstance?

yeah that sounds better. i wonder if the metric_dict is the only place where that needs to be done.

[fcogidi]

Instead of returning sometimes returning NoneType, would it be better to check if it's None before isinstance?

yeah that sounds better. i wonder if the metric_dict is the only place where that needs to be done.

There are a few cases of that in the models package. For example, in cyclops/models/torch_utils, there's

isinstance(X, (torch.Tensor, PackedSequence))

torch and PackedSequence are both optionally imported.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 66.66667% with 2 lines in your changes are missing coverage. Please review.

Project coverage is 75.14%. Comparing base (efb402d) to head (590b15f).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #597      +/-   ##
==========================================
- Coverage   75.15%   75.14%   -0.01%     
==========================================
  Files         128      128              
  Lines       11394    11399       +5     
==========================================
+ Hits         8563     8566       +3     
- Misses       2831     2833       +2     

Files	Coverage Δ
cyclops/utils/optional.py	100.00% <100.00%> (ø)
...clops/evaluate/metrics/experimental/metric_dict.py	92.13% <50.00%> (-0.48%)	⬇️
cyclops/models/torch_utils.py	75.60% <50.00%> (-0.65%)	⬇️