Add squeeze transform

[vpratz]

Very basic transform, just the inverse of expand_dims.
Fixes #479.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Files with missing lines	Coverage Δ
bayesflow/adapters/adapter.py	84.68% <100.00%> (+0.40%)	⬆️
bayesflow/adapters/transforms/__init__.py	100.00% <100.00%> (ø)
bayesflow/adapters/transforms/squeeze.py	100.00% <100.00%> (ø)

[stefanradev93]

Looks good! I would only change the example from:

sq = bf.adapters.transforms.Squeeze(axis=1)

to

sq = bf.adapters.transforms.Squeeze(axis=-1)

For these transforms, we should discourage the use of absolute axis when applied to inference targets, since these change dimensionality between training and inference (e.g., inference adds a sample dim and thus absolute axis index changes meaning).

[vpratz]

Thanks for taking a look. Good point, do we want to warn when we encounter non-negative values?

[stefanradev93]

It may be a bit too intense, since we should ideally only warn if the variables change shape.