Naive approach for IID potential evaluation for score estimators

[Kartik-Sama]

Addresses the issue #1450

Used Effective Sample Score (ESS) to evaluate the log_probs returned by estimated posterior. This way log_probs are being checked both for single observation and iid observations case through the test added via the parameter - iid_batch_size

[codecov]

Codecov Report

Attention: Patch coverage is 73.07692% with 7 lines in your changes missing coverage. Please review.

Project coverage is 79.31%. Comparing base (b5b4790) to head (f0da7e2).
Report is 9 commits behind head on main.

Files with missing lines	Patch %	Lines
sbi/inference/potentials/score_based_potential.py	70.00%	6 Missing ⚠️
sbi/inference/posteriors/score_posterior.py	83.33%	1 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             main    #1508       +/-   ##
===========================================
- Coverage   89.70%   79.31%   -10.40%     
===========================================
  Files         122      125        +3     
  Lines        9394     9828      +434     
===========================================
- Hits         8427     7795      -632     
- Misses        967     2033     +1066     

Flag	Coverage Δ
unittests	79.31% <73.07%> (-10.40%)	⬇️

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

Files with missing lines	Coverage Δ
sbi/inference/posteriors/score_posterior.py	80.95% <83.33%> (-11.13%)	⬇️
sbi/inference/potentials/score_based_potential.py	81.35% <70.00%> (-15.71%)	⬇️

... and 37 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[manuelgloeckler]

Please revert this change. Its unrelated, instead just import in the tests from sbi.simulators.linear_gaussian import ...

[manuelgloeckler]

For these two lines I would prefer not introduce a new function with a big docstring (except if its reused several times and then it should be in sbi.utils.

I would just move the calculations in the test above and add a one sentence inline comment.

[manuelgloeckler]

This test will be quite expensive now as one would have to retrain on each combination.

Please:

• move the test to linearGaussian_npse_test.py
• similar to test_npse_iid_inference use the fixture npse_trained_model (you can skip all the posteriors which are have a uniform prior

This fixture is only trained once and the used by all tests, avoiding this overhead.

[manuelgloeckler]

Thanks for implementing looks great.

Just a few changes i.e. moving the tests and using corresponding pytest fixture needs to be done.

[manuelgloeckler]

Thanks, great effort. Looks good.

This is done, but I will block merging this for now as it will have conflicts with #1497.