I haven't removed pandas and scipy bits from the QC model, because we want to test that my changes don't actually change the model at all. But this leaves us with still needing scipy and pandas at runtime, because the QC models are loaded during import stdpopsim. :(

Actually, I don't think we should be loading the QC models in stdpopsim/__init__.py, but I'm not sure the current scope of how these are used. We could instead import stdpopsim.qc from tests/__init__.py, which will ensure the QC models are used in testing. But the cli also uses QC models to only show help output for models that are QCed warn users the model hasn't been QCed. I didn't check elsewhere to see if QC models might be used at runtime.

Codecov upload failures. 😩

Thanks @grahamgower - agreed with all the above!

Me too - seems tricky!

Codecov Report

Base: 99.94% // Head: 99.94% // Decreases project coverage by -0.00% ⚠️

Coverage data is based on head (22075bf) compared to base (d2e4371).
Patch coverage: 100.00% of modified lines in pull request are covered.

@@            Coverage Diff             @@
##             main    #1416      +/-   ##
==========================================
- Coverage   99.94%   99.94%   -0.01%     
==========================================
  Files         109      109              
  Lines        3820     3807      -13     
  Branches      524      529       +5     
==========================================
- Hits         3818     3805      -13     
  Misses          1        1              
  Partials        1        1              

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

We should totally not be importing the qc models at runtime unless specifically requested - would be happy to help with workarounds to get them out of the import stdpopsim code flow

We should totally not be importing the qc models at runtime unless specifically requested - would be happy to help with workarounds to get them out of the import stdpopsim code flow

I think I'd rather have that as a separate PR, and I've opened #1418 to discuss/track.

Maybe for this PR, you'd be happy to look at the CI runs and satisfy yourself that I haven't changed the demographic model (the QC test still passes)? I can then push another commit making the analogous change to remove scipy/pandas from the QC model.

One additional point: time python -c 'import stdpopsim' takes about 1 second; after removing pandas/scipy from the import path, this drops to 500 ms. Loading the QC models (or not) makes negligible difference to import time once pandas/scipy is removed.

This is good to go now @jeromekelleher.