Issue 216: end-to-end CI

[SamuelBrand1]

This PR closes #216.

Steps:

• hewr function to generate count data from times series of exponential rates $r_t$.
• pipeline function to use above to create dummy data with this schema https://gist.github.com/damonbayer/63bfeab36514806896cd06dc0235d1ea @damonbayer
• gh actions basically as per (closed) Draft: Create test pipeline CI - do not merge #172

This PR adds a CI which installs relevant deps, then runs a test script which first creates fake test data from an exponentially mean-varying Poisson process and then runs the forecasting pipeline script. The target pathogen is covid-19 and the target state is "CA" (chosen arbitrarily).

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 8.17%. Comparing base (9884204) to head (8314ba1).
Report is 1 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff            @@
##            main    #247      +/-   ##
========================================
+ Coverage   7.96%   8.17%   +0.20%     
========================================
  Files         20      21       +1     
  Lines       1343    1346       +3     
========================================
+ Hits         107     110       +3     
  Misses      1236    1236              

Flag	Coverage Δ
hewr	22.53% <100.00%> (+0.51%)	⬆️
pipelines	0.00% <ø> (ø)
pyrenew_hew	1.91% <ø> (ø)

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

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[SamuelBrand1]

Blocked by an error

raise ValueError("Data frame appears to have date gaps")

Which is unexpected because the data is generated in sequence.

I'll parse this tomorrow.

[SamuelBrand1]

That error was due to differing names of disease across different data sets. Onto creating eval data.

[SamuelBrand1]

Through the eval data and into generate_epiweekly

Traceback (most recent call last):
File "/Users/samandfi/GitHub/CFA/pyrenew-hew/pipelines/forecast_state.py", line 499, in
main(**vars(args))
File "/Users/samandfi/GitHub/CFA/pyrenew-hew/pipelines/forecast_state.py", line 299, in main
generate_epiweekly(model_run_dir)
File "/Users/samandfi/GitHub/CFA/pyrenew-hew/pipelines/forecast_state.py", line 53, in generate_epiweekly
raise RuntimeError(f"generate_epiweekly: {result.stderr}")
RuntimeError: generate_epiweekly: b''

This was due to the pipeline wanting to be run from root dir

[SamuelBrand1]

pipeline now runs locally

[SamuelBrand1]

This seems to be running but noting that without the __pycache__/ dir it takes a while to compile.

[SamuelBrand1]

I've tried more "minimalistic" settings e.g. fewer look back points, but occasionally you get fails due to bad sampling propagating something unexpected along the pipeline. At this moment I'm using 90 days data for inference and drawing a single 500/500 chain.

Exploring the fastest option that 1) doesn't fail stochastically, 2) still covers the pipeline sufficiently to avoid false CI passes would be a good issue.

[SamuelBrand1]

I think this is now erroring because epipredict and epiprocess are not loaded.

[damonbayer]

@SamuelBrand1 After I added the quarto action, all checks are passing. On Teams you alluded to a potential other issue, so holding off on approval for now.

[SamuelBrand1]

@SamuelBrand1 After I added the quarto action, all checks are passing. On Teams you alluded to a potential other issue, so holding off on approval for now.

Nice! Thanks @damonbayer . The other issue was using deps that need to be installed with pak e.g. epipredict; I hadn't got to the Quarto problem yet.

This should be good to review now.

[damonbayer]

Thanks @SamuelBrand1. This is a huge improvement for the repo!