~~Test~~ Demo using data from WW R package

[gvegayon]

This PR does the following:

• It introduces the datasets submodule that includes test data generated by the WW team (see here)
• It includes a test asserting that the wastewater data is properly loaded. It uses polars to load the data.
  - [ ] It uses HospitalizationsModel to fit the data and compares it with the estimates obtained from the R package.
• It adds a demo using a polars dataset (with the WW data).
• Adds an example creating a weekday effect.
• Adds a padding argument to the hospital admissions and basic renewal model. The padding assumes the first padding observations are nan.

From the 🏰 Gargoyleosaurus sprint (April 12)

Part of the PR initially wanted to deal with adding a test comparing the results of the WW hospitalization model in R. Still, the plans changed as the number of codebase modifications grew substantially. Things that need to be addressed during the next sprint:

• Check the initialization of the model: The first half of NUTS estimated hospitalizations doesn't cover the data properly:

• Add Rt with damping effect (see More flexible Rt class (possibly meta) #45).
• Improve the current implementation of the weekday effect: This version of the package has only a toy example (see Day-of-week effect #14).
• Implement the latent infections with feedback (see Infection feedback in R(t) #9).

cc @dylanhmorris, @SamuelBrand1

[gvegayon]

I'm holding making progress on this one until #60 is merged.

[codecov]

Codecov Report

Attention: Patch coverage is 96.77419% with 3 lines in your changes are missing coverage. Please review.

Project coverage is 92.09%. Comparing base (fffbee0) to head (e72c000).

Files	Patch %	Lines
model/src/pyrenew/deterministic/deterministic.py	80.00%	1 Missing ⚠️
model/src/pyrenew/mcmcutils.py	94.73%	1 Missing ⚠️
model/src/pyrenew/model/hospitalizations.py	92.30%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #55      +/-   ##
==========================================
+ Coverage   90.57%   92.09%   +1.51%     
==========================================
  Files          28       33       +5     
  Lines         552      620      +68     
==========================================
+ Hits          500      571      +71     
+ Misses         52       49       -3     

Flag	Coverage Δ
unittests	92.09% <96.77%> (+1.51%)	⬆️

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.

[gvegayon]

@dylanhmorris and @AFg6K7h4fhy2, this PR is ready for review. The only problem is a test that checks the Model.plot_posterior function is able to reproduce an image. I had the good intention to check reproducibility (relevant to #81 (comment),) but the test fails. I'm not sure why. Any ideas?

[SamuelBrand1]

Some context: @gvegayon and I just re-ran with the

$$\log \mathcal{R}_t - \log \mathcal{R}_{t-1} \sim N(0., \sigma)$$

$\sigma$ value set at $\sigma = 0.1$ (large!) as opposed to $ \sigma = 0.025$, and that repairs a chunk of the posterior predictive fit; as expected by @dylanhmorris .

Obviously, that can't fix the whole problem because of how the infection -> hosp obs. delay works.

[gvegayon]

OK, @dylanhmorris, this is ready for review.

[dylanhmorris]

A few key questions, mostly looks ready @gvegayon