More testing (plus bug fix)

[afmagee42]

This PR's scope has increased slightly having found the below bug in matrix transposition. It now includes:

• Making pytest run on PRs
• A bug fix to matrix transposition in ngm.get_R()
• A 2x2 test against Keeling and Rohani (passes with bug fix)
• An additional two correctness tests in the spirit of NGM model demo #8, both of which lean on the newly-added ngm.run_ngm() which repeatedly applies the NGM (per slide 24 here)
  • test_one_step() tests that applying one step of the NGM from the nominal "stationary" distribution:
    1. Does not change the distribution
    2. Grows by a factor of the eigen-value-derived $R$
  • test_eigenvectors() plays out the NGM from one case for many (200) steps and compares to the eigenvector
• Updates to two existing tests
  • test_vax_beta is a correctness test which had the matrix transposed the same way as get_R
  • test_simulate is a consistency-of-output test, so I re-ran the code and updated the outputs

[swo]

I think this might be the problem: https://github.com/cdcent/cfa-ngm-widget/blob/main/ngm/__init__.py#L67

Should be?

return np.transpose(beta) * (s_i / n.sum()) * s_vax

[afmagee42]

That does in fact look suspicious. Checking numpy's broadcasting, that looks like it goes into columns, so

>>> np.ones((2,2,)) * np.array([0.25, 0.75])
array([[0.25, 0.75],
       [0.25, 0.75]])

But our docs say it should go into rows https://github.com/cdcent/cfa-ngm-widget/blob/f16d72085fcac4fe4f9af37733e28b2f4fd8750c/docs/approach.md?plain=1#L27, https://github.com/cdcent/cfa-ngm-widget/blob/f16d72085fcac4fe4f9af37733e28b2f4fd8750c/docs/approach.md?plain=1#L43

[afmagee42]

But I disagree on the solution, you need to transpose it back again at the end.

return (beta.T * ((s_i / n.sum()) * s_vax)).T

[afmagee42]

As of implementing my version in 8e6d602, we pass the KR test and fail several others (test_vax_beta, test_simulate)

[afmagee42]

We bootstrapped ourselves to the transposed matrix, though, so that's not entirely surprising?

[afmagee42]

Okay @swo, @paigemiller, @dinacmistry, this is ready for a (second) look