This is a minor re-organisation of some tests to improve the coverage we get from CI tests. Previously the tests which exercised the nuts method of the numpyro backend were hidden behind the RUN_SLOW flag, and these were not included in CI. I've now dropped this flag.
To keep a lid on the time it takes to run the entire suite, I've reduced the number of tests which use numpyro's nuts method. This was achieved by (a) switching to running some "parameter shape" tests using only the prior algorithm, and (b) splitting out cases related to "new data" from test_marginals_and_fitted_smoke. I think we lose very little by doing this -- I think it's a good trade-off.
This is a minor re-organisation of some tests to improve the coverage we get from CI tests. Previously the tests which exercised the nuts method of the numpyro backend were hidden behind the
RUN_SLOW
flag, and these were not included in CI. I've now dropped this flag.To keep a lid on the time it takes to run the entire suite, I've reduced the number of tests which use numpyro's
nuts
method. This was achieved by (a) switching to running some "parameter shape" tests using only theprior
algorithm, and (b) splitting out cases related to "new data" fromtest_marginals_and_fitted_smoke
. I think we lose very little by doing this -- I think it's a good trade-off.Fixes #48.