Closed bvdmitri closed 2 months ago
Attention: Patch coverage is 90.36145%
with 24 lines
in your changes are missing coverage. Please review.
Project coverage is 84.51%. Comparing base (
29061e5
) to head (10c6648
).
Files | Patch % | Lines |
---|---|---|
src/inference/autoupdates.jl | 90.82% | 21 Missing :warning: |
src/model/model.jl | 66.66% | 3 Missing :warning: |
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
@bvdmitri what if we make the autoupdates macro recursive such as constraints, meta and init? Might be nice to also update prior distributions for variables hidden in some submodel.
I like the idea @wouterwln , I think we can accomplish this in a separate PR, since we really need a generic code for those
for _something_ in submodel
end
pattern. As far as I understand now its mostly copy-pasted for all macros, lets not do that again. Also, this would require to extend the VarDict
, which now looks only at the top-level.
I've also come to realize that to support this feature in submodels, we'll need a more integrated approach with GraphPPL. This is because the autoupdates requires that certain arguments remain unspecified. However, I don't believe we can create a submodel with certain arguments underspecified.
This PR reimplements the
@autoupdates
macro byThis PR fixes https://github.com/ReactiveBayes/RxInfer.jl/issues/61