156 solve like function which dispatches on epiawareproblem and epiawaremethod

[SamuelBrand1]

A draft PR for an apply_method method.

Methods

The basic idea is that each behaviour of a method, which means either a sampling or optimization technique, has an _apply_method dispatch which implements this.

An important method is EpiMethod implies doing a sequence of optimizations passing infomation to each next level, then finally to a sampling method NUTSampler.

apply_method

apply_method has two methods:

1. ingests a Turing model object and applies the epi method.
2. Ingests an EpiProblem and a data object, as well parameters to fix and/or condition on, creates an appropriate model and then apply the epi method.

The output of apply_method is a NamedTuple which returns samples as well as gens, that is generated quantities if that makes sense for the returned object from _apply_method, and model which is the model (after any conditioning or fixing of parameters).

TBD

• Naming?
• Return objects?
• More unit tests

[seabbs]

In general I think this all looks good. Can look through in more detail or happy to wait until it goes from draft to full.

[SamuelBrand1]

Its failing the

Testing NegativeBinomialError against theoretical properties`

test on CI but not locally. Any thoughts?

[SamuelBrand1]

In general I think this all looks good. Can look through in more detail or happy to wait until it goes from draft to full.

I think a detailed review would be good.

[seabbs]

@SamuelBrand1 this is ready for review I think.

There is an outstanding issue with the manypathfinder into NUTs in the getting started example that I think is unrelated but I am really not clear on.

We also need to open some new issues to improve areas of this but as long as we do that I think we should move to merge asap as this is now very large, blocking, and complicated (none of which is ideal). I've closed some of the testing gaps but there are still a few I'd suggest we port to issues. I also think there could be someone better solutions for quite a few of the changes here but again I think we should split those out and implement as independent PRs

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 68.67470% with 26 lines in your changes are missing coverage. Please review.

Project coverage is 92.17%. Comparing base (6a6a1d4) to head (a99f900).

Files	Patch %	Lines
EpiAware/src/EpiAwareBase/apply_method.jl	0.00%	6 Missing ⚠️
EpiAware/src/EpiAwareBase/functions.jl	0.00%	6 Missing ⚠️
EpiAware/src/EpiAwareBase/EpiProblem.jl	55.55%	4 Missing ⚠️
EpiAware/src/EpiAwareBase/generate_models.jl	0.00%	3 Missing ⚠️
EpiAware/src/EpiAwareUtils/turing-methods.jl	86.36%	3 Missing ⚠️
EpiAware/src/EpiAwareBase/EpiAwareObservables.jl	33.33%	2 Missing ⚠️
EpiAware/src/EpiInference/ManyPathfinder.jl	81.81%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #169      +/-   ##
==========================================
- Coverage   99.31%   92.17%   -7.14%     
==========================================
  Files          33       37       +4     
  Lines         293      358      +65     
==========================================
+ Hits          291      330      +39     
- Misses          2       28      +26     

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

[SamuelBrand1]

OK. I'm looking at this and the getting_started example

[SamuelBrand1]

First, quick takeaway. I like the restructure in the files.

[SamuelBrand1]

Looking at getting_started now

[SamuelBrand1]

Want to make me a reviewer? I'm distant enough I think due to going on holiday in the middle of this PR?

[seabbs]

I can't because you opened it. We need to just mutually agree we are okay with merging and then can do

[SamuelBrand1]

I can't because you opened it. We need to just mutually agree we are okay with merging and then can do

I'm good. I approve.