Graphical Model w/ EP Output Structure

[Jammy2211]

The following script runs a simple graphical model using EP:

https://github.com/Jammy2211/autofit_workspace_test/blob/main/graphical/expectation_propagation.py

The output of the results is as follows:

Currently, it is unclear what result corresponds to what factor, and what factors were refit, and the order of those refits.

Folder Structure

We should add an input variable folder to AnalysisFactor that add the folder name to its output path, for a more concise API (we should not use the unique_tag of a search, as this will create nasty conflicts with paths):

dynesty = af.DynestyStatic(
    path_prefix=path.join("graphical"),
    name="expectation_propagation",
    nlive=100,
    sample="rwalk",
)

analysis_factor_0 = g.AnalysisFactor(
    prior_model=prior_model_0, analysis=analysis_0, optimiser=dynesty, folder="dataset_name_0"
)
analysis_factor_1 = g.AnalysisFactor(
    prior_model=prior_model_1, analysis=analysis_1, optimiser=dynesty, folder="dataset_name_1"
)
analysis_factor_2 = g.AnalysisFactor(
    prior_model=prior_model_2, analysis=analysis_2, optimiser=dynesty, folder="dataset_name_2"
)

Now each AnalysisFactor has a folder we can think about how to structure the output results. First, I think each set of results should be in the folder corresponding to its folder=dataset_name_# (based on how paths work I think this is actually what will happens as soon as the search :

expectation_propagation
   -> dataset_name_0
   -> dataset_name_1
   -> dataset_name_2

If the AnalyisFactor's search does not have a folder , we should instead use an index with some generic folder names:

expectation_propagation
   -> factor_0
   -> factor_1
   -> factor_2

Factor Optimization Order

After adding folders, it is still not clear which unique tag corresponds to which optimiztion of the factor

expectation_propagation
   -> factor_0
         sdijdfiogjioudfnguidfshgfsd
         sdlkf;dlf;aktojopjfophgkokopf
   -> factor_1
        ncjkncjknckjnjkfnkjfnkjfnfkj
        sdpodpslfokfdospkfopkds
   -> factor_2
      dsoapsomdspomfopsdd
      dsfniodngfioujhoifdsj

We have two options, eiither add another folder expressing this:

expectation_propagation
   -> factor_0
      ->   optimize_0
                   sdijdfiogjioudfnguidfshgfsd
      -> optimize_1
                  sdlkf;dlf;aktojopjfophgkokopf
   -> factor_1
      ->   optimize_0
                ncjkncjknckjnjkfnkjfnkjfnfkj
      ->   optimize_1
                  sdpodpslfokfdospkfopkds
   -> factor_2
      ->   optimize_0
               dsoapsomdspomfopsdd
      ->   optimize_1
             dsfniodngfioujhoifdsj

Or append the unique tags with this index for conciseness (as the risk of making database interfacing more complicated)

expectation_propagation
   -> factor_0
         0_sdijdfiogjioudfnguidfshgfsd
         1_sdlkf;dlf;aktojopjfophgkokopf
   -> factor_1
        0_ncjkncjknckjnjkfnkjfnkjfnfkj
        1_sdpodpslfokfdospkfopkds
   -> factor_2
      0_dsoapsomdspomfopsdd
      1_dsfniodngfioujhoifdsj