JuliaBUGS v0.10

[sunxd3]

The todo item is marked as complete does not reflect the main branch, but sunxd/v0.10 branch.

Issues to Address

• CompatHelper: bump compat for JuliaSyntax to 1, (keep existing compat) #287
• Clarify imported functions can be used in BUGS model #294
• condition is doing too much: maybe separate it into condition + create sub-model for #252
• Correctness issue with MHFromPrior sampler #250
• [feature request] one-arg tempered_evaluate  #260
• Add Aqua tests #285
• Add a test for parallel and distributed sampling #118
• Rename register_primative to bugs_primative #302 (by @arnavk23)
• of for specifying parameter type #324

Refactoring

• Graph: more correctness tests
• Gibbs: improve performance and make the implementation more clear
• LogDensityProblems.logdensity_and_gradient probably can be implemented with DifferentiationInterface and thus simplify code related to ADgradientWrapper (we would still need to be compatible with the latter, but introduce a get_model function) (future)

More Tests

• Properly test AD packages:(opt for better documentation)
  • currently the story is: use ReverseDiff (compiled) for performance when evaluating with graph, Mooncake or Enzyme with the generated source because the generated source mutates. It'll be better to make this clear by printing instructions after the model compilation.

Documentation

• Update with the new Julia model macro syntax
• Use DifferentiationInterface and Mooncake in documentation
  • There should be a section on picking AD backends
• Source Generation: make clear it only work with parameter transformation and with Mooncake

Others

• Remove archive folder and find alternative home for some of the idea here

[yebai]

We should also add #285 to this milestone.

[sunxd3]

Given that #305 is huge as it is, we are not going to address all the doc related items. The documentations will be improved gradually over near future.