More streamlined context caches usage

[ijpulidos]

Description

Context caches for propagation and energy computations are expected to be handled only via MultiStateSampler instance attributes. Default behavior is to use global cache. If users want to avoid cycling they are expected to specify different context caches via the instance attributes.

Since we are not serializing context cache information, resuming simulations requires users to specify context cache attributes upon resuming.

Resolves #546

Todos

• Implement feature / fix bug
• Add tests
• Update documentation as needed
• Update changelogNotable points that this PR has either accomplished or will accomplish.

Status

• Ready to go

[ijpulidos]

Hmm, the test I designed for this is failing. It seems that using context_cache._lru._n_access is not the way to test this. Any ideas how to test this behavior? I'm just basically trying to test that propagating the replicas and the energy computation are using the context caches being specified.

[ijpulidos]

I found out that ParallelTemperingSampler had its own _compute_replica_energies method that was hardcoded to use the global context cache. Fixed that and now all tests are passing locally. I'll wait for GHA workflow to give me the green flag, if so, this should be ready for revision.

[jchodera]

Let's also make sure we document energy_context_cache and sampler_context_cache as class attributes of MultiStateSampler, note that they default to the global context cache, and potentially also add an example of how to override them with your own ContextCache in the docstring example.

[ijpulidos]

@jchodera This should be ready for review, I updated the documentation as requested. Thanks!

[jchodera]

Looks really good! Thanks!

[ijpulidos]

I guess the change to main for the main branch messed up the codeclimate check. I think this has to be changed from the codeclimate side of things, @mikemhenry @jchodera can you check that please? I could also just merge it without this check using admin capabilities, in case we want to do that.

[jchodera]

Not sure I know how to fix that,---can we continue without it and have @mikemhenry look into it in a separate PR?