Sampler implementation in process

[jchodera]

Starting basic samplers implementation.

Very much in progress and untested. Just figuring out APIs right now.

[jchodera]

I've temporarily ingested ThermodynamicState and SamplerState and written a minimal functioning placeholder for MCMCSampler until we can switch over to the openmmmcmc package.

[jchodera]

Some decisions on issues #90 and #89 are needed before I can go further.

[jchodera]

This is mostly done. Once some of the other issues I raised are cleaned up, I can finish the testing and we're off to the races with a very basic functional implementation.

This currently only replicates the NCMC-geometry-NCMC scheme we have in test_sampler.py, but we can add more schemes (geometry-ncmc-geometry and geometry-ncmc) later.

[jchodera]

Added tests for the whole sampler stack. Hopefully this will work once #103 is resolved!

[jchodera]

Did a bit more cleanup and documentation. Just waiting on @juliebehr's changes to finish.

[jchodera]

Also, the API for this shouldn't be considered final. I'm more than happy to change this however we want to make things easier. This is just a first stab at the whole stack.

[pgrinaway]

This is just a first stab at the whole stack.

That's fine. I think the first stab is important because we want to get the paper out as well as get things started for collaborators. Taking a look now while the debugger takes its time...

[jchodera]

Probably easier to link to the whole new files:

• perses/samplers/samplers.py
• perses/tests/testsystems.py
• perses/tests/test_samplers.py

[jchodera]

OK, I'm now stuck waiting on #93 too.

[pgrinaway]

What do you need from #93? I can just do that now and merge. Is there any reason we can't merge this first?

[jchodera]

See test failures for master. There were some thing being cleaned up in
geometry or TopologyProposal, or so I thought?
On Feb 11, 2016 9:46 AM, "Patrick Grinaway" notifications@github.com
wrote:

What do you need from #93 #93?
I can just do that now and merge. Is there any reason we can't merge this
first?

—
Reply to this email directly or view it on GitHub
#92 (comment).

[jchodera]

Or maybe these are things I forgot to change after merging in changes from master?

[pgrinaway]

Looks like there are 3 failures in master:

1. GeometryEngine failure because of a change in the TopologyProposal API. This is fixed in my PR and shouldn't really affect this PR.

2. test_run_example fails but this is basically deprecated by this PR. (also, the failure is the same change as in 1)...perhaps the changes weren't pulled in?

3. The last one looks like an oversight on my part. I need to change the test for the SmallMoleculeSetProposalEngine, but again I don't think it should affect this PR.

[jchodera]

Should we merge this then?

[pgrinaway]

I think that would be best.

[jchodera]

Go for it then! Will have time to clean up anything I missed tonight.

[pgrinaway]

Ok, I'll do that now. Hopefully I can finish up #93 tonight as well (if there is an API change I can fix that in #93)