Update calibration methods, add tests.

[bas-rustenburg]

MBAR calibration currently not working because it depends on context in the initialization. Will fix this at another time.

[jchodera]

MBAR calibration currently not working because it depends on context in the initialization

Can you elaborate?

[bas-rustenburg]

Sure. The calibration class that uses MBAR currently uses context as an argument in it's __init__ function. However, since we now use a compound integrator, and generate a context using that integrator, we have to define the Titration object first before we have a context.

I may be able to move this so context isn't needed in __init__. I could potentially merge it into the other calibration class.

[bas-rustenburg]

Here is the test case that shows the issue:

https://github.com/choderalab/openmm-constph/blob/master/constph/tests/test_explicit.py#L132-L137

[jchodera]

Ah, so you just need to fix up the API.

Why not just replace context with system, and take an optional platform argument?

[bas-rustenburg]

For the sake of unifying the APIs, I think with a little bit of work, we can include an mbar option as an alternative to eq9 and eq12, as opposed to fixing the MBAR class with a separate api.

The one thing in which the classes fundamentally differ -- which will be the part that takes some work --is storing all the previous states. Right now MBarCalibrationTitration use this adaptation_tracker dictionary entry to keep track of all past samples. Its __init__ requires a context because I need to extract the potential for the initial state. I would have to incorporate that into the CalibrationTitration code.

It'll take some time, and for now it isn't high priority, since we can already use SAMS on-line. MBAR was just an interesting at-line alternative, and we could just run MBAR off-line if we wanted to. Not fixing this right now shouldn't hold up our development.

[bas-rustenburg]

Added reminder to #20.

[gregoryross]

You've probably already done this, but have you compared the free energies you get with SAMS with the free energies you get with MBAR? This is an important validation test for the SAMS code.

[bas-rustenburg]

@gregoryross Very qualitatively yes, they seemed to match up, but I'd like to set up a test suite/example set that proves this.

[bas-rustenburg]

Added reminder for that to #20.