Don't use online analysis to set offline analysis matrix dim

[mikemhenry]

We want to make sure the _offline_analysis works. _online_analysis is experimental and can be neglected for now. We need to make sure the matrix/array dimensions include the endstates for the offline analysis.

[zhang-ivy]

I think we had discussed that this can be part of the next milestone, actually.

[mikemhenry]

Yes sorry! I made this issue as we were talking about it, I've bumped it!

[zhang-ivy]

We need to make sure the matrix/array dimensions include the endstates for the offline analysis.

Just going to elaborate on this.

The problem: Currently, we get the following error when online_analysis_interval != None and we use unsampled endstates: initial_f_k must be a 5-dimensional np array -- which is raised here on the pymbar side. The initial_f_k must be a 5-dimensional np array error comes from this line.

We basically want to avoid initializing initial_f_k (in offline analysis) with the self._last_mbar_f_k used for online analysis.

Going to assign this to @ijpulidos , but we can work together on this if needed.

[zhang-ivy]

We may also want to add to the docstring an explanation of online vs offline analysis -- i.e. both compute the free energies, but online is a more a crude estimate.

[zhang-ivy]

We'll also want to open an issue concerning adapting the online analysis to include unsampled endstates.

[zhang-ivy]

Note that we may also want to consider opening an issue for renaming self.online_analysis_interval to self.offline_analysis_interval

[ijpulidos]

Note that we may also want to consider opening an issue for renaming self.online_analysis_interval to self.offline_analysis_interval

Yes, I think this is already a TODO in the code, but I'll be opening an issue for this. The problem with this change is that it could break backwards compatibility (which we'd probably have to anyways, at some point).