Assert dtraj lengths are consistent with inputs

[maxentile]

• Check that the length of each discrete trajectory is the same as the length of its corresponding input trajectory
• Also remove the unused source_full object (I had added that before I realized that save_trajs writes out all coordinates, even if passed source object that only looks at a subset of coordinates while featurizing)

[sonyahanson]

Is this related to this: https://github.com/choderalab/msm-pipeline/issues/36 ?

[maxentile]

Sorry I missed replying! Yes, this is the most reasonable check I could think of for Issue 36.

Another possibility I considered after discussing this with @marscher last week is that the indices returned by hmm.sample_by_observation_probabilities are somehow mismatched with the dtraj lengths. However, I think this is unlikely. I think this is tested internally in pyemma / bhmm, and I also did a randomized test that should be very likely to fail if this is the culprit. We could add a runtime check here to be sure.

[marscher]

I would then conclude that the trajectory length is wrong. Can you send me the ~/.pyemma folder of the system running the analysis as a zip please? However this also very unlikely.

[jchodera]

@sonyahanson @maxentile ?

[sonyahanson]

Sent the files to Martin, and we are currently sending some files back and forth via email to see what we can figure out.

[sonyahanson]

From Martin:

Thank you Sonya, with the aid of these files, we were able to track down the problem. Expect a new release soon.

[sonyahanson]

Looks like this is here: https://github.com/markovmodel/PyEMMA/pull/969

[jchodera]

Awesome! Thanks for moving this along.

[marscher]

I think you can safely close this now.