Updated PSA module

[sseyler]

Changes to the PSA module now reflect proposed features in the PSA paper (in review). These additions include:

• Three new distance functions: two based on averages (hausdorff_avg and hausdorff_wavg), one based on computing nearest neighbor pairs and distances (hausdorff_neighbors)
• Two global functions in the psa module for computing msd (get_msd_matrix, used by all distance functions) and the dimensions of numpy arrays of trajectories (get_coord_axes)
• A new plotting method plot_annotated_heatmap to generate annotated heat maps (in addition to the heat map-dendrograms)
• A new analysis method in the PSA class class to perform nearest neighbors/Hausdorff pairs analysis
• A new plotting method in the PSA class class to generate plots of nearest neighbors distances as a function of (normalized) frame progress
• A class PSAPair to facilitate the storage and extraction of heterogeneously sized data generated by nearest neighbors/Hausdorff pairs analysis
• Additional getter methods in PSA class
• Various changes/tweaks to documentation

[richardjgowers]

Heya

Add changes to CHANGELOG please

[kain88-de]

some tests would be nice as well.

[dotsdl]

@kain88-de I agree. I don't think this analysis module has any tests yet, but they should be added before this is merged. @sseyler, can you add a test class to test_analysis.py that ensure the class returns expected values for, say, an ADK transition compared to itself? It's not ideal, but I think to test this more strictly we'd need at least two different ADK transitions in the test data. Perhaps you could add a second one? No more than 10 frames or so for space considerations.

[orbeckst]

@sseyler , you can add a second AdK DIMS trajectory to the test data, they are small.

[sseyler]

@kain88-de @dotsdl Most definitely. Minimal tests will be added very soon so that it can be merged; as planned, a full test suite will be added in the near future.

[richardjgowers]

You might need to put seaborn in as an optional dependency here and maybe document that it's potentially needed.

[kain88-de]

Does seaborn change the matplotlib style also when imported in a module? If yes that might lead to surprises for some users.

[sseyler]

@richardjgowers Got it.

@kain88-de seaborn does change the default matplotlib style, unless one uses import seaborn.apionly. This was what I did in the plot_annotated_heatmap method, but not the plot_nearest_neighbors method. I am going to change this so that plot_nearest_neighbors behaves the same way as plot_annotated_heatmap and doesn't catch users by surprise.

[dotsdl]

I'd recommend using seaborn.apionly in general, since this means a user's chosen style isn't disrupted by using this method.

[sseyler]

@kain88-de @richardjgowers I'm happy to convert to numpy-style docs in the near future. Just trying to get a working version of this module out so those who come across it in the associated paper can start tinkering.

[richardjgowers]

@sseyler just tell me when you're happy with it and I'll merge and release 0.11.1 to pypi

[sseyler]

@richardjgowers I will be rolling out periodic enhancements (and fixes) in the near future, but I'm happy with it as it currently stands (once I get the final Travis CI build to not fail).

[sseyler]

@richardjgowers The tests appear to be failing, but neither I nor @dotsdl can figure out exactly why. It appears it might have something to do with coverage checks.

[richardjgowers]

Coverage got bumped to version 4.0 in the last few days. Probably gotta rename something in coveragerc (or install the old version in travis)

e: fixed it, so stuff should/might/could/won't (delete according to optimism level) work