Fixed the run() method of analysis.hydrogenbonds.hbond_analysis method to use atom indices instead of ids

[bieniekmateusz]

Changes made in this Pull Request:

• The run() method of analysis.hydrogenbonds.hbond_analysis.HydrogenBondsAnalysis currently stores the ids of atoms in a hydrogen bond, rather than the indices. The documentation states, and the helper methods assume, that it is atom indices that are stored.
• We have fixed this and added test cases that check the ids and indices of atoms in a hydrogen bond.
• Also Fixes AttributeError when using MDAnalysis.analysis.hydrogenbonds.hbond_analysis with topology file without bonds information #2396

PR Checklist

• Tests?
• Docs?
• CHANGELOG updated?
• Issue raised/referenced?

Co-authored-by: @p-j-smith paul.smith@kcl.ac.uk

[codecov]

Codecov Report

Merging #2572 into develop will decrease coverage by 0.01%.
The diff coverage is 100.00%.

@@             Coverage Diff             @@
##           develop    #2572      +/-   ##
===========================================
- Coverage    90.71%   90.70%   -0.02%     
===========================================
  Files          174      174              
  Lines        23555    23554       -1     
  Branches      3073     3073              
===========================================
- Hits         21369    21365       -4     
- Misses        1565     1569       +4     
+ Partials       621      620       -1     

Impacted Files	Coverage Δ
...DAnalysis/analysis/hydrogenbonds/hbond_analysis.py	97.29% <100.00%> (+1.84%)	⬆️
coordinates/base.py	94.33% <0.00%> (-0.63%)	⬇️
util.py	88.06% <0.00%> (-0.02%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update da67789...2801f2d. Read the comment docs.

[orbeckst]

Minor issues, see comments.

Please also update CHANGELOG as this fixes an issue.

[zemanj]

Maybe replace hasattr(u, 'bonds') by hasattr(u._topology, 'bonds') for performance reasons (see comment below). Since I don't know how long the analysis usually takes (per frame), I don't know if this is an issue. So I leave the decision up to you. If you decide to follow my suggestion, the reason should be commented in the code.

[richardjgowers]

The bonds lookup is because it actually calculates them too right? I think that example is why I wanted to make Topology first class, so you could do “‘bonds’ in u.topology”, else there’s not a clean/canonical way to check for existence of attributes.

…

On Sun, Mar 8, 2020 at 21:14, Johannes Zeman ***@***.***> wrote: ***@***.**** requested changes on this pull request. Maybe replace hasattr(u, 'bonds') by hasattr(u._topology, 'bonds') for performance reasons (see comment). Since I don't know how long the analysis usually takes (per frame), I don't know if this is an issue. So I leave the decision up to you. If you decide to follow my suggestion, the reason should be commented in the code. ------------------------------ In package/MDAnalysis/analysis/hydrogenbonds/hbond_analysis.py <#2572 (comment)> : > @@ -408,7 +409,7 @@ def _get_dh_pairs(self): # If donors_sel is not provided, use topology to find d-h pairs if not self.donors_sel: if not (hasattr(self.u, 'bonds') and len(self.u.bonds) != 0): I also posted <#2396 (comment)> this in the related issue for reference: hasattr(u, 'bonds') is an *incredibly slow* test because it involves a lot of object instantiations under the hood. hasattr(u._topology, 'bonds') is *much* faster: import MDAnalysis as mda from MDAnalysis.tests.datafiles import GRO, TPR u = mda.Universe(GRO) # universe *without* bonds %timeit hasattr(u, 'bonds') # 758 µs ± 4.33 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each) %timeit hasattr(u._topology, 'bonds') # 216 ns ± 2.88 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each) u = mda.Universe(TPR) # universe *with* bonds %timeit hasattr(u, 'bonds') # 88.8 ms ± 524 µs per loop (mean ± std. dev. of 7 runs, 10 loops each) <== AAARGH! %timeit hasattr(u._topology, 'bonds') # 86.4 ns ± 0.266 ns per loop (mean ± std. dev. of 7 runs, 10000000 loops each) You see that in the latter example, hasattr(u, 'bonds') is *a million times slower* than hasattr(u._topology, 'bonds'). This may be irrelevant for a setup routine that is run only once, but when being called in a loop, hasattr(u, 'bonds') is a nice way of very thoroughly killing performance. 😁 — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#2572?email_source=notifications&email_token=ACGSGB5L4LVQ27RQ5Y6P7DTRGQDEPA5CNFSM4K7F64K2YY3PNVWWK3TUL52HS4DFWFIHK3DMKJSXC5LFON2FEZLWNFSXPKTDN5WW2ZLOORPWSZGOCYNNRQA#pullrequestreview-370858176>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACGSGB7WAQRDWCWR4RY5NXTRGQDEPANCNFSM4K7F64KQ> .

[orbeckst]

@zemanj are you happy with the changes? I'll leave it to you to shepherd the PR to the final merge.

[zemanj]

Looks awesome, just found some minor thingies.
Your tests are really neat, especially the ASCII art 😁👍

[zemanj]

Thank you @bieniekmateusz @p-j-smith!

[bieniekmateusz]

@zemanj are you happy with the changes? I'll leave it to you to shepherd the PR to the final merge.

Thanks @zemanj, would you mind merging it when you get a chance? Thanks, Mat