faster way to loop over atoms

[Daniel Konstantinovsky]

Hi MDAnalysis!

I'm writing a custom hydrogen bonding analysis code and need to loop a lot over atoms. Right now my code is doing this:

for atom in atomGroup:

         pos1 = atom.position

         for atom2 in atomGroup2:

                 pos2 = atom2.position

                 doStuffWithPositions(pos1, pos2)

I tried looping over indices, like this

for i in range(atomGroup.n_atoms):

         pos1 = atomGroup.positions[i]

         for j in range(atomGroup2.n_atoms):

                 pos2 = atomGroup2.positions[j]

                 doStuffWithPositions(pos1, pos2)

but it was (even) slower. I noticed with the first version that just looping over the atoms took up 1/3 of the total time in my case (doStuffWithPositions etc are optimized Cython routines)! How can I loop over atoms faster? Why is there so much overhead? I need to extract positions, ix, and other information from each hydrogen bond so I can't just loop over position arrays. I am not doing any select_atoms() inside the loop of course.

Thank you!

Dan

[Lily Wang]

Dear Dan,

A hidden gotcha with MDAnalysis is that any time you call atomGroup.positions, it returns a copy of the underlying positions array. Does the following rearrangement help speed anything up?

positions1 = atomGroup.positions

positions2 = atomGroup2.positions

for pos1 in positions1:

    for pos2 in positions2:

        doStuffWithPositions(pos1, pos2)

Cheers,

Lily

--
You received this message because you are subscribed to the Google Groups "MDnalysis discussion" group.
To unsubscribe from this group and stop receiving emails from it, send an email to mdnalysis-discus...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mdnalysis-discussion/ec3add3f-ddb5-495c-9cc0-d6c62cf0cd51n%40googlegroups.com.

[Daniel Konstantinovsky]

Hi Lily,

Thank you for the tip, I got a 60% speedup from changing it as you said!

Dan

You received this message because you are subscribed to a topic in the Google Groups "MDnalysis discussion" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/mdnalysis-discussion/LUdmxMV83Xk/unsubscribe.
To unsubscribe from this group and all its topics, send an email to mdnalysis-discus...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/mdnalysis-discussion/B724A98B-951D-4E52-BED9-9B04E0D758BB%40minium.com.au.