pdb2pqr runs not deterministic

[electrostatics]

pdb2pqr runs not deterministic

[keith923]

@kmonson, can you comment? Do you know off the top of your head where this is happening?

[tonigi]

In this branch I tried to fix it with liberal use of OrderedDicts (4bad162a6c03d79e109d6e9c9d5b69312bb59881).

[kmonson]

I'm not sure where the problem is in this case. We try to be deterministic everywhere it matters. I know that for debumping we use the same order. In the graphcut code in pdb2pka order doesn't matter if the algorithm is implemented correctly.

I could see a case (based on what changed in @tonigi's branch) where the hbond optimization could happen out of order and cause a problem. The algorithm checks each half of a potential bond and sometime the results of the first half causes the second half to be skipped. I was under the impression that it was symmetrical, but that may not be the case.

It would certainly explain the one pdb2pka test case we saw that was platform dependent.

[keith923]

It seems like this must have been a problem even before @tonigi ported the code to Python 3. Right? I mean, dict ordering has never been guaranteed. At least not that I've seen.

[tonigi]

Yes. Python3 only made it easier to spot because it enforces randomization between runs (and not just between platforms).

[tonigi]

By the way, Py 3.6 again changed the behaviour of dictionaries, which are now ordered . Still, it is probably a good idea to be explicit in the use of OrderedDict if the goal is to be deterministic.

[sobolevnrm]

Thanks!

…

On Thu, Apr 6, 2017 at 2:51 AM, Toni G ***@***.***> wrote: By the way, Py 3.6 again changed the behaviour of dictionaries, which are now ordered <http://stackoverflow.com/questions/39980323/dictionaries-are-ordered-in-cpython-3-6> . Still, it is probably a good idea to be explicit in the use of OrderedDict if the goal is to be deterministic. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <https://github.com/Electrostatics/apbs-pdb2pqr/issues/439#issuecomment-292124798>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AB1_2FFmu-eSCZo3-vFUcqRXOx9k3m6-ks5rtLWKgaJpZM4IgDoe> .

[sobolevnrm]

@intendo -- can you see if Electrostatics/apbs-pdb2pqr@4bad162 can be merged?

[sobolevnrm]

The -R option is no longer supported so I am closing this issue.