Don't call Model.update() from restraint get_output() methods

[sethaxen]

Most (if not all) restraints call IMP.Model.update() at the beginning of their get_output() methods. Consequently, the outputting process repeatedly updates the model unnecessarily (see here). For systems where the scoring function itself is inexpensive, these redundant calls can dominate the runtime. It seems it would be better if either IMP.pmi.output.Output updated the model once before getting outputs or if it assumed that the model was updated, leaving it to the method that triggers saving of outputs (e.g. IMP.pmi.macros.ReplicaExchange0) to update the model once before outputting.

Do the PMI architects see any fatal flaws here?

[Pellarin]

It is ok with me

Sent from my iPhone

On Jun 20, 2018, at 10:32 PM, Seth Axen notifications@github.com wrote:

Most (if not all) restraints call IMP.Model.update() at the beginning of their get_output() methods. Consequently, the outputting process repeatedly updates the model unnecessarily (see here). For systems where the scoring function itself is inexpensive, these redundant calls can dominate the runtime. It seems it would be better if either IMP.pmi.output.Output updated the model once before getting outputs or if it assumed that the model was updated, leaving it to the method that triggers saving of outputs (e.g. IMP.pmi.macros.ReplicaExchange0) to update the model once before outputting.

Do the PMI architects see any fatal flaws here?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.