> My biggest concern is the runaway process that used up 26 GB of memory before > I killed it. The molecules that take a very long time to canonicalize are > also problematic, but with the "timeout" feature, they're not deadly. We > should verify that structures like ferrocenes, and buckyballs are handled in > a reasonable amount of time since they're so common.
I will try to verify that the process was runaway on that one particular SMILES, and if so, what's causing the allocations. Apple's "Instruments" debugging feature is very useful, since you can follow the allocations over time. You mentioned a timeout for ferrocenes and buckyballs. With graphene and nanotubes and buckyballs, most of the time is spent in the Kekule assignment. There's a timeout when we attempt to find the largest fused ring system, but not for the recursive expand_kekule function. This can consume a lot of stack space and time. Do you think there's a good way to insert a timeout there? Obviously the recursion happens if you can't find a good bond assignment. When do we say timeout? > PubChem is also a useful test case. It's much larger but tends to have a lot > of redundancy because they include things like Zinc (which has multiple > automatically-generated stereoisomers for each of the input molecules) as > well as some "virtual libraries" (from suppliers who say they have in-stock > compounds but actually have never synthesized them). I've been using ZINC, since I can get it as SDF and SMI. There seem to be a few cases with SMI that don't pop up with SDF. TJ seems to have been using PubChem. So I think between the three test sets, we've got a good combination for a release criteria. As you say, the worst category is the time/memory hogs. So let's start there. Any ideas about limiting kekulize.cpp? -Geoff ------------------------------------------------------------------------------ Beautiful is writing same markup. Internet Explorer 9 supports standards for HTML5, CSS3, SVG 1.1, ECMAScript5, and DOM L2 & L3. Spend less time writing and rewriting code and more time creating great experiences on the web. Be a part of the beta today. http://p.sf.net/sfu/beautyoftheweb _______________________________________________ OpenBabel-Devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/openbabel-devel
