Keren Lasker wrote: > Ben - I might be missing something here but the centroid changes at > each optimization step - so this means that the particle will converge > slower in the O(N) suggestion than in the O(N^2) suggestion - since > the centroid of the previous iteration will hold it back. > The score state is updated before the restraints are evaluated, so it will always have the position of the current centroid.
> On Dec 3, 2008, at 7:36 PM, Ben Webb wrote: > > >> Friedrich Foerster wrote: >> >>> for the 26s project i would like to use a maximum diameter restraint >>> on the shape of proteins. so it'd be a restraint similar to the >>> connectivity restraint: all distances in a protein are evaluated and >>> if the largest distance is above a specified threshold, a harmonic >>> restraint would be applied on the respective pair. >>> >> The solution already proposed by Daniel looks perfect to me as a >> literal >> solution to your problem, but is that really what you want to do: >> constrain every particle pair in the protein to less than the >> diameter? >> If you have a lot of particles, that's going to become expensive >> rather >> quickly (O(N^2), obviously). Another option to consider would be to >> constrain every particle to be less than the radius from the protein >> center. That would give you far fewer restraints and scale as O(N). >> You >> could do this with a bunch of DistanceToSingletonScores if the protein >> center is at a fixed point, or use a GravityCenterScoreState to >> stick a >> particle at the center of gravity of your protein, and then use >> regular >> distance restraints between each particle and that center. >> >> Ben >> -- >> ben@salilab.org http://salilab.org/~ben/ >> "It is a capital mistake to theorize before one has data." >> - Sir Arthur Conan Doyle >> _______________________________________________ >> IMP-dev mailing list >> IMP-dev@salilab.org >> https://salilab.org/mailman/listinfo/imp-dev >> > > _______________________________________________ > IMP-dev mailing list > IMP-dev@salilab.org > https://salilab.org/mailman/listinfo/imp-dev >