May 2008 - IMP-dev - salilab.org

Ordering among score_states (and other things)
by Daniel Russel 08 May '08

08 May '08

I am ending up with various score states which have to be updated in the right order. Currently this requires that the user add them to the model in the right order and non-obvious things happen if this is not the case. I think adding an iteration counter to the update call would make this simpler (and the change could be implemented without changing individual score states much). Model maintains a counter which is incremented each time evaluate is called. It then calls ScoreState::update(int) for each score state passing the counter. This update is a method in the ScoreState base class which simply compares the passed counter to the stored counter and calls the existing update method if the counters are different. ScoreStates which depend on other score states can simply call update on their dependencies with the counter. Each ScoreState is updated exactly once and, unless there are circular dependencies, everything is updated before it is used. The same thing can be used for after_evaluate, although there is currently no need. With slightly more logic we could catch circular dependencies too. Comments?

2 3

Cleanup of attribute table
by Daniel Russel 07 May '08

07 May '08

I cleaned up and simplified the attribute table handing. You can now declare keys types anywhere (although the injunction against static initialization of keys is still there). And there is some explanation of what is going on.

1 0

Remove unused python class in test
by Daniel Russel 07 May '08

07 May '08

I removed an unused class from the test case

1 0

Limiting the size of moves with conjugate gradients
by Daniel Russel 07 May '08

07 May '08

In an attempt to decrease the frequency with which CG blew up due to stepping to far and getting floating point overflow, I added a parameter which restricts the maximum change for any attribute. It helps is my code, but is a bit of a hack.

1 0

Finer control of monte carlo moves
by Daniel Russel 07 May '08

07 May '08

Added a parameter to the montecarlo optimizer to limit the number of individual moves which are made. You can set the move probabiity and then each particle is moved with that probability. Also added some methods to track and get statistics about the number of accepted and rejected moves.

1 0

Gravity center cleanup
by Daniel Russel 07 May '08

07 May '08

Remove an unused member variable, check that the particles have the needed attributes on addition, and then don't check later.

1 0

Patches for boost 1.35
by Daniel Russel 07 May '08

07 May '08

Frank installed boost 1.35 (since nobody told him not to). The changes to support it are quite minor, just replacing some incomplete definitions with complete ones for objects used in ObjectPointers. The patch is against the current head, so is independent of the other changes.

2 1

Summary with regards to external libraries
by Daniel Russel 06 May '08

06 May '08

After a discussion with Ben it seem like the real holdups are 1) ease of commercially licensing of IMP as a whole. It is better to license something with a fast non-bonded list than without (and CGAL wants money from commercial users). The license of the package itself wouldn't be affected meaningfully by linking with QPL or LGPL or similar dependencies as it is easy enough to isolate contaminated code in particular .cpp files. Personally, I don't think we should put much effort towards the future until it arrives. As long as we abstract from the external libraries well, reimplementing the algorithms later won't be more work than writing it now (and in fact, should be easier since we have a good implementation to compare against). And wrapping it typically quite cheap to write an maintain. 2) Other people probably don't want to install another external library. Fine, as long as other people want to help develop the functionality :-) I would only propose only using libraries that are really easy to install (rpms on redhat, packages on ubuntu, and some sort of installer for mac os and windows and no more effort than "install- command --prefix=path" on the rest). My estimate is that installing a decent library isn't generally more work than than writing and maintaining a couple of lines of code. But maybe others have different weights. We also should not depend on libraries that are evolving too fast since those require the user to update frequently. 3) It is faster to test two separate runtime code paths rather than two separate compile time code paths (since the unchanged tests don't run more than once and the compilation doesn't have to be repeated). I think this is most substantial concern. The best solution I have to this so far is that we should make sure optional dependencies are independent so that simply testing with all of them and without all of them is enough. Strictly speaking, we already need tests without the dependencies (such as boost) to make sure that the installer fails properly. Any other thoughts? A couple less related ones: - We should adopt the policy of building as many things as possible at once-- e.g. impem should be built/installed along with everything else when embed is found (same way modeller support is done now). - I personally don't plan to put much effort into the grid-based nonbonded list and will use the CGAL one. With the yet-to-be-check-in cleanup, much of the code is the same and so many of the parameter tweaks will benefit both. - we need a cleaner way of adding search paths to scons. I had to go figure out what paths IMP already uses to search for boost and find one that wasn't owned by some software in order to tell Frank were to install boost (since mac ports choked while trying to build it on his machine). This is a bit silly.

1 0

More minor patches
by Daniel Russel 06 May '08

06 May '08

Since we are doing the patch game, here are some trivial ones: Add addition and multiplication by a scalar to vector3D (python support needs to be added separately). Just throw an exception on a bad cast (no error message) in the decorator. Currently catching an exception is the only way to check if a particle has the required fields or not. We may want to add a cleaner mechanism. Make sure that bonds connect two different particles.

2 2