As evoked in a former message, I couldn't manage to have GC optimizer working with the binary MAC OSX version (IMP 1.0). So I compiled the latest version on one of our linux machines. The compilation was easily achieved with the default parameters, though I guess it could be possible to enhance IMP speed by tweaking some magic compilation parameters… I'll see this point later.
I am thus using IMP version SVN 6211
With this version CG optimizer seem to work fine, though I always get warning messages WARNING could not find parameters for radius:1.85 WARNING Setting params for radius :1.85 WARNING could not find parameters for radius:2.275 WARNING Setting params for radius :2.275 WARNING could not find parameters for radius:2 WARNING Setting params for radius :2 WARNING could not find parameters for radius:1.7 WARNING Setting params for radius :1.7 WARNING could not find parameters for radius:2.175 WARNING Setting params for radius :2.175 WARNING could not find parameters for radius:2.06 WARNING Setting params for radius :2.06 WARNING could not find parameters for radius:1.77 WARNING Setting params for radius :1.77 WARNING could not find parameters for radius:1.9924 WARNING Setting params for radius :1.9924 WARNING could not find parameters for radius:1.8 WARNING Setting params for radius :1.8 WARNING Object "Nameless" was never used. See the IMP::Object documentation for an explanation.
But when trying to use CG in conjunction with MC ( mco.set_local_optimizer( cgo ) ), I still experience problems; sometimes it seems to pole in an infinite loop, sometimes it prematurely ends with this message :
Traceback (most recent call last): File "./fitWithRestraints.py", line 222, in <module> opti.optimize(params["optimizer"]["maxSteps"]) File "/usr/local/imp/lib64/python2.6/site-packages/IMP/__init__.py", line 1569, in optimize def optimize(self, *args): return _IMP.Optimizer_optimize(self, *args) _IMP.UsageException: Attempt to use uninitialized vector.
Ultimately, I am wondering if trying to get CG working is not a waste of time for me… I mean, maybe there are better approaches... So, If you have any comment or general advice that I should check, feel free to do so.
--Ben