[Objcryst-foxx] Fox v1.2 testing
Brought to you by:
vincefn
From: Vincent Favre-N. <vi...@us...> - 2002-07-01 21:35:32
|
Hi, I guess it's time to prepare for a new Fow release. I have fixed a nasty bug which was introduced 2 months ago and caused random crashes (especially under windows...). I also removed a bug when removing a scatterer an re-launching an optimization (thanks Lachlan for the report). Finally, the 3D crystal structure update during optimization should now work under win NT and above. Also, you will notice two new options in global optimization objects. The first is to save the evolution of all parameters during optimisation ; this is of course very slow and only for testing purposes, so don't select it (it won't be in the release, but if you want to have fun doing statistics on the evolution of parameters to see how the drunkard algorithm evolves...). The second is to save regularly the best configuration as a full xml file (every 10', 1h, 24h), in case you are making a very long run (did you use the sin(theta)/lambda limit to run faster ? Hmm??). The auto saved files are put in the directory where you launched Fox (in the Fox directory if you launched it from its icon). You can either update from the current CVS, or download a precompiled version from ftp://ftp.unige.ch/pub/soft/crystal/fox/test/. In the same directory there are two textured sample data (brucite & tartrate). There are two things I still have to work on before release, (i) the manual, and (ii) improving the convergence rate when searching texture parameters (for the tartrate, the correct solution is rarely found; even if the texture is extreme in that case it should be improved). If you have data with relatively strong preferred orientation and average complexity (something which is typically solved in about 100.00 moves, more than brucite), I'm looking for such data for tests ! If you generated a xml file with the previous version and are using texture, I'd suggest right-clicking on the "March Coeff" name and reduce the limits to [0.1;10], which will be more reasonable and help in the search (it's the default now). Cheers, Vincent -- Vincent Favre-Nicolin ESRF- Grenoble, France http://v.favrenicolin.free.fr ObjCryst & Fox : http://objcryst.sourceforge.net |