Many times you use XYZDecorator to get the initial Vector3D. And the results of the Rotations/Transformations are stored back in particles - so the double precision will be lost in the optimization process. Can one change FloatKey to be really "double" in compilation time using a flag ? On Jan 30, 2009, at 7:13 AM, Daniel Russel wrote:
> I was just thinking about that. Since the Vector3D (and rotation) is > basically a computational convenience class which stores > intermediates in computations rather than being used to store data > in particles, it should probably use doubles. Easy enough to change. > > > On Jan 30, 2009, at 2:16 AM, Seung Joong Kim wrote: > >> I have another question with regard to maintaining double precision. >> >> Currently all vector related classes are defined on the basis of >> float data type (single-precision), so I'm losing accuracy whenever >> I use double data type in a Vector3D class. >> >> Do you have any update plan to implement Vector3D class in a double >> precision? I think it should be very useful for higher accuracy. >> >> Many Thanks! >> Seung Joong Kim >> >> >> Ben Webb wrote: >>> Seung Joong Kim wrote: >>> >>>> By default, Float data type is defined as "typedef float Float;" >>>> and it >>>> can have a corresponding FloatKey. >>>> >>>> But in the case of "double" data type for higher accuracy, there >>>> is no >>>> DoubleKey assigned currently. >>>> >>>> I'm just curious about it. Do you think is it necessary to define >>>> DoubleKey as well? >>>> or is it safe to use FloatKey instead? >>>> >>> >>> If all you want to do is to use double-precision floating point >>> data in >>> your calculations, by all means go ahead and use the "double" data >>> type. >>> We would only need a "DoubleKey" if we wanted to explicitly store >>> double-precision floating point data in the IMP Particles (but we >>> probably don't need to do that). >>> >>> Ben >>> >> >> _______________________________________________ >> 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