Re: [eigen] some problems with building with AltiVec enabled

[ Thread Index | Date Index | More Archives ]

Quoting Gael Guennebaud <gael.guennebaud@xxxxxxxxx>:

On Fri, Aug 22, 2008 at 4:18 PM,  <jacob@xxxxxxxxxxxxxxx> wrote:
I also have numerical stability issues. This is getting annoying because
determining if a test passes or not often requires to run it several times
(each time with 10 repeats) to see if failure occurs everytime. This is
really unfortunate as it could make us ignore a real error that would be
only occasionally triggerred.

exactly. So I have a suggestion (maybe wrong): what about generating
floating point random matrices with a limited number of decimals ?
this way the loss of precision should be significantly reduced and we
should be able to reduce the tolerance thresholds to more decent
values. What do you think ? (we would do that only in the unit tests)

Sounds like an excellent idea. One clarification though (I know that you know!) here by decimals you mean binary decimals. Something like some_random_integer * 2^-8.


This message was sent using IMP, the Internet Messaging Program.

Mail converted by MHonArc 2.6.19+