Re: [eigen] non-linear optimization test summary

[ Thread Index | Date Index | More Archives ]

Hello List!

I was wondering what the status of the non-linear optimisation tests is. I found this old thread from 2010, which seems to suggest that the code has major issues.

Today I enabled the disabled NonLinearOptimization test routines and they still failed. However, I also implemented the failing tests into the tester of the original MINPACK routines from netlib, and the results are almost identical to the Eigen results.

I guess my question is really, why the tests are disabled instead of adapted to the new results, if the MINPACK code (which is the basis for the original implementation as far as I understand) produces very similar results.

I used gcc-4.6 and gfortran-4.6 to compile.

If anybody is interested in the results, I can post them along with the modified code that produced them.


2010/6/15 Benoit Jacob <jacob.benoit.1@xxxxxxxxx>:
> Oh wait I know why it must now be looking "better" on the dashboard: I
> un-split the NonLinear test, so instead of 4 separate test failures,
> you now see 1.
> I unsplit it because it was split in such a way that all the channels
> were compiling the same code (partly because the subtests weren't
> templated, hence were always compiled !!) making this most very test
> be compiled 16 times.
> Benoit

Mail converted by MHonArc 2.6.19+