Re: [eigen] New Levenberg Marquardt stuff

[Thomas Capricelli <orzel@xxxxxxxxxxxxxxx>]

Il 07/12/2012 22:16, Sameer Agarwal ha scritto:
> I am curious as to what was your evaluation criterion for various
> algorithms, and 
> what were the ways in which minpack was the best.
> I am asking because if  ceres is missing something along the lines of
> performance, numerical stability and or robustness we would like to
> fix it.

* written in c/c++
* heavily tested/validated/proof-read, as in "used in lot of places
since long ago"
* accuracy
* speed

By order of decreasing importance, but minpack was better in all of them.

Then again, i didn't know ceres by the time. I still dont know much
about it to be honest. It was announced around may 2012, right ? The
eigen LM code was written in 2009...

Thomas

-- 
Thomas Capricelli <orzel@xxxxxxxxxxxxxxx>
http://www.freehackers.org/thomas/