Non-linear least squares?

Oct 27, 2013 at 5:45 PM
Hi,

Great project!

Are there any plans to add any non-linear least squares fitters? Perhaps something based on Levenberg-Marquardt / other trust-region method as a start, as there is also a native trust region solver available through MKL (which would then perhaps be a good fit given the general C# / with native alternative approach)? Is this already-there-you-just-missed-it / under-development / not-wanted / desirable-if-someone-volunteers-to-add-it?

Thanks,
Joe
Coordinator
Oct 28, 2013 at 4:56 PM
Hi

Yes, this has been on the wish list for quite some time. We even got some code contributed (e.g. the whole csmpfit by its author), it has just not been integrated yet (mainly because some existing parts of the project have had priority).

MKL support would be highly appreciated, so it would be good if the implementation would fit to it nicely. But we would also need a fully managed implementation.

Hence: planned-later & help-appreciated-to-speed-up-integration & desirable-if-someone-volunteers-to-add-it

Thanks,
Christoph
Marked as answer by cdrnet on 1/28/2014 at 1:31 AM
Oct 28, 2013 at 10:13 PM
Hi Christoph,

I'm happy to help out adding MKL support (then perhaps lend a hand with managed version integration if needed).

Thanks,
Joe
Nov 13, 2013 at 9:23 PM
Hi Christoph,

Have forked and I think new code is about complete enough to ask a maintainer to take a look (bunch of decisions to make but thought it would be easier to put something in place as a starting point). What is the next step? (pull request?)

Thanks,
Joe
Coordinator
Nov 14, 2013 at 9:24 PM
Great!

Yes, a pull request at GitHub would work well, to discuss it and coordinate the next step - and of course to actually pull it.

Thanks,
Christoph
Jan 23 at 7:18 PM
Hello,

I would be also very interested by this feature and would be willing to help.

What's the status of the fork that has been created? Is there a to-do list or something?

Thanks!
Xavier
Coordinator
Jan 28 at 8:30 AM
You have already found it, but if someone else is looking, the issues where this is discussed is #173.

Thanks,
Christoph