www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.announce - mir-optim v0.0.1: betterC nonlinear least squares

reply 9il <ilyayaroshenko gmail.com> writes:
https://github.com/libmir/mir-optim

This work has been sponsored by Symmetry Investments and Kaleidic 
Associates.

http://symmetryinvestments.com
https://github.com/kaleidicassociates
Jul 09
next sibling parent jmh530 <john.michael.hall gmail.com> writes:
On Monday, 9 July 2018 at 13:54:17 UTC, 9il wrote:
 https://github.com/libmir/mir-optim

 This work has been sponsored by Symmetry Investments and 
 Kaleidic Associates.

 http://symmetryinvestments.com
 https://github.com/kaleidicassociates
It looks like the docs are not correctly handling the parameters you have in the struct. https://mir-optim.dpldocs.info/mir.least_squares.LeastSquaresLM.html
Jul 09
prev sibling parent reply Shigeki Karita <shigekikarita gmail.com> writes:
On Monday, 9 July 2018 at 13:54:17 UTC, 9il wrote:
 https://github.com/libmir/mir-optim

 This work has been sponsored by Symmetry Investments and 
 Kaleidic Associates.

 http://symmetryinvestments.com
 https://github.com/kaleidicassociates
great! do you have any plans of algorithms to be supported?
Jul 09
parent reply 9il <ilyayaroshenko gmail.com> writes:
On Monday, 9 July 2018 at 21:52:22 UTC, Shigeki Karita wrote:
 On Monday, 9 July 2018 at 13:54:17 UTC, 9il wrote:
 https://github.com/libmir/mir-optim

 This work has been sponsored by Symmetry Investments and 
 Kaleidic Associates.

 http://symmetryinvestments.com
 https://github.com/kaleidicassociates
great! do you have any plans of algorithms to be supported?
The algorigbms from https://github.com/JuliaNLSolvers are good candidates. No plans to implement them for now, but PRs are wellcome.
Jul 09
next sibling parent jmh530 <john.michael.hall gmail.com> writes:
On Tuesday, 10 July 2018 at 02:10:56 UTC, 9il wrote:
 The algorigbms from https://github.com/JuliaNLSolvers are good 
 candidates. No plans to implement them for now, but PRs are 
 wellcome.
Dlangscience has headers for the nlopt and glpk C libraries that I've used in the past. ipopt is another one that I've used, but I don't think there's a D interface. IMHO, you'd need a good reason to implement all of the functionality of a non-linear optimizer on your own. Simple linear and quadratic programming are probably a useful start, but beyond that the ability to interface with C/Fortran libraries goes a long way.
Jul 10
prev sibling parent reply jmh530 <john.michael.hall gmail.com> writes:
On Tuesday, 10 July 2018 at 02:10:56 UTC, 9il wrote:
 The algorigbms from https://github.com/JuliaNLSolvers are good 
 candidates. No plans to implement them for now, but PRs are 
 wellcome.
Another type of functionality that would be useful: https://www.mathworks.com/help/matlab/ref/fzero.html This can be done with a non-linear least squares, but I don't think that's the most efficient method.
Jul 13
parent 9il <ilyayaroshenko gmail.com> writes:
On Friday, 13 July 2018 at 16:20:30 UTC, jmh530 wrote:
 On Tuesday, 10 July 2018 at 02:10:56 UTC, 9il wrote:
 The algorigbms from https://github.com/JuliaNLSolvers are good 
 candidates. No plans to implement them for now, but PRs are 
 wellcome.
Another type of functionality that would be useful: https://www.mathworks.com/help/matlab/ref/fzero.html This can be done with a non-linear least squares, but I don't think that's the most efficient method.
D's findRoot probably is the world's most efficient and robust implementation. For single initial point fzero we can be implemented as IEEE set binary search of second initial point x1 such that sign(f(x0)) != sign(f(x1)) plus D's findRoot.
Jul 14