Dear all, I released levmar-1.0, an implementation of the Levenberg-Marquardt minimization algorithm that can be used for least squares, curve fitting and nonlinear programming.
The most important change in this release is the use of vectors and matrices from the hmatrix package (which uses the high performance vector package). An application at work became 25 times faster when switching to the new levmar! I'm not sure this major speedup can be entirely attributed to the switch from lists to vectors. I also switched the FFI types from CDouble to Double in the internal bindings-levmar package. This made the expensive realToFrac conversions from CDouble to Double unnecessary which I believe had a big impact. API Documentation: http://hackage.haskell.org/package/levmar-1.0 I also switched the repository from darcs to git and hosted it on github: https://github.com/basvandijk/levmar/ Regards, Bas _______________________________________________ Haskell mailing list Haskell@haskell.org http://www.haskell.org/mailman/listinfo/haskell
