Jaroslav Hajek wrote: > On Mon, Mar 31, 2008 at 5:38 PM, Jonathan Stickel <[EMAIL PROTECTED]> wrote: >> Firstly, I would like to ask about the spline-gcvspl package, which >> provides a nice way to smooth data. What is the origin of the non-free >> license? I looked at the source link for the fortran dependency and >> performed a google search, but I cannot find the posted license >> anywhere. Does the non-free license (dated 1986) still apply? Other >> references on the web, including a Matlab mex interface, seem to imply >> that this code is public domain, or at least provide no >> copyright/license information at all. >> >> Secondly, I have written a set of functions that implements data >> smoothing by Tikhonov regularization. For most cases it performs at >> least as well as the gcvspl method, especially when looking at >> derivatives of the data. The code will be under the GPL and is written >> completely in m-files. My question is: should I put the functions in >> an existing octave-forge package or create a new one? They shouldn't go >> in splines since a spline method isn't used. How about a >> "data-smoothing" package or perhaps a "regularization" package? >> > > Is this a kernel-based smoothing method? If so, you may also check my > recently-created > OctGPR package in main/ that is dedicated to regression & smoothing of > multidimensional > data. It currently implements the Gaussian Process Regression (kernel > smoothing) with > kernel width estimation via trust-region. I'm also planning to add RBF > regression > using GCV or BIC with the same optimization method in the near future > (currently > in progress). The package is targeted to medium-scale mildly multidimensional > (say, up to a few thousand points in up to 10-15 dimensions). > In any case, a little duplicate work is normal for Open Source :) >
Although I have come across the concept of kernel-based methods, I still do not understand exactly what that means. I am an engineer who dabbles in applied mathematics. I do not think the method I have implemented is kernel-based. It is sometimes known as "penalized least squares"[1]. The idea is to minimize a function that is the sum of a goodness of fit term and a data roughness term. I have implemented it for one-dimensional data only, and I have not yet seen a multi-dimensional implementation in the literature, although it may be possible. Jonathan [1] Anal. Chem.; 2003; 75(14) pp 3631 - 3636 ------------------------------------------------------------------------- Check out the new SourceForge.net Marketplace. It's the best place to buy or sell services for just about anything Open Source. http://ad.doubleclick.net/clk;164216239;13503038;w?http://sf.net/marketplace _______________________________________________ Octave-dev mailing list Octave-dev@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/octave-dev
