Re: Polynomial Fit Program

Helmut Jarausch wrote:
Thomas Smid wrote:
Hi,

Does anybody know an open source code for a one dimensional polynomial
(n-th order) least squares fit program (preferably C++ or Fortran)?
Something similar to the Java applet under
http://www3.sympatico.ca/mcomeau/webpublic/javapage/reg/reg.htm would
do, but I need the source code, as I need to modify it to perform a
weighted fit (if this isn't built in already). Alternatively, if
somebody knows the contact details of the author of this particular
Java applet (Michel Comeau), it might help me as well.

I don't know if it helps, but free "Matrix Packages" like Octave or Scilab
have it built in.

What you really need is a Q-R-decomposition from any linear algebra package in C++.

In the simplest case (polynomials of moderate degree), if your data is contained
in the two column vectors x and y, build the matrix

A=[ones(x),x,x.^2,x.^3] // for a polynomial of degree 3
This is Matlab/Scilab/Octave jargon :

ones(x) : a column vector of as many ones as x has components

x.^n : (note the dot in front of ^) a column vector which has
x_i^n as i-th component

Now solve the least squares problem A*c ~ y and c contains the coefficients
of the polynomial (lowest first).

If you have weights, e.g. w_i, then first formally
build the diagonal matrix W= diag(w_1,...) and
solve the least squares problem W*A*c ~ W*y

In practice this is done by multiplying the i-th row of A and the
i-th component of y by w_i.

As you can see, if you have a standard (i.e. non weighted) least squares
solver you don't need to modify it.



This works okay as long as you don't try to go to very high order--that matrix [x_i^j] is a subset of the Hilbert matrix, which is very very ill conditioned for large orders.

Cheers,

Phil Hobbs
.

Relevant Pages

  • Re: Polynomial Fit Program
    ... Does anybody know an open source code for a one dimensional polynomial ... least squares fit program? ... In the simplest case (polynomials of moderate degree), ...
    (sci.math.num-analysis)
  • Re: Polynomial Fit Program
    ... Does anybody know an open source code for a one dimensional polynomial ... least squares fit program? ... In the simplest case (polynomials of moderate degree), ...
    (sci.math.num-analysis)
  • On limitations of the Mautsch Principle (was Re: ineqality)
    ... > squares. ... I would prove the homogeneous polynomial inequality ... polynomials (which would necessarily have to be linear combinations ... So symmetry on the way in does not imply symmetry on the way out. ...
    (sci.math)
  • Re: How to invert the sigular matrix.
    ... use interpolating polynomials at all, ... If this is a least squares problem, ... polynomial models is the scaling of your data. ... A better solution anyway is to use a spline again. ...
    (comp.soft-sys.matlab)
  • Re: Polynomial Fit Program
    ... least squares fit program? ... In the simplest case (polynomials of moderate degree), ... If you have weights, e.g. w_i, then first formally ... RWTH - Aachen University ...
    (sci.math.num-analysis)
Quantcast