Wednesday, May 14, 2008

SciStat.org regression models

Added a few regression models from the SciStat.org nonlinear regression datasets. They are solved by the linear and non-linear regression solvers GAMS/LS and GAMS/NLS.

  1. Dugongs implemented as dugongs.gms This is actual a linear regression problem, so we solve with LS. The log/log fit is better, see also dugongs.png and dugongsplot.gms. What is a Dugong? See: http://en.wikipedia.org/wiki/Dugong.

  2. Dolphin in gams formulated as dolphin.gms. If we add a constant term the problem converges to a singular solution. The problem may be related to an outlier visible in the plot dolphin.png (created by dolphinplot.gms).

  3. brunhild is implemented as brunhilda.gms. The first non-linear fit. The seconds one is better. The third regression is a linearized version of model 2.

  4. troutpcb is implemented as troutpcb.gms. The linear regression problem provides initial values for the non-linear regression problem.

  5. chloride is implemented as chloride.gms. See also chloride.png, chlorideplot.gms

  6. rabbit is implemented as rabbit.gms. See also rabbit.png, rabbitplot.gms
  7. chlorine: chlorine.gms

I used GNUPLOT to plot results, e.g.: