home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: sci.math.stat
- Path: sparky!uunet!charon.amdahl.com!pacbell.com!ames!ncar!csn!yuma!agropyron!steve
- From: steve@agropyron.cfnr.colostate.edu (Steve)
- Subject: how-to: nonlinear surface fitting
- Sender: news@yuma.ACNS.ColoState.EDU (News Account)
- Message-ID: <Nov18.061759.17190@yuma.ACNS.ColoState.EDU>
- Date: Wed, 18 Nov 1992 06:17:59 GMT
- Reply-To: steve@agropyron.cfnr.colostate.edu
- Nntp-Posting-Host: agropyron.cfnr.colostate.edu
- Organization: Colorado State University, Dept. of Mech. Eng.
- Keywords: non-linear, regression, surface, model
- Lines: 31
-
- I'm trying to fit a model to historical data, and keep hitting
- limitations and problems with various approaches/packages.
- This doesn't seem to be too weird of a problem, but I'm getting frustrated.
- I'm not a statistics guru, so I could be (probably am) wrong.
-
- The problem, I have several samples of data in x and y that I need
- to fit a surface/model to. X and Y are independent. For each sample,
- x ranges from 30 to 60 by increments of 2, while y goes from 130 to
- 260 by increments of 5, I have values for each point of this matrix,
- although the border/outer edge data values are unreliable. When plotted,
- the surface created by the seperate data samples looks like a fairly
- "steep" poisson density function (tall hump in the middle). I've been
- unable to find a reference/package for this type of a fit. I've found
- linear/quadratic surface algorithms, but this is exponential. I've found
- exponential curve algorithms, but no surfaces except where y is a dependent
- variable ( if I am understanding things correctly, I'm refering to ODRPACK
- here).
-
- I'm guessing a model of the form (roughly):
-
- F G M N
- B D(1-Ex ) I K(1-Ly )
- f(x,y) = Ax * Ce + Hy * Je + Pxy
-
-
- Any ideas/pointers/references would be GREATLY appreciated!
-
- thanks
-
- Steve = steve@bouteloua.cfnr.colostate.edu
-
-