| |
 |
|
|
Science Forum Index » Math - Numerical Analysis Forum » nonlinear regression
Page 1 of 1
|
| Author |
Message |
| Tim vor der Brück |
 Posted: Thu Jan 04, 2007 12:19 pm |
|
|
|
Guest
|
Hi, I have the following nonlinear regression problem
y=c1Fermi(a1,b1)(x1)+c2Fermi(a2,b2)(x2)
Fermi(a1,b1)(x)=1/(1+e^-((x1-a1)/b1))
c1,c2 are constants. a1,b1,a2,b2 should be determined using
nonlinear regression with least squares. Unfortunately the algorithm
does not converge. It works if I have only one summand.
Is it not possible to solve this with nonlinear least squares or made I
a mistake somewhere? Is there an alternative possibility?
Thanks for any help,
Tim. |
|
|
| Back to top |
|
| Ray Koopman |
| Posted: Thu Jan 04, 2007 4:11 pm |
|
|
|
Guest
|
Tim vor der Brück wrote:
Quote: Hi, I have the following nonlinear regression problem
y=c1Fermi(a1,b1)(x1)+c2Fermi(a2,b2)(x2)
Fermi(a1,b1)(x)=1/(1+e^-((x1-a1)/b1))
c1,c2 are constants. a1,b1,a2,b2 should be determined using
nonlinear regression with least squares. Unfortunately the algorithm
does not converge. It works if I have only one summand.
Is it not possible to solve this with nonlinear least squares or made I
a mistake somewhere? Is there an alternative possibility?
Thanks for any help,
Tim.
In situations like this it sometimes helps to rewrite (x-a)/b
as c + d*x, iterate for c and d, then take a = -c/d and b = 1/d. |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Thu Jan 08, 2009 6:30 pm
|
|