non linear regression algorithms
hello all,
I am looking for a non linear regression algorithm to find constants (a,b,c,d) for my function
Y=Exp^(a*x+b)+Exp^(c*x+d)
Where Y and X i already have the data
Using non linear regression i need to find the constants which would be the best guesses
Could some body provide some informations.
Regards biji
[360 byte] By [
baijupaca] at [2007-11-27 4:04:43]

# 1
That's an unusual form for a relationship, so I presume you got it from some theoretical process and you're trying to match your data with your theory. Does your theory give any hints as to the rough sizes of the parameters? E.g. it's probably not too bad to try a hillwalking-type approach if either c~=0 or c~=a.
# 2
A further thought:
If you know a and b then you can do linear regression to find c and d, via the transform
ln (y - exp(ax+b)) = cx+d
You can then verify a and b by plugging c and d into the original function and doing the same transform under tr a-d cdab
So you could try using linear regression on
ln y = ax+b
to find values for a and b, use those to find values for c and d, and then iterate a few times checking for convergence.
# 6
My aim was to find the right constants to get the best curve fit. i tried couple of experiments with different sets of data it gives best guesses.
Currently i am plotting the graph using applets. Do any one have a idea how would i use it in webapplication. Want to use Open source graph solutions .But drawing a functional chart all these utilities will be of any help !!!
Regards
RA
# 12
> I could use JFreeChart ...but i should give some
> feature to even edit or ignore specific points
> which i plot. If i display as image i cant provide
> the feature right.
If you want it interactive and using Java then a Applet is the obvious solution.
>
> baiju and PAc are the same person ...hehehehe
I worked that out but I don't find it funny! It just caused confusion.