A long time ago, I used a Texas Instruments SR-50A to do linear regression in high school. This calculator had a single memory location, and the manual explained the procedure. Good times.

So, if you're trying to do exponential regression, how about just doing linear regression on x, log( f( x ) ) values? You can convert that back into an exponential afterwards.

That's what I'd try.

Thanks much for the tip!

(and the groklaw link)

halfcountplus:

I noticed Algorithm::CurveFit, and it looks like it would do the job, though it seems you have to define your error function with Math::Symbolic. It's too bad they didn't simply specify the interface to a function you could provide. (Well, they may, but a brief review of the docs didn't make that obvious to me.)

...roboticus

When your only tool is a hammer, all problems look like your thumb.