Problems with exponential fit.

I start with the following data:

x,y
0.0025,10
0.001818,100
0.001429,10000

And try to fit to rate=a*exp(bx) using the following code:

p=fit(x,y,'exp1');

p2=coeffvalues(p);

yfit=p2(1)*exp(p2(2)*x);

semilogy(x,y,'.',x,yfit)

and get a poor quality fit along with values that do not make sense. It really underestimates y for larger x.

Excel’s exponential fit seems reasonable.

Any ideas why the Matlab fit does not work. I was working with small x values so I multiplied by 1000 but it had no effect.

Answer 1

[Walter Roberson]

http://www.mathworks.com/matlabcentral/answers/221316-how-can-i-find-these-values-using-polyfit-command#comment_288984

[Todd Gross]

Using transformed coordinates e.g. log x and log y with a polynomial fit often leads to values other than A for x=0 when y = A*exp(bx). I am interested in learning why exp1 fit does not work well in Matlab when the similar function in Excel seems to work well.