How to curve fit to a log curve in Gnuplot?

Question

I have some code in a Gnuplot file that does curve fitting, currently to a line:

f(x) = m * x + b
fit f(x) "data.txt" using "Days":"Data" via m, b

This works great, but the data looks like it will fit a log curve better. So I tried the following:

f(x) = b + m * log(x)
fit f(x) "data.txt" using "Days":"Data" via m, b

This results in the following error:

line 46: unknown type in real()

What am I doing wrong?

Not sure about the internals of how fit works, but if x ever goes negative, log(x) becomes a complex number, which fit may not be prepared to deal with (assuming gnuplots log() implementation even works for negative numbers, instead of returning a "domain error" or other similar condition)... — twalberg
– twalberg, Commented May 19, 2015 at 18:45
@twalberg Thanks. x will never go negative for me. Is there a way to constrain it? — user1002430
– user1002430, Commented May 19, 2015 at 19:20
Fitting to non polynomials is usually a bad idea if you can avoid it. If your data looks like a log, then try fitting to a linear function but using preprocessed data, e.g. fit a*x+b "data" using (log($1)):2 via a,b or something like that. Show your data if you want better help figuring out the actual problem here. — Miguel
– Miguel, Commented May 19, 2015 at 19:55

san4d · Accepted Answer · 2015-12-09 21:32:36Z

2

I just had a similar issue. My data set was small (8 points) on a semi-log plot. When trying to fit, I also received `

unknown type in real()`

I changed my initial parameter estimates and everything works. Hopefully, that helps.

answered Dec 9, 2015 at 21:32

san4d

916 bronze badges

Sign up to request clarification or add additional context in comments.

1 Answer 1