gnuplot-beta

limitations of fit?

[<pl...@pi...>]

Hi,

I have about 2300 data points that I need to fit a simple cosine to. 
However, fit says it is converging at the first iteration and 
effectively does nothing.

The data has a good signal to noise and I have primed the parameters 
with values which are pretty close. It just needs to shift the constant 
term closer to zero.

I had this working fine with less data points (about 200)  on 
essentially the same dataset.

Is the large number of points causing fit to fail to converge properly?

TIA, Peter.



  fcos1(x)= a1*cos(2*pi*(x-yz1)/p1)+c1
a1=c1=1; yz1=1990; p1=4.5
  fit [:1991.5] fcos1(x) datafile1  u 1:2 via  a1,c1,p1,yz1

After 1 iterations the fit converged.
final sum of squares of residuals : 3036.32
rel. change during last iteration : 0

Re: limitations of fit?

[Hans-Bernhard B. <HBB...@t-...>]

Am 31.01.2015 um 20:47 schrieb pl...@pi...:

> I have about 2300 data points that I need to fit a simple cosine to.
> However, fit says it is converging at the first iteration and
> effectively does nothing.
>
> The data has a good signal to noise and I have primed the parameters
> with values which are pretty close. It just needs to shift the constant
> term closer to zero.

The RMS you got appears to disagree with that assessment.

 > final sum of squares of residuals : 3036.32

For 2300 points, and a function that only varies between 0 and 2 with 
the given starting parameters, you're getting an average deviation 
between function and data of more than 1.0 per data point.  That's quite 
a bit removed from being "pretty close".

> I had this working fine with less data points (about 200)  on
> essentially the same dataset.

200 data points cannot ever be "essentially the same dataset" as 2300.

> Is the large number of points causing fit to fail to converge properly?

Almost certainly not.

For a more meaningful answer you'll have to divulge the dataset in question.