On 16/01/12 11:01, Markus wrote: > Hi there, > > it might be a silly question, but I think I haven't understood the fit widget > in > Veusz yet. I tried to plot a regression line to my scatter plot with the fit > widget. I selected the x- and y dataset (both without errors) and evaluated > the > "fit function" button and received the following: "Warning: No errors on y > values. Assuming 5% errors. Fit data not equal in length. Not fitting." > Looking at the example document, I realised that the fit function there > (specified as a+b*x) does not look linear at all. So I guess it is doing > something completly different from what I thaught it would do. Can anybody > give > me some advice? Is there any other way to plot regression lines in Veusz?
Hi Markus The plot doesn't look linear in the example because one of the axes is log (I can't remember why I did that). The fitting does a chi^2 minimisation on the model to the data, using the model, points and the Y error bars (or assumes 5% errors if they not given). It sounds like your X and Y data don't have the same length, so I'd check that in the data edit dialog box. It should work if they are equal in length. There's also the option of using pyminuit which does a more sophisticated minimisation. There's clearly an opportunity to add better fitting facilities to Veusz, to try other methods, investigate uncertainties, etc. You can use the command line, however, to use scipy fitting in Veusz, if required. Jeremy _______________________________________________ Veusz-discuss mailing list [email protected] https://mail.gna.org/listinfo/veusz-discuss
