SUMMARY When doing a fit, checking "Use given data uncertainties" on fit options has no effect. STEPS TO REPRODUCE 1. Create a fit, making sure the spreadsheet has a column for the Y uncertainties. 2. Now disable the "Use given data uncertainties" option, re-run the fit, and compare the results OBSERVED RESULT Nothing changes. EXPECTED RESULT For instance, one would expect the uncertainties to influence the χ² calculation, since we should be dividing each squared residual by the variance of each point, which changes according to its uncertainty. SOFTWARE/OS VERSIONS Operating System: Arch Linux KDE Plasma Version: 5.27.5 KDE Frameworks Version: 5.106.0 Qt Version: 5.15.9 Kernel Version: 6.3.6-arch1-1 (64-bit) Graphics Platform: Wayland
Did you select the weights for the data uncertainties to use? When i select "Instrumental (1/col²)" and the third column (Y uncertainties) for Y-Weight, the fit changes accordingly.
(In reply to Stefan Gerlach from comment #1) > Did you select the weights for the data uncertainties to use? > When i select "Instrumental (1/col²)" and the third column (Y uncertainties) > for Y-Weight, the fit changes accordingly. Ohh, I see... So that's what I was missing. To be fair, I don't think it's very intuitive right now: I add a column with the uncertainties, and then the only obvious uncertainty-related option I see on the fitting window is that "Use given data uncertainties" checkbox; I see that it's checked, so I figure that means it must be using the uncertainties I previously added (so what does it do, then?). I never would've guessed I had to manually add them as a weight (and I've been a proud LabPlot user for years!). Sure, retrospectively, I get the logic: the uncertainties will be treated as a generic data weight by the fitting algorithm (more uncertainty = less weight). But could we perhaps rework things a little so the workflow is slightly more intuitive? For example, on the dialog that pops up when we attempt to create a fit from a spreadsheet, we could add a data input for the uncertainty (just like with X-Data and Y-Data). Then, if a column is selected, we could ask for a weight (or maybe just set it to "instrumental" by default, that's probably what most users expect when working with uncertainties).
Added issue https://invent.kde.org/education/labplot/-/issues/577 to fix this.