search

OxfordIonTrapGroup / ndscan

N-dimensional scans for ARTIQ
GNU Lesser General Public License v3.0
18 stars 15 forks source link

Asymmetric Confidence Intervals #133

Open pathfinder49 opened 5 years ago

pathfinder49 commented 5 years ago

Much of our experimental gata is not gaussian distributed. Symmetric standard errors are therefore not accurate. It would therefore be desirable to plot non gaussian asymmetric confidence intervals.

This would go along with an OITG fitting implementation for non-gaussian errors and asymmetric interval calculation tools.

dnadlinger commented 4 years ago

We should possibly just implement this as an extra array dimension of size 3 (value and upper/lower bounds) rather than the current channel display_hints crutch.

pmldrmota commented 11 months ago

I would've just created another two display hints ("upper/lower_error_bar_for") - I think we are already very used to those. The implementation on the applet side should be trivial.

In the majority of cases, we plot binomially distributed data. Wouldn't it be better if we added a display hint to the result channel, e.g. "error_distribution" key with "binomial" value, and calculated the error bars in the applet / the analysis code instead of using an extra result channel?