bias correction

[vuillaut]

Hi

Is bias correction something wanted in pyirf?

If so, where should it be implemented?
Directly as an option in the resolution functions?
Or as separate functions?

[maxnoe]

@vuillaut Yes, I think this is the right place here. But what exactly do you have in mind?

What you'd need is some kind of transformation function f(reco_energy) -> reco_energy' that is fit on a test data set. I think MARS uses splines? @moralejo

[vuillaut]

My first though is a very simply correction of the bias to centre distributions to 0.
Here is an example of what I had implemented in ctaplot:
https://github.com/cta-observatory/ctaplot/blob/dcb2e332e0ad8138119d543ac81b579b2b5d5ce7/ctaplot/ana/ana.py#L739

[maxnoe]

I don't think this will work. The bias correction should definitely be energy dependent for the energy bias correction.

[HealthyPear]

Thanks @vuillaut for having activated Discussions - I think now pyirf is already in an advanced status to make this useful

@maxnoe if you didn't know, next time you can use the option in the bottom right menu in the sidebar to create an issue from a discussion. We could migrate to the issue you created or make a new one from this discussion, same for me.

---

I think we could generalise this option in a similar way to what we did for resolution:

• a module or a section of it for bias correction methods (or applied to specific cases, like energy)
• the resolution function calls a keyword argument which sets which function to use

Regarding specifically what CTAMARS does see here.
This happens after cuts optimisation, so in that case it fits with pyirf's scope.

[maxnoe]

I actually used the "REference in new issue feature" to create #185, let's continue there.

[HealthyPear]

I think if you use the create issue option the entire chat migrates and not only the reference