I have looked through the repo but not really found anything about it. Is there already a function for blind fits implemented? If so, how do I invoke it? Otherwise, can we set this up?
Designs
Child items
...
Linked items
0
Link issues together to show that they're related or that one is blocking others.
Learn more.
We have to set this up, it was one of my outstanding issues to solve. How do you want to go about it? I would say we give an interface to the RooFit blinding methods.
What level of interface do you think of? I think it would be nice to have it in the fit config. In our case it is of course rather an easy thing because we simply have to blind a region from m_lower to m_upper in one fit of a simultaneous/simple fit. This could be incorporated inside the config file easily (as far as my case/complexity goes).
What do you think of the other cases, is it feasible to use this approach or does that fail somewhere?
I was thinking of allowing to blind parameters, for example, instead of regions. In our case it's easy, just set a fit range (I don't think we support split fit ranges though) and some sort of selection when loading the data, but blinding parameters could be a bit more involved.
All in all, I was expecting to write some kind of blind keyword that could be used in fit config (or even on data loading config) that does whatever is relevant for each case.
on which level should we specify this? If the example should work, this will be inside each pdf on the same level of parameters, right? Otherwise, using mu as a name would be unambiguous or cumbersome.
what are possible other cases to use the blinding for? I just slightly dislike to create an additional interface (or syntax) if there is no need for it (but of course, keeping it flexible for the future is a good argument, though). Is there already a similar thing from which the syntax of the interface could be taken? A similar functionality like blinding (I have nothing in mind...)?
So, it depends a bit on the analysis... For Kpipiee we want to blind the mass, but once we go to calculate the R we will need to blind a parameter in the fit (the efficiencies), so it's probably worth implementing some interface.
I'm not sure about the level of the interface: sometimes it's at PDF level, sometimes at data level... Maybe it should just go at the same level as those...
PS: I'll be tomorrow (Wednesday) in Zürich, let's reserve some time for this.
I agree, that makes sense. If we detect an often used case where the syntax is quite cumbersome (say a lot of blinded fit variables), we could still think of some simpler syntax alias ("BLIND VAR ...").