RooStats development meeting

Hello,

I am glad that this forum is now open. Several people participating in the RooStats project are around for the summer, and I was hoping that would could organize an informal workshop at CERN sometime in June.

We already have several pieces of code (examples and higher level classes), some of which should be reviewed given the recent additions to RooFit. For example:
plone4.fnal.gov:4430/P0/phystat … es/0803001
www-ekp.physik.uni-karlsruhe.de/ … sKarlsruhe
isscvs.cern.ch/cgi-bin/cvsweb.cg … taTTooLLHC

As a reminder, the CVS area “Statistics Tools For The LHC” (the last URL) was setup as a sandbox for collaborative development. Once tools are ready and have been agreed upon, we can add them to the ROOT SVN. I’ve talked with René about the initial import, which will be managed with care. Then we can make a development branch. Again, only code that is well tested will go into ROOT SVN, and the LHC Statistics Tools for The LHC CVS area should be used for initial code development.

Can others that are interested please respond to this thread to comment on good/bad dates.

June 9-12 there is an ATLAS workshop, which may have several people already at CERN, but it may also be a very busy time.

All the best,

Kyle

Hi,

I’m also glad we now have a way of communicating together. I’m just stating to use this forum and in particular wondering if one gets an email when new topics are created. I couldn’t find out how I could subscribe to the “stat and math topic” in particular so I guess I (and everybody, ie. anybody) will either receive an email for every root related posting or none (except the ones I contributed). In that later case, one would have to regulary scan the forum for new posting!

Anyway, to answer you about good/bad dates: June 18-22 would definitively be a bad date (CMS week outside CERN).

Cheers,

– Gregory

I agree that it’s unfortunate that you can’t subscribe to the forum. The BB does have a Usergroups option, but this forum is not in there. It is also possible to be notified when someone replies to a particular thread (eg. this one).

Gregory, has your group made any modifications to the code you developed given Wouter’s addition of the “constraint” concept?

Hi,

It is also possible to be notified when someone replies to a particular thread (eg. this one).

That’s only working if you posted once into the thread, so except you and me no one else is beeing notified for our postings.

has your group made any modifications to the code you developed given Wouter’s addition of the “constraint” concept?

I tried it but at the time saw some differences to what I expected (probably I am wrongly using the code of wouter). So I thought I would wait for the documentation wouter mentionned in his email. On the other hand we have for long time a working implemanation of constraints so we still stick to that.

Another issue we mentionned already is that we still miss features in roofit to perform our statistical analysis. Now that we have a forum I can write about it.

In our toy MC experiments we generate alternatively S+B and B-only datasets and fit them with S+B and B-only models. Then we look at the lnQ distribution of the 2 types of sample. The problem is that:

  • in ToyMCStudy one cannot pass 2 PDFs, one for generation and one for fitting (or generating with one PDF and fitting with the 2 PDFs)
  • in a RooWorkspace one can’t specify what is signal and what is background model
  • I have also some other limitation in mind but to be forced to use ToyMCStudy if I want to use constraints is the main one for the moment

– Gregory

Hi again,

I also had to write my own ToyMC study to make sampling distributions for the (profile) log-likelihood ratio. I understood from Wouter, that the MCStudy does allow for ‘plugins’ that would make this possible.

Wouter, can you provide an example?

Since this will be a very common use-case, I think this is a candidate for a high-level RooStats class (which may use RooMCStudy internally).

[quote]> It is also possible to be notified when someone replies to a particular thread (eg. this one). [/quote]Yes. At the bottom left of the page is a link called “watch this topic” that will setup the forum to send you mail every time somebody post in the topic.

Cheers,
Philippe.

Sorry I made a mistake:

Anyway, to answer you about good/bad dates: June 18-22 would definitively be a bad date (CMS week outside CERN).

CMS week is the one following (23-27). In Karlsruhe we would however not be much available in the week (16-20) although we would be at CERN in the beginning of that week. The dates Bob Cousins proposed:

Indeed we need to converge on dates for meetings of RooStats, joint
CMS/ATLAS stats, and maybe more. Can Eilam+Glen and Luc help
organize these? A couple days that Wouter and I suggest are Monday
June 30 or Tuesday July 1.

Are fine for us.