RooStats Developer Meeting

Thursday 9 Feb 2012, 17:00 → 18:30 Europe/Zurich

CERN

Dial-in numbers and access codes Dial-in numbers: +41227676000 (English-US, Main) Access codes: 0193780 (Leader) 0138629 (Participant) Leader site: https://audioconf.cern.ch/call/0193780 Participant site: https://audioconf.cern.ch/call/0138629

RooStats Developer Meeting (09 Feb 2012)

Lorenzo, Kyle, Wouter, Sven

 

Release 5.33 at end of February.  Aim for Wed. Feb 22 to have code in.

 

SVN:

merge trunk to dev branch (don't forget about the user's guide)

merge 532-patches version of HistFactory into trunk and dev branch

 

5.32 patch aim for Tuesday Valentines day (Lorenzo will be gone)

 - misc bug fixes [Wouter ]

 - probably include the ability to control constraint term NormSet [Wouter]

 - check if factory->importClassCode [Wouter]

 - HistFactory fixes [George & Kyle]

 - More protections in HistFactory parser [George & Kyle]

 - SamplingDistPlot fixes (Sven)

 - AsymptoticCalculator sigma(0) vs sigma(mu)  ?? [Kyle & Lorenzo]

 - Bug effecting simultaneous PDF is generating composite data store ?? [Wouter]

 - TestStatistic (switch for signed version)??

 

 

Wouter still to send summary of many changes

 

Sven for 5.33

Wouter:

- Interface changes for auto-detection of constraint terms

- RooProdPdf Set vs. List (proxy)

- Simultaneous PDF is generating composite data store 

Sven (away on 17th):

- importance sampling : new class inheriting from ToyMCSampler (re-)committed

- multiple nulls [not done, to do next]

- FrequentistCalculator can specify which nuisance parameters are used for toys [done]

- removed options for importance sampling

- over-ride specific nuisance parameters [not done, priority]

- Detailed output from toy MC sampler [re-committed]

- SamplingDistribution, SamplingDistPlot, HTR & HTRPlot also changed

- follow up on TF1 and line color

- would be nice to store sigma(0) and sigma(mu) for each point to draw asymptotics

- MarkovChain (minor changes to merging and burn-in, non-integer weights)

- huge memory usage, should be looked into

- MultiNest (technical stuff worked out, can be committed)

Lorenzo (gone week of 13th):

- HypoTestInverter (requests) [Gena, Lorenzo]

- signed version of profile likelihood ratio

- autoscan for 2-sided intervals

- extend autoscan for median and bands

- HypoTestInverter

Gena, Max, everyone

- Simple high-level user interface for specific calculations

- Check PROOF

 

Next meeting Thursday 23rd

 

All the best,

 

Kyle