-
Notifications
You must be signed in to change notification settings - Fork 4.2k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
71 x taus quality cuts improvement #2843
71 x taus quality cuts improvement #2843
Conversation
jpavel
commented
Mar 13, 2014
- Speeding up the execution of tau quality cuts (commits by @makortel)
- Reducing inefficiency of anti-mu discriminators at high PU (authored by @veelken)
- Clean up to follow cmssw guidelines (i.e. removing couts where possible)
- do not veto track segments in innermost CSC chamber - count number of muon stations containing matched segments rather than summing matched segments of different muons in the same muon station (in the old implementation a tau was failing the anti-muon Loose discriminator in case there are 2 reco::Muons each having 1 matched segment in the innermost CSC chamber. The new implementation requires that the matched segments are in at least 2 different chambers before the tau fails the anti-muon discriminator)
A new Pull Request was created by @jpavel (Pavel Jez) for CMSSW_7_1_X. 71 x taus quality cuts improvement It involves the following packages: RecoTauTag/RecoTau @nclopezo, @cmsbuild, @anton-a, @thspeer, @slava77, @Degano can you please review it and eventually sign? Thanks. |
@@ -252,7 +252,7 @@ double AntiElectronIDMVA5GBR::MVAValue(Float_t TauEtaAtEcalEntrance, | |||
{ | |||
|
|||
if ( !isInitialized_ ) { | |||
std::cout << "Error: AntiElectronMVA not properly initialized.\n"; | |||
edm::LogError("ClassNotInitialized") << "Error: AntiElectronMVA not properly initialized.\n"; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If it's a configuration problem, an exception is more appropriate. Else we'll just keep running, which is probably not what's intended.
Hi Slava, thanks for suggestions - all are implemented. |
nm, I see that changes are expected, as mentioned in the PR submission message |
Are there any plots describing changes in performance? BTW, plots with discrimination by ChargedIsolation, by MuonRejection (2 and 3), by CombinedIsolationDBSumPtCorr, and by Isolation ... what's more disturbing on a separate level is that nothing showed up in jenkins tests in |
@@ -48,13 +48,12 @@ | |||
} | |||
|
|||
void RecoTauConstructor::addPFCand(Region region, ParticleType type, const PFCandidateRef& ref, bool skipAddToP4) { | |||
//std::cout << "<RecoTauConstructor::addPFCand>:" << std::endl; | |||
//std::cout << " region = " << region << ", type = " << type << ": Pt = " << ref->pt() << ", eta = " << ref->eta() << ", phi = " << ref->phi() << std::endl; | |||
edm::LogInfo("TauConstructorAddPFCand") << " region = " << region << ", type = " << type << ": Pt = " << ref->pt() << ", eta = " << ref->eta() << ", phi = " << ref->phi(); |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
these should be replaced as well.
I suggest you grep for LogInfo in all the code and change to LogDebug
Nice, I will check results later today (including the Vertex::trackWeight in #2894 for CPU performance comparison) |
Hi Matti, this is not exactly true. Although reco::TrackBaseRef is avoided by quality cuts, it is still being used heavily by https://github.com/jpavel/cmssw/blob/71X_taus_QualityCutsImprovement/RecoTauTag/RecoTau/src/RecoTauVertexAssociator.cc where the weight of tracks in vertex is used to find the tau vertex. |
Hi Pavel, Thanks for the correction (I didn't look outside RecoTauQualityCuts.cc for the uses). However, in RecoTauVertexAssociator.cc you pointed, an intermediate |
I think that if reco::TrackBaseRef is never dereferenced one gets the speed-up anyhow. |
I agree - but it would work only after #2894 is merged, no? Otherwise the templates would not be defined in the code, or am I missing something? |
Results based on tests in CMSSW_7_1_X_2014-03-18-0200
Here is one diff (red is new) that looked particularly large, just to check with you that they make sense. |
Hi Slava, something like this is indeed expected - the aim of Christian's fix was to improve efficiency at larger eta regions, while the fake rate should not change dramatically. I will also let @veelken to comment on this plot. |
Hi Slava, the plot is the probability for muons to pass the discriminatorAgainstMuons3Loose, right ? Cheers, Christian On Mar 22, 2014, at 3:49 PM, Pavel Jez wrote:
|
Hi Slava, thank you very much for sharing the plot. It is good to see that the Tight discriminator changed much less. Thanks again, Christian On Mar 23, 2014, at 8:24 AM, slava77 wrote:
|
71 x taus quality cuts improvement