[94X] DeepFlavour Negative Tagger

[emilbols]

backport of #23467

The only difference is that the DeepFlavour model is no longer updated

[cmsbuild]

A new Pull Request was created by @emilbols for CMSSW_9_4_X.

It involves the following packages:

PhysicsTools/PatAlgos
RecoBTag/TensorFlow

@perrotta, @monttj, @cmsbuild, @slava77, @gpetruc, @arizzi can you please review it and eventually sign? Thanks.
@TaiSakuma, @gouskos, @rappoccio, @HeinerTholen, @seemasharmafnal, @mmarionncern, @imarches, @ahinzmann, @smoortga, @acaudron, @jdolen, @drkovalskyi, @ferencek, @jdamgov, @nhanvtran, @gkasieczka, @schoef, @clelange, @JyothsnaKomaragiri, @mverzett, @gpetruc, @mariadalfonso, @pvmulder this is something you requested to watch as well.
@davidlange6, @slava77, @fabiocos you are the release manager for this.

cms-bot commands are listed here

• Backported from DeepFlavour Negative Tagger + DeepFlavour 10X training #23467

[slava77]

@cmsbuild please test

[cmsbuild]

The tests are being triggered in jenkins.
https://cmssdt.cern.ch/jenkins/job/ib-any-integration/29045/console Started: 2018/07/08 16:31

[cmsbuild]

+1
Tested at: ee07dfe
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-23763/29045/summary.html

[cmsbuild]

Comparison job queued.

[cmsbuild]

Comparison is ready
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-23763/29045/summary.html

Comparison Summary:

• No significant changes to the logs found
• Reco comparison results: 96 differences found in the comparisons
• DQMHistoTests: Total files compared: 27
• DQMHistoTests: Total histograms compared: 2721493
• DQMHistoTests: Total failures: 103
• DQMHistoTests: Total nulls: 0
• DQMHistoTests: Total successes: 2721228
• DQMHistoTests: Total skipped: 162
• DQMHistoTests: Total Missing objects: 0

[slava77]

Reco comparison results: 96 differences found in the comparisons

changes in 94X are not allowed at this point.
Please adjust the implementation of this backport to preserve the default behavior.

[emilbols]

OK, so i guess we should not update the DeepFlavour model?

[slava77]

On 7/10/18 12:56 PM, emilbols wrote: OK, so i guess we should not update the DeepFlavour model? —

right, not in the default setup

[emilbols]

So the discriminator distribution slightly changed, which it shouldn't have. I think i found the issue, i was calculating deltaR using the vertex flight direction as done in deepCSV, but it should be done using the vertex momentum for deepflavour. If there are no differences with the baseline now, i will propagate this to the master as well.

[slava77]

@cmsbuild please test

[cmsbuild]

The tests are being triggered in jenkins.

[slava77]

@smuzaffar @mrodozov
I believe that this is the right console for the tests
https://cmssdt.cern.ch/jenkins/job/ib-any-integration/29079/console
this appears to be stuck.
Please check.

[slava77]

@cmsbuild please test

hoping that a restart will work

[cmsbuild]

The tests are being triggered in jenkins.
https://cmssdt.cern.ch/jenkins/job/ib-any-integration/29083/console Started: 2018/07/12 09:47

[smuzaffar]

things have not still recovered from the yesterday's CVMFS issue https://cern.service-now.com/service-portal/view-outage.do?n=OTG0044912

looks like VM cmsbuild82 is in bad state, I am talking out of system.

[cmsbuild]

+1
Tested at: 3dc27e4
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-23763/29083/summary.html

[cmsbuild]

Comparison job queued.

[cmsbuild]

Comparison is ready
https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-23763/29083/summary.html

Comparison Summary:

• No significant changes to the logs found
• Reco comparison results: 0 differences found in the comparisons
• DQMHistoTests: Total files compared: 27
• DQMHistoTests: Total histograms compared: 2721493
• DQMHistoTests: Total failures: 104
• DQMHistoTests: Total nulls: 0
• DQMHistoTests: Total successes: 2721227
• DQMHistoTests: Total skipped: 162
• DQMHistoTests: Total Missing objects: 0

[slava77]

+1

for #23763 3dc27e4

• the backport of a combination of DeepFlavour Negative Tagger + DeepFlavour 10X training #23467 b626752 and DeepFlavour deltaR calculation fix #23781 319c283 (current state of the master branch) is correct; this backport excludes the change to the new training made for the master to preserve backward compatibility in this production release.
• jenkins tests pass and comparisons with the baseline show no differences.

[fabiocos]

@slava77 I understand that the default behaviour is kept with this backport, but in principle the user has in case the possibility to access the new training by changing the input file that here is omitted, right?

[slava77]

On 7/18/18 8:20 PM, Fabio Cossutti wrote: @slava77 I understand that the default behaviour is kept with this backport, but in principle the user has in case the possibility to access the new training by changing the input file that here is omitted, right?

yes, the training files are interchangeable in a sense that the algorithmic code changes will not be required.

[fabiocos]

+1

the update has been made backward compatible for the default proposed configuration

[fabiocos]

merge