Updated parameters of the Mustache SC algorithm, optimized for Run 3 #31823
Conversation
|
The code-checks are being triggered in jenkins. |
|
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-31823/19124
|
|
A new Pull Request was created by @rchatter (Rajdeep Mohan Chatterjee) for master. It involves the following packages: RecoEcal/EgammaCoreTools @perrotta, @jpata, @cmsbuild, @slava77 can you please review it and eventually sign? Thanks. cms-bot commands are listed here |
|
@cmsbuild please test |
|
The tests are being triggered in jenkins.
|
| @@ -68,25 +68,25 @@ namespace reco { | |||
| bool inDynamicDPhiWindow( | |||
| const float seedEta, const float seedPhi, const float ClustE, const float ClusEta, const float ClusPhi) { | |||
| // from Rishi's fits 06 June 2013 in log base 10 | |||
There was a problem hiding this comment.
The new parameters are not taken from the Rishi's fit of more than 7 years ago, I imagine: either you update the comment, or better to remove it
There was a problem hiding this comment.
Updated the comment to add the correct reference.
Thanks,
Rajdeep
|
+1 |
|
Comparison job queued. |
cmsbuild
added
code-checks-pending
tests-pending
and removed
code-checks-approved
tests-approved
labels
|
The code-checks are being triggered in jenkins. |
|
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-31823/19129
|
|
+1 |
|
Comparison job queued. |
cmsbuild
added
comparison-pending
tests-approved
and removed
comparison-available
tests-started
labels
|
Comparison is ready Comparison Summary:
|
|
@rchatter we ran some additional validation on this PR with gun samples to be sure. We see an increase of photon multiplicity per jet (16.0): We also see a slightly better jet resolution (19.0): Here's a list of all the DQM plots with MET or JetValidation: Let me know if this is in line with your expectations and if you see anything else out of the ordinary in the DQM plots. Thanks! |
|
Here's phase2 upgrade workflow DQM, please check this one, instead of the Run1 I posted above (thx Slava for noticing!)
http://jpata.web.cern.ch/jpata/reco/31823/11705.0/ |
|
Hi @jpata, do you have DQM plots for ECAL superclusters or pfPhotons/Electrons with a e/gamma gun sample? The mustache is really only used in the reco of e/gamma so any effect on jets in an indirect one. My guess here is that having a tighter parameters in the mustache superclustering means that we cluster less energy from PU into superclusters and therefore the jets peaks up more photon. Thank you, simone |
|
Hi again, there should be a Hgg sample run as part of the standard validation matrix, @nancymarinelli could have a look at that if you saved the DQM from that workflow. Here https://lzygala.web.cern.ch/lzygala/Envelope_OptPFRecHitThresholds34/mustache_comparison_SingleSetParams_Weighted/h_SC_nPFClusters.pdf you can also see the number of pfClusters (i.e. the same thing as photon multiplicity but for ECAL SC) with the old and new set of parameters. For ECAL SC the number of constituents decreases with the new parameters so it can be reasonable to expect that the rejected pfClusters are clustered by nearby jets, this would explain the plot above. Could you also please confirm that these are Run2 DQM plots and which release is "ref"? Thank you, simone |
|
Thx for the suggestion @simonepigazzini! The idea here was to check downstream things that might not otherwise have been in the DPG view in the validations, but we can also produce plots for specific EGamma validation. For the photon gun sample, here are a few that see bigger changes with this update. The rest of the plots are at the cern link above.
11602.0 and 11705.0 are run3 workflows as far as I understand (2021), ref was From the reco point of view, I would say as long as the changes are understood and noticed by the relevant POGs also downstream, this is acceptable and we can proceed. :) |
|
Just to be sure of no surprises downstream, I ran some extended tests. While I don't find any major differences in downstream jet or met outputs, there are some fluctuations. For example, The full DQM output for Overall, I think we can proceed with this PR from the reco side. |
|
Hi Joosep, Just to understand the second plot: Is the comparison on precisely the same events with and w/o this update, in which case the difference seems a "little" large. Also I don't precisely recall how exactly the photon multiplicity in a jet is evaluated. But most likely these are not photons using the Mustache SC algorithm so like you said shouldn't be a cause for worry [unless JETMET thinks otherwise]. Best, |
|
These should be the same events, generated with the workflows mentioned above. It is my understanding that running the GEN-SIM workflow twice with the same seed should give exactly the same outputs, modulo changes brought in by this PR. |
|
Hi All, I think that despite the events being the same a change in what goes into the superclusters means that we can end up looking at different jets. I would say that there are no major trend differences w/ and w/o this PR. On the ECAL side we are confident that the new parameters are ok for the superclusters and we will soon have a deeper look at what happens with jets (validation with more events). I think the PR can be merged, keeping in mind that soon we will have a new PR moving these parameters to the database, at which point we will be easily able to fix any problem with the parameters on the fly through a new GT. Best, simone |
|
+reconstruction
|
cmsbuild
added
fully-signed
reconstruction-approved
and removed
pending-signatures
reconstruction-pending
labels
|
This pull request is fully signed and it will be integrated in one of the next master IBs (tests are also fine). This pull request will now be reviewed by the release team before it's merged. @silviodonato, @dpiparo, @qliphy (and backports should be raised in the release meeting by the corresponding L2) |
|
+1 |
This PR updates the hard coded parameters defining the Mustache superclustering algorithm used to cluster electrons and photons in the CMS ECAL. Here the optimization has been performed based on expected Run 3 conditions (but doesn't break anything if used with Run 2 conditions).
A more elegant implementation is underway where these parameters would be loaded from the DB, but this will not meet the last 11_2_X pre-release deadline next Wednesday, hence the hardcoded parameters are being updated.
These parameters have been approved by the ECAL DPG. The details of derivation and validation of the same may be seen in these slides from N. Marinelli and L. Zygala https://indico.cern.ch/event/949294/contributions/3988389/attachments/2091573/3514649/2020_08_26_Clustering.pdf
@nancymarinelli @lzygala @thomreis @fcouderc @simonepigazzini @mdonega