MTD TDR Reconstruction + MTD Track Association BDT #28815
Conversation
… requirement with a beamspot constraint or vertex + path length. Possibility also to use the SimVertex Add configurable parameters, separated for BTL & ETL Addpossibility to reproduce PFPackedCandidates with and error from an external map
|
The code-checks are being triggered in jenkins. |
| #define VAR_ENUM(ENUM) ENUM, | ||
| #define VAR_STRING(STRING) #STRING, | ||
| #define MTDTRACKQUALITYMVA_VARS(MTDBDTVAR) \ | ||
| MTDBDTVAR(pt) \ |
There was a problem hiding this comment.
the arguments didn't need to have MTD in the prefix, but it's fine.
| t_vtx, | ||
| t_vtx_err, //put vtx error by hand for the moment | ||
| false); | ||
| TrackTofPidInfo tof = computeTrackTofPidInfo(p.mag2(), |
There was a problem hiding this comment.
const TrackTofPidInfo tof = or event auto tof =
|
Comparison is ready Comparison Summary:
|
|
it looks like this PR is adding about 5% to the total RECO time in PU200 (in a production setup, writing AODSIM), which is coming from about 5-fold increase in CPU use by the trackExtenderWithMTD (from 5.7 s to ~28 s/ev). [this is still based on a partial set of events in the test that's still running] @pmeridian @kpedro88 please take a note for something to follow up. |
|
+1
|
|
+upgrade |
|
some inclusive plots from the root files for the ttbar PU200 D49 setup, 200 events At the track level, it looks like most of the population changes are in the ETL pf CH times are closer to 0 pf electrons show the most significant effect the time error of the 4D PVs is better (with some migration to the default no-time-set value [displayed at 0]) The time is closer to zero as well with significantly reduced tail on the positive side x (and y) errors are a bit better, but there is also a small component that gets worse At a glance, this all looks like going roughly in the direction of improving the performance. |
|
merge |
|
@silviodonato , this also needs cms-sw/cmsdist#5538 which is not merged yet. I am going to merge it now. You might want to restart 11.1.0.pre3 release to get latest cmsdist |
|
@silviodonato , Ok, I have restarted the 11.1.0.pre3 release jobs , so every thing should be fine now |
|
+1 |
cmsbuild
added
analysis-approved
fully-signed
and removed
analysis-pending
pending-signatures
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 be automatically merged. |
PR description:
Reopening of PR #28279: adds the tuning of the MTD track extender done for MTD TDR (using also a beam spot/vertex time constrain when finding new hits) and add a track quality MTD associator score based on a BDT classifier.
Most of the comments raised for PR #28279 are addressed (not addressing all the style comments made for TrackPIDinfo, as this will need to completely evolve once MTD reconstruction is fully integrated into the tracker)
The TrackPUIDMVA Producer and TOFPIDProducer are now stored into a new package in RecoMTD/TimingIDTools as suggested by @slava77 .
Also the BDT evaluation now used TMVAEvaluator with the GBRForest evaluation.
Future backport into 11_0_X can be imagined once this is fully integrated.
PR validation:
Validated this technically rerecoing 1000 events from RelValZMM_14TeV. To run a real test is required to make a repository to store the BDT xml file (now residing under RecoMTD/TimingIDTools/data). Please create a cms-data repository for this package.
@parbol @hatakeyamak @bendavid @simonepigazzini