Kalman per-iteration hit timing cuts, plus SVT truth-relation diagnostics - #1175
Draft
mgignac wants to merge 7 commits into
Draft
Kalman per-iteration hit timing cuts, plus SVT truth-relation diagnostics#1175mgignac wants to merge 7 commits into
mgignac wants to merge 7 commits into
Conversation
Make mxTdif a per-iteration array and add three new per-iteration timing
knobs alongside it:
seedTimeSpread gate on tmax-tmin across the 5 seed hits
hitTimeWindow pick up hits within a window of the running mean of
accepted hit times, replacing the tMin/tMax envelope
maxTotalSpread backstop on total tMax-tMin over the track
All three default to 0, which selects the existing behaviour, and
setMaxTimeRange() still sets both iterations, so current steering files
are unaffected. removeHit()/addHits() take max(mxTdif[0], mxTdif[1])
since they have no iteration context.
TrackCandidate.reFit() gains a trial argument to select the time window.
Also expose iteration-1 seed cuts (setMaxKIter1, setMaxdRhoIter1,
setMaxdZIter1, setMaxTanLIter1), which must be applied after the tier-2
setters because those clamp the iteration-1 tier downward; and
edgeTolerance, the seed detector-bounds check that acts as an acceptance
cut at low tan(lambda), previously hard-coded at 1 mm.
KalmanPatRecHPS.debug is now armed per event from KalmanParams.debugEvent
(default -1, off) instead of being a compile-time constant.
Under pulser overlay many raw hits reach readout with no truth relation. Two independent gates can drop one: G1 (totalContrib > 4*meanNoise), which attaches simHits to the channel, and G2 (readoutCuts), evaluated on the combined waveform whose baseline need not match the DB pedestal. Add counters for both, indexed by whether the channel carried a pulser hit, plus a pulser-baseline-vs-pedestal measurement, reported at endOfData(). These are always accumulated and change no behaviour; setDebug() additionally prints per-hit loss records. Add opt-in writeHitOriginCollections (default false), writing SVTHitOriginPulser and SVTHitOriginMCContrib. Combined with SVTTrueHitRelations these resolve "no truth relation" into NOISE, MC_PURE, MC_PURE_SUBTHRESH, PULSER_PURE, MERGED or MERGED_SUBTHRESH, separating a lost relation from an absent one. Enabling it widens the output MCParticle collection to cover sub-threshold contributors, so output is only bit-identical with it disabled.
Two analysis drivers consuming the provenance relations written by SvtDigitizationWithPulserDataMergingReadoutDriver; both require writeHitOriginCollections=true. SvtHitProvenanceDriver answers the aggregate question: are the untruthed hits on a track adjacent to channels that did receive MC charge? Since high occupancy makes adjacency alone meaningless, it is measured against the baseline rate for any pulser hit in the event. SvtEventForensicsDriver dumps named events hit by hit, separating a track whose particle never produced a truthed raw hit from one where truthed raw hits existed but pattern recognition used different ones.
Count events by the reason they were kept or dropped (EPICS, scaler, pulser, FaradayCup, missing TSBank, other) and print a summary at endOfData(). Optional per-event logging via debug=true in the steering file. Pulser and FaradayCup were tested in a single condition and are now separate branches so they can be counted apart; the accept/reject outcome is unchanged.
The extra tag followed the v0 skim driver block and did not match any open element.
…3.2) Shade plugin output picking up the junit version already declared in the parent pom; no source change.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds per-iteration control over the Kalman hit timing cuts, and the
diagnostics used to work out why tracks lose their truth relations under
pulser overlay.
Kalman pattern recognition
mxTdifbecomes a per-iteration array, joined by three new per-iterationknobs:
seedTimeSpread— gate ontmax-tminacross the 5 seed hitshitTimeWindow— pick up hits within a window of the running mean ofaccepted hit times, replacing the
tMin/tMaxenvelopemaxTotalSpread— backstop on totaltMax-tMinover the trackAll three default to 0, selecting existing behaviour, and
setMaxTimeRange()still sets both iterations, so existing steering filesare unaffected.
TrackCandidate.reFit()gains atrialargument toselect the window;
removeHit()/addHits()use the looser of the twotiers since they have no iteration context.
Also exposes the iteration-1 seed cuts (
maxPtInverseIter1,maxD0Iter1,maxZ0Iter1,maxTanLambdaIter1) andedgeTolerance— the seeddetector-bounds check, previously hard-coded at 1 mm, which acts as an
acceptance cut rather than a quality cut at low tan(lambda). Seed-by-seed
tracing is armed per event from
debugEventinstead of a compile-timeconstant.
SVT truth-relation diagnostics
A raw hit can lose its truth relation at either of two gates: G1
(
totalContrib > 4*meanNoise) or G2 (readoutCuts, evaluated on themerged waveform, whose baseline need not match the DB pedestal).
SvtDigitizationWithPulserDataMergingReadoutDrivernow counts both,indexed by whether the channel carried a pulser hit, and reports at
endOfData(). No behaviour change.Opt-in
writeHitOriginCollections(default false) writesSVTHitOriginPulserandSVTHitOriginMCContrib, which together withSVTTrueHitRelationsresolve "no truth relation" into NOISE / MC_PURE /MC_PURE_SUBTHRESH / PULSER_PURE / MERGED / MERGED_SUBTHRESH —
distinguishing a lost relation from an absent one. Enabling it widens the
output MCParticle collection, so output is bit-identical only with it off.
Two consumers:
SvtHitProvenanceDriver(aggregate — are untruthed trackhits adjacent to MC-charged channels, against the event baseline?) and
SvtEventForensicsDriver(per-event hit dump separating "digitizationdropped the relation" from "patrec used different raw hits").
Also
UnbiasedTriggerFilterDriver: accept/reject counters and summary;pulser and FaradayCup split into separate branches for counting, same
outcome.
</driver>in the pass5 physics recon steering file.detector-model/dependency-reduced-pom.xml.