Track-by-track analysis of Victor's proposed tracking code changes
Updated on Fri, 2016-06-24 13:21. Originally created by genevb on 2016-06-23 15:02.
Track-by-track analysis of Victor's proposed changes
Same dataset and methodology as in other recent studies (e.g. 1,2).
I. Victor's changes to Sti vs. DEV:
Fig. 1: Δ(Nhits) for modified matched tracks
Fig: 2: ratio of event reconstruction times for Victor / DEV


II. Victor's changes to StiCA vs. DEV:
Fig. 3: Δ(Nhits) for modified matched tracks
Fig. 4: fraction of tracks modifed vs. q/pT for matched tracks
Fig. 5: Δ(q/pT) vs. q/pT for matched tracks
Fig. 6: Δ(DCA) vs. q/pT for matched tracks
Fig. 7: ratio of event reconstruction times for Victor / DEV





III. Victor's bug fix to StiCA vs. Victor's StiCA
Fig. 8: Δ(q/pT) vs. q/pT for matched tracks
Fig. 9: Δ(pT) vs. q*pT for matched tracks
Fig. 10: ~ratio of event reconstruction times for fixed / unfixed



-Gene
Same dataset and methodology as in other recent studies (e.g. 1,2).
I. Victor's changes to Sti vs. DEV:
- 25 out of 131k (0.02%) matched tracks have differences in hits (and consequently tiny differences in track parameters; more lose hits than gain, ranging from [-7,+1] except for one with +5, see Fig. 1)
- 17 unmatched from DEV
- 7 unmatched from Victor
- No obvious phase space biases for unmatched or modified tracks other than low number of hits
- Victor's appears to run ~4% faster (see Fig. 2)
Fig. 1: Δ(Nhits) for modified matched tracks
Fig: 2: ratio of event reconstruction times for Victor / DEV


II. Victor's changes to StiCA vs. DEV:
- 884 out of 154k (0.6%) matched tracks have differences in track parameters; larger fraction of low-pT tracks (3%) affected than high-pT tracks (0.1%) (low-pT means large magnitude for q/pT as plotted, see Fig. 3)
- Modified tracks have negligible mean change in q/pT and DCA, with spreads of up to ~1 MeV/c and 100 μm respectively at low pT (see Figs. 5 & 6)
- 48 unmatched from DEV
- 21 unmatched from Victor
- No obvious phase space biases for unmatched tracks other than low number of hits
- Victor's appears to run ~4% slower (see Fig. 7)
Fig. 3: Δ(Nhits) for modified matched tracks
Fig. 4: fraction of tracks modifed vs. q/pT for matched tracks
Fig. 5: Δ(q/pT) vs. q/pT for matched tracks
Fig. 6: Δ(DCA) vs. q/pT for matched tracks
Fig. 7: ratio of event reconstruction times for Victor / DEV





III. Victor's bug fix to StiCA vs. Victor's StiCA
- Nearly all matched tracks (129k) are modified
- On average, 0.05 hits lost per track
- On average, ~1.5 MeV/c in pT gained per track, with an odd pT dependence (see Figs. 8 & 9)
- 26k unmatched from Victor's StiCA
- 28k unmatched from Victor's bug-fixed StiCA
- Unmatched fractions higher at large DCA (~30% at DCA of ~2cm!), low pT, and low number of hits
- Bug-fixed appears to run ~8% faster (see Fig. 10)
Fig. 8: Δ(q/pT) vs. q/pT for matched tracks
Fig. 9: Δ(pT) vs. q*pT for matched tracks
Fig. 10: ~ratio of event reconstruction times for fixed / unfixed



-Gene
»
- genevb's blog
- Login or register to post comments