Minutes of 2008-Mar-10 S5 QPipeline Review Teleconference
This telecon was rescheduled from March 6, 2008. The call started at 11:00 Eastern.
Shourov Chatterji, Jonah Kanner, Dave Reitze
Minutes by Jonah Kanner.
Dave: should we look again at coincidence window lengths, as suggested by erik? Shourov: no, i think it's fine. Erik's concern was based on how other pipelines do coincidence. for example, a white noise burst with a 1 s duration could have peaks at different times at different detecotors. so, clustering algorithms might fail. i'm of the opinion that clusters should not be used for testing coincidence. If something is 1 s long, all of those tiles are tested for coincidence. Corrected link: ldas-jobs.ligo.caltech.edu/~shourov/s5qreview/production_playground/old/H1H2L1/window/ Reviewing these tests, it seems clear that the 10 ms window is fine. Dave: Coming back to white noise burst waveforms, a long waveform does show increased efficiency as the window increases: why are they "worst case"? Shourov: Because they are long duration and high bandwidth, so hard for coincidence Dave: what does "minor improvement" mean? Shourov: Scroll down. It is of order 1 more injection out of ~38 injections x multiple amplitude scales Dave: Perhaps you could quantify the rate of "accidental coincidences" ? Shourov: hmmm...that might be hard - it's probably wave-form dependant. Shourov: Try adding /report.txt to the web address Dave: there are 100's in some burst tables, but not others Shourov: it seems consistent with the frequency span of the waveforms: more freq. space suggests more accidental coincidences. dave and jonah: ok, we're convinced this is ok Shourov: status: did trigger production for h1h2 zero lag. if we want to test zero-lag background h1h2 time-shifted with L1 jonah: does that run a danger of 'peeking' in the box? shourov: well - maybe. if you do limited time lags, it might be ok. one reason i moved things around is to clean up go to ~shorov/s5qreview/triggers/s5 shorov: these are various levels of trigger posproduction. step 5 is coincidence. go to 5/H1H2L1/RAW some interesting things are that for high scale, you get more triggers than you did for singe detector H1L1 and H2L1 are options, but they are low priority. Re-mapping is step 6 that is in progress One issue in previous versions of code is that i was using H1H2 coherent as measure of significance - one consequence was that mainly H1 level triggers were left over, and those depended very little on time slides - so time slides were not independent Now that significance is based partly on correlated energy, as you time-slide time shifts are more independant, and you can use the time-shifts to train the re-mapping The only worry is that real GW's would show up in time slides, and so devalue themselves I've played around to convince myself that the issue is better - i hope to show that, using the playground day, the correlated energy is better with the re-mapping. Shourov: go to /s5qreview/old/scatter/likelihood here we see some notes about a bayesian type approach the problem is that the probabities we get depend on the probabilities we inject the nice property is that in coherent/correlated space, injections lie on a line so, as a measure of 'how GW like', we can use a measure of how far from line in hard-corr space So, i think that's we're things stand. the final step is defining this re-mapping and then i should be ready to open the box. Dave and Jonah: we need time to digest this http://ldas-jobs.ligo.caltech.edu/~shourov/s5qreview/old/scatter/likelihood shourov: ok, well, this is what i'm working on now as well. dave: should we meet this thrusday?? Shourov: i will likely be travelling on thurday dave: ok - then we'll wait to hear from you.