Minutes of 2008-May-08 S5 QPipeline Review Teleconference
Shourov Chatterji, Jonah Kanner, Isabel Leonor, Dave Reitze
Minutes by Jonah Kanner.
Isabel's list of questions for the review
These comments refer to plots and data given in the following link: http://ldas-jobs.ligo.caltech.edu/~shourov/s5qreview/triggers/s5/box.html 1) For injections, there are outlier events with large H1H2 coherent energy but with small correlated energy. These events fall well outside the main distribution and below the chosen thresholds. There also seems to be more of these events for f <= 200 Hz. Can we have more information about the actual and recovered frequencies of these events (and maybe other morphology)? I think Shourov has previously suggested that these outlier events might be due to injections with very small antenna factors. In this or any case, why would these events have large coherent energies (some as large as ~1000)? In light of these injection outliers, what can be said about the raw data, zero-lag "outlier" in the plot of H1H2 correlated energy vs. coherent energy (f < 200 Hz) that was pointed out by Dave in a previous telecon? Where do these injection outliers fall in the plot of L1 normalized energy vs. H1H2 correlated energy? In relation to this, I would like to request the following plots for the simulations: a) Plot of H1H2 coherent energy vs. LHO antenna factor. b) Plot of H1H2 correlated energy vs. LHO antenna factor. c) Plot of L1 normalized energy vs. LLO antenna factor. d) Plot of L1 normalized energy vs. LLO antenna factor. e) Plot of L1 normalized energy vs. H1H2 correlated energy, but with the injection outliers in question distinguished from the main disbribution. f) Plot of L1 normalized energy vs. H1H2 coherent energy (real data and injections). Again, please distinguish the injection outliers. 2) The frequency boundary for the different thresholds was chosen to be 200 Hz, with a higher threshold for f <= 200 Hz. Given that the sweet spot for S5 is at or near ~150 Hz, does it make sense that the higher threshold includes frequencies near and at the sweet spot? How was the choice of 200 Hz made? Is this the default value used by the burst group? Is it worthwhile to rethink this choice? Is it because of the frequency-dependent threshold that the efficiency of the pipeline is best near 235 Hz instead of 150 Hz, and that for some waveforms, the efficiency at 153 Hz is comparable to the efficiency at 361 Hz? Is the same pattern seen by other pipelines after applying frequency-dependent cuts? For convenience, below are tables of the sine-gaussian efficiencies which I copied from Shourov's web page, but sorted according to increasing frequency. H1H2L1 detection efficiencies H1H2 detection efficiencies ----------------------------- --------------------------- HSG1_S5_R2 hrss50 hrss90 HSG1_S5_R2 hrss50 hrss90 ========================= ========================= SG70Q8d9 3.1e-21 1.4e-20 SG70Q8d9 3.9e-21 2.1e-20 SG100Q8d9 1.3e-21 5.9e-21 SG100Q8d9 1.6e-21 8.4e-21 SG153Q8d9 8.3e-22 3.6e-21 SG153Q8d9 9.9e-22 5.3e-21 SG235Q8d9 6.3e-22 2.5e-21 SG235Q8d9 8.1e-22 4.0e-21 SG361Q8d9 1.1e-21 4.2e-21 SG361Q8d9 1.4e-21 7.0e-21 SG554Q8d9 1.2e-21 4.8e-21 SG554Q8d9 1.5e-21 7.6e-21 HSG5_S5_R2 hrss50 hrss90 HSG5_S5_R2 hrss50 hrss90 ========================= ========================= SG70Q3 2.5e-21 1.1e-20 SG70Q3 3.2e-21 1.7e-20 SG100Q3 1.3e-21 5.8e-21 SG100Q3 1.6e-21 8.0e-21 SG153Q3 9.0e-22 3.9e-21 SG153Q3 1.1e-21 5.5e-21 SG235Q3 6.7e-22 2.6e-21 SG235Q3 8.5e-22 4.3e-21 SG361Q3 9.5e-22 3.7e-21 SG361Q3 1.1e-21 5.8e-21 SG554Q3 1.2e-21 4.9e-21 SG554Q3 1.5e-21 7.7e-21 HSG6_S5_R2 hrss50 hrss90 HSG6_S5_R2 hrss50 hrss90 ========================= ========================= SG70Q100 4.0e-21 2.9e-20 SG70Q100 4.9e-21 8.0e-20 SG100Q100 1.5e-21 8.1e-21 SG100Q100 1.8e-21 1.6e-20 SG153Q100 9.8e-22 5.0e-21 SG153Q100 1.2e-21 7.5e-21 SG235Q100 7.7e-22 3.2e-21 SG235Q100 9.4e-22 4.6e-21 SG361Q100 1.3e-21 5.1e-21 SG361Q100 1.8e-21 8.6e-21 SG554Q100 1.4e-21 5.6e-21 SG554Q100 1.7e-21 8.3e-21 3) What happened to the planned significance calculation and remapping? Was remapping found to be unnecessary after all? 4) (For this item, suppose that the QPipeline results will not be combined with results from other analyses.) How would the QPipeline S5 first year results be structured for a paper? What quantitative conclusions can be made from these results? It would be good to have sample upper limits derived from the results so we can evaluate if the limits make sense on their own, independent of combining with other analyses.
Action items from Dave and Jonah's questionsShourov
- Fix broken documentation links
- Cleanup, update, and fix technical documentation
- Links to discussion of clustering
- MDC caliibration uncertainty study
- Plot statistics in terms of event number in addition to event rate
- Run H1H2 vs. L1 time lags at irrational time steps
- Improve documentation of whitening filter issues
- Document how vetoed area increases with null stream significance
- Complete code review
- Validation of post-processing code: compare with other codes?
- Request non-linearly polarized injections from the burst group
Dave: To Do: Isabel's comments Review status regarding action items from last week Isabel, let's walk through your list Isabel: I was looking at the open box results. Item 1) pertains to injections Shourov claimed that outlier injections are related to weak antenna patterns. Can we see that plot? Shourov: Could be an interaction between noise and injections: glitch + injection gives a large coherent but low correlated. For the H1H2, sky position does not play a role. For L-H, sky position DOES matter. The small misallingment makes some funny points Isabel: How does the number of green dots compare to the black S: There is nothing in these plots that enforces that the green are also injections In the past, I have tried to make plots where I enforce only the loudest trigger for each injection - that removed a lot of triggers in this region. I'll try to identify a way to look at these. Isabel: Were these time shifted? S: No. I: Are these all the waveforms lumped together? S: No. These are HSG1 set - it may be a subset of amplitude. I'll try to go back and relabel these plots with more detail. S: I guess 1) has 2 components: the h1h2 and the triple coincidence. For H1H2 it should not be antanna patterns. maybe the thing to do is run Q-event for some of them and see what they look like. i could also try picking out the point with the highest energy for each injection and see if it cleans up these plots. Maybe we should make it an action item to do Qevents of 100 of these injections and also background events with high coherent energy I: The outlier in the H1H2 low freq plot - was it thrown out by any cuts? Shourov: If you look at the bottom of the page, there is a table with corr energy as ranking. Maybe it would be good to do this using coherent energy, and also for background events. Then we could do it for injections too, picking injections with high coherent and low correlated energy. I: For the antanna factors, it might be better to plot the hrss detector S: sure, that should be available in the mdc log file I: let's move to 2). Why 200 Hz for the boundary? Your most sensitive at 250 Hz instead of 150 Hz. S: I think it has become a default for the burst group in the sense that block normal and waveburst are also using. it is motivated by the low freq glitch population. i think i looked at 150, and there you start to push into the glitch population. this is very course - 2 chunks. you could imagine a finer scale, which is what the remapping was intended to do. I: So, for other searches, they see this too? that is, they are most sensitive at some frequency above 200 Hz Dave: The argument would be the instrument is more glitchy at low freq, so you get better sens. at higher freq. however, i don't think that's true for cWB. Does anyone know? Shourov: i don't think a lot of work went into choosing 200 Hz - other than looking at scatter plots and saying 'that's about right'. you'd have to go to the wave burst page to see what the values were S: I think we talked about 3). For number 4),I think that wouldn't be too hard to do. There are some things I would need to do - i didn't really do any sigmoid fitting. I need to run Laura's code. J: What is Laura's code? S: scripts for sigmoid fitting, sigmoid errors, and other things to produce a final upper limit. In the past, laura was the one who produced those final upper limit curves. For S5, we have Patrick's method, but we will still need to fold in uncertainties etc. D: How much effort would it be for you to dust off laura's code. S: I can ask Laura to do it. D: Is this quick? S: Yes, but I could also take what is shown here, and get a rough upper limit curve D: That might be adequate. I: Yes. Just so what goes into upper limit calc. are things which actually come out of the pipeline Dave: Maybe we can talk about action items and where we stand S: I haven't made any progress on these. I'll endeavor to make progress over the next week. Other items, such as rerun and writing, are taking a lot of time right now. D: Can you remind me of the time frame for the S5 1st year paper? S: The plan was to have a draft by Orsay. This might not be a reality. There is a lot of work to do on the re-run. There is a call tomorrow to discuss. I: Why rerun? S: To combine results w/ same amp. injections D: Code review? I: I'm not done yet, but maybe this weekend. D: Who else is on that? I: Lindy and Shantanu S: It might be a good idea to remind them of this. D: Under the heading "all", validation of post-processing code. What's the plan? S: I put it under all for other peopele to think of ideas. One idea is that in current astrowatch, Erik is running coincident code. Maybe Erik's code can be run on these triggers, and try to reproduce results. D: Non-linearly polarized injections. J: I think we should remove that item. D: Ok, I agree. D: Anything else that we didn't cover, that we might want to add to this list. No? Ok, then the road is clear. Next week, I'll be travelling. Let's tentatively say we will NOT have a telecon unless you hear from me. I'll let you know by tuesday. D: I'll send a link to the recording when avaliable. Bye.
Make the following plots for the simulations: a) Plot of H1H2 coherent energy vs. LHO antenna factor. b) Plot of H1H2 correlated energy vs. LHO antenna factor. c) Plot of L1 normalized energy vs. LLO antenna factor. d) Plot of L1 normalized energy vs. LLO antenna factor. e) Plot of L1 normalized energy vs. H1H2 correlated energy, but with the injection outliers in question distinguished from the main disbribution. f) Plot of L1 normalized energy vs. H1H2 coherent energy (real data and injections). Again, please distinguish the injection outliers. Note: For plots above, may be better to use detector hrss rather than antenna patterns Q events for 100 high coherent/low corr. energy injections Re-plot low freq. injections, keeping only 1 trigger for each injection and see if it "cleans up" the outlier injections Note: The goal of many of the above items is to understand the origin of the "outlier" injections in the low-freq., H1H2 plot, as well as the ouliers in the triple coincident plot. Check wave burst sensitiviy: is it best at some freq. above 200 Hz or below? Q events for loud events (both loud corr and loud coherent) Table of loudest events ranked on coherent energy Use known information to estimate upper limit curves from Q-pipeline only and/or run Laura's upper limit code