Minutes of 2008-Aug-28 S5 QPipeline Review Teleconference
Shourov Chatterji, Jonah Kanner, Isabel Leonor, Dave Reitze
Minutes by Dave
Recording available at:
Detailed Minutes: The main agenda item today is to develop a time table for the 'in-progress' and 'still to be done' items in Jonah's table http://www.lsc-group.phys.uwm.edu/bursts/review/projects/s5-qpipe/summaryTable.html 1. Table element - H1H2 Coherent Transform shows 'in progress'. The remaining task is for the reviewers to go back over the implementation documentation one last time. Shourov points out the documentation is not up-to-date - the flow chart needs to be updated. the TechDoc link is not correct, and some of the documentation needs to be modified/augmented. J: this element should read that 'we are waiting for documentation' S: In the table, "**Current implementation is now documented in Tech Doc" is not correct; action item for me. Target date: realistically two weeks after Amsterdam L-V meeting. J: can Isabel and I look at the code and try to get what we need that way, by sort of reverse engineering? And then add some documentation. Is that crazy? S: not completely crazy. Relevant codes are 'Q-transform' (single) and 'Q-colocated transform' (multiple). Could do a diff to see differences. Section to look at is 'construct coherent signal sum terms', pretty straightforward if you wanted to look at it. J: I might want to do this; don't want to commit to it yet though. This issue will come up again in the post-processing scripts. 2. Table element - Coincidence. Shourov needs to include the freq. coincidence window values in the documentation, but the main item is that the reviewers need verify that the post-processing scripts are correct and document it. (Currently, we think 'probably') J: link to Shourov's code. Shourov and I have been in touch on this. The codes exist here at UMD. Shourov has written C++ functions and shell scripts. JONAH - PLEASE ADD YOUR NOTES!! These functions and scripts are being used by the HF Search (Brennan). One of the functions 'detect' hasn't been tested, but I can do what Brennan has done - make up test inputs and check that the outputs are correct. Pretty straightforward, in CVS, well documented. On top of this there are shell scripts, 6 of them, that rely on the C++ functions. Not so easy throw in inputs and see outputs. Better to check code line-by-line. D: How many lines of code? Is it possible? J: A page per script, not really sure, though. J: not sure! Never done this before... I: the right way to do this is look line by line to get a feel for the code and catch errors, and then put in a known input and convince yourself that the output is correct. J: I could pick two specific injection of high and low amplitude and try to pass these through each script. I: Doesn't sound like it's too long, not a significant amount of code. J: I agree, good idea to put in test inputs. S: I checked on the length - roughly 60-100 lines depending on script. J: I estimate it would take me a couple days of work. D: you have until 2 weeks after the L-V meeting! J: Right! D: Why not focus on this rather than try to reverse engineer the analysis code and document it. J: Yes, you talked me out of it. I can commit to doing the shell scripts, but not look at the coherent transformation. I: I can a crack at the H1H2 Coherent Transform documentation. 3. Table Element Thresholds - Johan did a nice estimate of the hrss vs the Q-pipe 50% effic. Works quite well except at highest (1797) and lowest (100) freqs. This looks pretty good. But why is the high f off by 21%? S: I think I understand. Related to bandwidth of injections. At edges, the Q3 are pretty wide bandwidth, and therefore the some of the freq content of the signal falls outside the bandwidth of the search. J: Note that I used the median curves that Shourov sent around. I: On p 3 the presentation you sent around, how did you do the first point "The H1H2 hrss50% point for the injections is predicted from the set thresholds and median noise curve for select freq."? J: Look on slide 2, two equations (actually the same equ), a relationship between the hrss of the signal and the PSD of the IFOs. Z is the normalized correlated energy, the output statistic of Q-pipe, and corresponds to the threshold. I took the Z_th, and the actual median noise curve for the PSD, and the hrss of the signal. We know all of these. I: on p 4 for the errors, why is there better agreement at the sweet spot? D: Why higher threshold at lower freqs? S: The instrument is glitchier there. Different thresholds at low frequency, so threshold is higher. Go to last bandwith of: http://ldas-jobs.ligo.caltech.edu/~qonline/sensitivities_s5_year1/ This is how to set the SNR threshold. Look at Q3 (red), get pretty big at high f, whereas the green (Q9) are lower bandwidth and less impacted. Note also can get ratio of thresholds from these graphs. Note, all pipelines use 200 Hz threshold. J: my estimates are pretty simple - no bandwidth in my analysis. D: So I think we're happy with Jonah's estimates for H1H2. What about H1H2L1? Should be easy to do? J: No! I don't have an easy algorithm since thresholds are not simple straight lines in the L1 normalized vs H1H2 correlated. Vertical line in H1H2 correlated energy is easy, but the slanted line is not easily analytical related to an SNR. S: I can think about how to do this, too. J: How different are the triple coin 50% hrss from the H1H2 results? S: Good question. J: My feeling is that is good enough; this is a sanity check, and it worked. D: Could estimate the threshold by just assuming a straight line of constant H1H2 correlated energy? J and S: Yes, possibly. J: Could do at H1H2 at SNR of 15 and 50 (50 is already done actually). OK, good, I'll do that. Simple, can be done quickly. 4. Table Element - Examine Loud Events - action item is for Shourov to make table of loud events based on coherent energy. NOTE: NEEDS TO BE CORRECTED IN SUMMARY TABLE; CURRENTLY SAYS CORRELATED D: Shouldn't take a lot of time. Perhaps before Amsterdam meeting? S: Actually, the action item should read to sort on coherent energy. The table already has correlated. Correlated energy is more important. Script to generate table is easy to modify. Can do right after f2f meeting. S: Please include the timelines on the tables. 5. Table Element - Outstanding Action Items, all for Shourov. S: calibration check may take some time. 'Make Q-pipeline only upper limit' is basically done, I think. Look at Laura's web page and get the efficiencies for Q-pipeline, multiply by 2.3 and divide by livetime. D: Can you add the link to Laura's page? J: Don't have it. S: Will send the link. Timelines: Plots - haven't thought about it. Can't get to it til after the Amsterdam meeting. MDC calibration check - need to run code, will take a week to run. Will have to be postponed til after the Amsterdam meeting - three weeks after the meeting. Q-pipeline UL - basically done, just need to do final calculation, quick, anyone can do it. Plot statistics in terms of event number in addition to event rate - not too difficult. Before Amsterdam meeting. S: Documentation is the highest priority? D: Post-processing script check important, too. 6. Table element - Write review document D: Started writing it, made a little bit of progress, but not a lot. Goal is to have the review document move at same pace as other elements. Should be ready for the Burst group by mid-October. May slip, but no later then end of October. How far along is the cWB review? J: Look at the review document, still work to be done. I: Also AstroBurst review. Not sure where that review stands, but still holding telecons. D: Next week, let's look at review document and assignments - and time table for getting them done. Will circulate minutes and review document. Can we move telecon? J: Conflict! D: Will do a Doodle poll to find a new time.
Note's on Post-Processing Scripts
The Q Pipeline post-processing rerun live on the uwm cluster:
in this directory: /scratch3/shourov/s5_qpipeline_rerun_postprocess/
The "post processing code", written by Shourov and checked in to CVS, that is used are detect, select, mask, and coincide.
Basic sanity checks on select, mask, and coincide have been performed by Matt Pitkin and Brennan Hughey in the course of the high freq. analysis. These check are linked from the review summary table.
I have not seen any sanity checks on detect. It should be straight forward for me to perform some simple tests.
In addition, there are six shell scripts that are used to execute the appropriate commands:
coincide.sh, cut.sh, detect.sh, mask.sh, select.sh, and collect.sh
These shell scripts are called by a wrapper for each waveform.
I'm not sure what is the best way to review these fairly simple shell scipts. Maybe the thing to do is go through them by eye, and check for bugs. I am open to suggestions.
Note: the steps are as follows (from documentation in same directory) 0. collect 1. downselect overlapping triggers 2. apply final h1h2 and l1 cuts (part of select script) 3. mask out category 2 and 3 data quality and vetoes a. apply mask b. skip mask 4. coincide h1h2l1 a. apply mask b. skip mask 5. apply final h1h2l1 cuts a. apply mask b. skip mask 6. compare aginst list of injections a. apply mask b. skip mask
h1h2 in h1h2l1 time
h1h2 in h1h2 only time
- Look at 100 loudest events: October 3
- Write review document based on assignments: Now through Nov.1
- Make Q-pipeline only upper limit: September 19
- Verification of Post-Processing Shell scripts: Oct 15
- Rough H1H2L1 Efficiency estimate: Sept. 29
- Add Laura's link to "oustanding action items": Now would be good
- Document H1H2 Coherent Transform: Oct 15
- Dcoument Coincidence - freq. coincidence window values: Oct 15
- Provide 100 loudest events sorted by coherent energy: Sept 12
- Make plots suggested by Isabel's first comment : Oct 22
- MDC study of calibration uncertainty: Oct 22
- Plot statistics in terms of event number in addition to event rate: Sept 19