Minutes of 2008-Jul-17 S5 QPipeline Review Teleconference
Shourov Chatterji, Jonah Kanner, Isabel Leonor, Dave Reitze
Minutes by Isabel Leonor.
* Announcement Dave: Next week might not be good time for telecon. * Discussed table that Jonah created which summarizes things to be done to finish review: http://www.lsc-group.phys.uwm.edu/bursts/review/projects/s5-qpipe/summaryTable.html Jonah leads discussion. * Whitening Jonah: Concerned this has not been discussed enough. Isabel: What's in Chapter 4 of Shourov's thesis is implemented in Shourov's code. Has also run the whitening code and confirms it does whiten data. Jonah: I need to read and understand Chapter 4 then. Dave: Has read Chapter 4. Jonah only needs to convice himself after reading Chapter 4. Shourov: The whitening implemented is a tried and true technique. Prior implementation has also been used by other analyses, e.g. r-statistic. Dave: What can go wrong? Shourov: Need to pick filter length to be longer than data segment. Narrow lines will still appear if long FFTs are taken. h(t) data has large increase at low frequencies due to seismic wall. This causes dynamic range problem for whitening. Solution is to highpass-filter before whitening. Phase delays are avoided by implementing zero-phase method. * H1H2 coherent transform Jonah: Concerned that implementation is not fully understood by reviewers and that this needs further documentation. Isabel, can you explain this to us? Isabel: Will try to document this. Dave: Shourov, can you add links to documenation to this table? Isabel: Notes that there is already documentation in the technical document link. * Null stream veto Jonah: Wants additional documentation on null stream inflation factor. * Coincidence Jonah: Verification of Shourov's scripts has been done with Laura's scripts. Shourov, walk us through how these were verified. Dave: Are live times the same as for cWB analysis? Same DQ flags? Shourov: There are minor differences due, for example, to differences in minimum segment durations. In combining results, take union, not intersection, of live times. There are only minor deviations. Should be verified. Jonah: We need list of all postprocessing scripts and to do some sort of verification. Shourov, if I poke through the directory [see links in table], can I figure out which scripts were used? Shourov: There is C++ library used for coincidence, overlap, thresholding. There are also simple shell scripts, awk. Jonah: Tempted to poke around and check. I might learn something. Isabel: How many post-processing scripts are there? Shourov: A lot of small scripts. Jonah: A list of these scripts is necessary. Isabel: Agrees that a list is necessary. Are these scripts in CVS? Shourov: Not sure. The rerun (injections) was more organized, so will point to those scripts. Dave: Modify table. Shourov to make sure that scripts are in CVS. * Thresholds Jonah: Sent around document with sensitivity estimate. Not sure if it's right. Shourov sent an email around. Shourov: Use average instead of best performance curves. Did study using 1000 4-sec long segments throughout year, then average spectra. Resulting average was similar to best noise curves, which was surprising. Patrick suggested that, instead of averaging S, take average of 1/S, then calculate 1/<1/S>. Resulting spectrum is a factor of 2 lower. Dave: Try another random 1000 segments. Average shouldn't go down with second method. Shourov: It's possible there was a mistake. Possible option, used in S3 to determine threshold of detectability: Look at each injection and calculate power spectrum at time of injection, and evaluate SNR integral for that injection. Then create table of detection/non-detection; this determines SNR threshold. Dave: Regarding Jonah's calculations, numbers are not crazy. SNR of 12. Jonah: Overestimated H2. Shourov: Might also look at documentation for coherent follow-up. Jonah: One question regarding QPipeline normalized energies. Are correlated energies normalized by coherent energy values? Shourov: Correlated energy is not like other energies. Can't answer that question right away. Jonah: Should get written down somewhere. * Loud events Jonah: What's on table? Dave: Why the loudest events table? Shourov: Had initially planned loudest event analysis instead of using threshold. Part of motivation was lost when we switched to counting statistics. All: We all agree that manually examining a table of 100 loudest events is excessive and unnecessary. A table of the 10-20 loudest events is more reasonable. * Outstanding items Jonah: At some point, we need to see if we still care to get these things done. What do we need to do with frequency-dependent results? Isabel: I think the frequency-dependent results make sense. Once thresholds are chosen, what we get is what we get. Just need to compare these results to those of other analyses and point this out to general review committee. Jonah: We need to go back and see what we really want Shourov to do. Dave: Propose different meeting time next week: Monday, July 28, 1 PM PDT. I'm working on assignments for review document. Will circulate next week.