- Chad: To amend the documentation to warn about fixed SNR threshold of 4. At some point in the future consider making SNR threshold an adjustable parameter
- Kipp: Need to fix distance quantization for the construction of SNR PDFs
- Chad: You have a hard-coded 4096 Hz, find out where that is; need to have a variable sampling rate (perhaps this is not hardcoded any more???).
- Chad: Time-dependence of horizon distances can be captured in the computation of the 'numerator'
\section Action Items:Makefile:
- Chad: Indicate that overlap should be 10% of the num_split_templates
- Chad: People shouldn not have to remove Multicore=True
- Chad: The length of autocorrelation chi-squared in sample points is set to AC_LENGTH = 351. It might be worth exploring how the sensitivity volume changes by varying AC_LENGTH
- Chad: Please specify the boundaries of the various parameters in Makefile, especially if it is desirable to explore how the sensitivity changes for different values of these parameters
- Chad: Makefile for NSBH and BNS-spinning on the documentation page all point to the BNS Makefile; please correct them.
\section minutes minutes
- Looked at the unit test. The output graph looks OK.
- snglconic.py:
-- Looked at init function and dictionary
- far.py:
-- It would be good to histogram the distances as a function of GPS times
-- Sathya: How many places is the SNR threshold is hard-coded?
-- Kipp/Chad: In three places, DAG script, this code and binning
-- Chad: Motivation for SNR threshold of 4 was to get 1 trigger per second and so it would not be any worse off even if templates are short and look like glitches
-- Kipp: For the early release use a fixed SNR PDF using median PSD which could be pre-computed and stored on disk. (Store database of SNR pdfs for a variety of horizon)
-- Jolien: Could use simple scaling.
-- Sathya: Plots of SNR in different instruments do not really look simple enough to do simple scaling
-- Florent: The binning parameters are hard-coded too; Could it be a problem
-- Kipp: (What was your precise response Kipp?)
-- Jolien: Line: 640: Why not use finish() method inside iadd().
-- Kipp: It is too time consuming
-- A bunch of questions on how the likelihood is calculated. We need to better understand this; to be presented at the Rates and Significance call next week.
-- Jolien: What changes when you go to high masses is not the prefactor but the number of bins.
-- Jolien: Suggests that time-dependence of horizon distances can be captured in the computation of the "numerator"
-- Jolien: Example in 1165 could print the output.
-- Kipp: Chisquare binning has not been tuned to be a good representation of the PDFs; could be improved in future
- Makefile:
-- Kent: what changes are needed to get a new approximant to work:
-- Chad: Make changes to line 65/66 templates.py
-- Note that some of the variables in the Makefile are defined in Makefile.offline_analysis_rules
-- --sort can choose almost any parameters in the template bank params (masses, chirp mass, etc.) Changes would be needed in gstlal__bank__splitter_source if you need to sort on other parameters.
-- --bank-cache You would need to know the string that specified the way the bank was made
-- --peak-time used to be 8 s but for MDC it is turned off (i.e. uses 0) (computational cost could be reduced by tens of percents by using a peak time of 4s or 8s without significantly changing the efficiency).
