- Likelihood ratio, background and foreground sampling
- Optimization and throughput benchmarking
- Data format wrangling
- Enabling running on OSG/IGWN grid
- DQ dev
- HM search
- Exploratory development (5 minutes)
- Misc projects
* Operations (5 minutes)
* LL CBC operations Divya : nodes were broken over weekend but they are fixed. Becca has worked on having multiple aggregater for non-inj and inj jobs. Surabhi : I saw the same issue should I just relaunch the analysis? Divya : yes
* Offline CBC operations
* LL IDQ operations no update.
* O4 Dev (30 minutes)
* Low latency integrated testing and Monitoring Becca : update LL tests suite to connect to igwn alert. Mostly got it working. The results look consistent with gracedb
* Template bank CIT shows file_trasfer error and reached out to admin. LLO, LHO still have some errors in a dag. bank sim for negative large spin cut has been done.
* Likelihood ratio, background and foreground sampling chad : about LR dev, we started re-working LR code. We cannot guarantee putting this in production soon, but the idea is factorize each term as a node/method in the joint prob of likelihood.
* Optimization and throughput benchmarking Alex : some runs are going. Cleaned up benchmarking repo and added docs. Chad : would you able to present the opt results for SVD files in LL ops call today? Alex : yes
* Data format wrangling Chad : we will discuss the sim_inspiral in east call.
* Enabling running on OSG/IGWN grid Divya : I have some dag to launch, will work on that today.
* DQ dev none
* HM search Stefano : nothing
* Exploratory development (5 minutes)
* Misc projects
* Focus session
- Aaron Zimmerman on GstLAL chi-square | [slides](https://dcc.ligo.org/G2200635)
Aaron : Definition about auto-corelation should be pure real version. Hope that is not used in the code.
Chad : no, I don’t think it is used in the code.
Aaron : If you take the pure real version of chisq, that ends up being not what you want to do in the test.
Chad : the blue trace in the slide4 is the real part after rotating by some phase that maximize the SNR peak.
Chad : that was off-the-shelf thing we were aware. We were hoping that was a good approximation. It would be reasonable to tune params so that it is still a good approx.
Aaron : to be fair, at least that would be good to be described somewhere.
Chad : Cholesky transformation is effectively same as fourier transformation whose bases function is circualr function.
Chad : Gaussian approx would be dream. We could model the mis-match and put that into machinary to get the better model. While we are doing this correctly is good, it is clear that df is a function of physical param but we don’t understand it well. If we can work out more rigorously, that would be great.
Aaron : when conditions like PSD changes, we can go back to U-matrix and apply the change and follow the same procedure.
Aaron : it would be good to have a plan to describe what has been done with chi-square.
Chad : As a concrete suggestion, it would be good to have someone work on auto-correlation matrix and derive mean and std for approximated Gaussian distribution.
<09:48:28> "Shio.Sakon": the number of templates is 881478 for the limited negative spin template bank. (for comparison, the template bank with the same configuration, but without the large negative spin cut had 1326927 templates)
<09:48:46> "Shio.Sakon": this is the bank sim: https://ldas-jobs.gwave.ics.psu.edu/~shio.sakon/O4banksim/python3_limited_negative_spin_nbank50_broad/
<09:49:02> "chad.hanna": agreed
<09:49:06> "ryan.magee": +1 . I don't know the proposed one at this point
<09:49:14> "Divya Singh": +1
<09:49:14> "SurabhiSachdev": @shio is there a place where all the proposed banks are documented?
<09:49:17> "Shio.Sakon": the new limited negative spin template bank and mass models are here; https://git.ligo.org/gstlal/offline-configuration/-/tree/o4-template-bank-dev/o4-template-bank-dev
<09:49:47> "Rachael Huxford": Maybe we can switch Shio's focus session slot with Victoria and get an update next week on template bank?
<09:49:56> "ryan.magee": also @shio are you planning to do a bank sim with more than 300 inj at some point? Just so we have a better idea of how well it covers the space?
<09:50:03> "Shio.Sakon": @surabhi @ryan @becca the proposed tempalte banks are https://git.ligo.org/gstlal/offline-configuration/-/tree/o4-template-bank-dev/o4-template-bank-dev/bank
<10:13:07> "jolien.creighton": (historical note: Bruce's "r^2" stat was not two-phase, and was not chisquared distributed -- the test described in Allen 2005 is what I proposed)
<10:13:14> "chad.hanna": hand up
<10:15:37> "jolien.creighton": the gstlal reviewers "complained" about this too... which is why it is called xi^2 rather than chi^2
<10:15:56> "chad.hanna": yes
<10:15:58> "chad.hanna": indeed
<10:16:03> "chad.hanna": more than fair
<10:22:09> "Leo Tsukada": hand up in the end of this slide
<10:22:10> "chad.hanna": hand up
<10:24:31> "Leo Tsukada": yes
<10:25:44> "Leo Tsukada": yes, thanks!
<10:28:14> "chad.hanna": quick response to that if I can
<10:29:01> "Leo Tsukada": hand up
<10:29:33> "Leo Tsukada": hand down
<10:30:46> "Leo Tsukada": hand up about singular-ness
<10:32:20> "chad.hanna": SVD?
<10:34:55> "chad.hanna": hand up
<10:38:37> "chad.hanna": agreed
<10:42:54> "chad.hanna": exactly
<10:43:47> "chad.hanna": when the SVD is made we have all that info availavle
<10:45:46> "Leo Tsukada": Thank you Aaron, we will get back to you soon!
<10:45:47> "Aaron Zimmerman [h]": my guilt is less
<10:46:05> "chad.hanna": hand up
<10:46:09> "Shio.Sakon": @ryan @surabhi @becca This is the quick overview of the template banks we have: https://git.ligo.org/shio.sakon/o4-template-bank/-/blob/master/README.md
<10:48:04> "Leo Tsukada": Andre and I can work on the latter, but I guess Gaussian approx would be interesting too
<10:48:15> "Leo Tsukada": I will ask him if he is intersted in that too
<10:49:08> "chad.hanna": yes, def.
<10:49:25> "chad.hanna": great
<10:50:04> "chad.hanna": hand up for concrete sugg
<10:52:01> "Leo Tsukada": Andre has already got some part of it coded up.
<10:52:08> "Leo Tsukada": okay
<10:53:05> "Aaron Zimmerman [h]": that all sounds fine to me
