011723.md

@@ -15,41 +15,105 @@ Shio
 Chair / Minute Taker / Focus Session [Rota](https://git.ligo.org/groups/gstlal/-/wikis/Meetings/West-call/Rota-table)
 
 ## Action Items for next week
-
+* [ ] Divya will present the test suite at the upcoming R&P call to make a case for real time R&P injections.
+* [ ] Assigned to Cort/Leo : get the "file list too large" fixing commit on O4-offline-dev merged into master
+* [ ] offline meeting with Chad, Anarya, Becca, Yun-Jing for snr maximization in composite event is the right thing to do. Anarya to schedule the meeting ([poll](https://www.when2meet.com/?18322371-Etkpw))
+* [ ] Leo, Becca, Divya, Rachel: In the next couple weeks we should make a new template for the west dev call with updated projects 
+* [ ] Cort: reach out to Shio for injection distribution plotting script
+* [ ] Leo, Stefano: Work on installing latex dependencies for plotting job failures. Leo have an instruction note to follow. Chad: maybe we can look into dropping some math fonts and not use LaTeX. Jolien: you can probably tell programs to not use LaTeX. 
 (**Leave action items here for next week.**)
 
 ## Last week's call
 
-* [ ] Divya(+Leo) will present the test suite at the upcoming CBC call to make a case for real time R&P injections.
-* [ ] Assigned to Cort/Leo : get the "file list too large" fixing commit on O4-offline-dev merged into master
-  - Leo: Not done yet, but will bring this up this week/next week.
-* [ ] offline meeting with Chad, Anarya, Becca, Yun-Jing for snr maximization in composite event is the right thing to do. Anarya to schedule the meeting ([poll](https://www.when2meet.com/?18322371-Etkpw))
+* [x] Divya(+Leo) will present the test suite at the upcoming CBC call to make a case for real time R&P injections. :arrow_forward: Divya presented at the CBC call. Discussions were that people from the CC pipelines would be needed and we would still need to run injections offline. Divya will present it in the R&P call this week. 
+* [x] Assigned to Cort/Leo : get the "file list too large" fixing commit on O4-offline-dev merged into master
+  - Leo: Brought it up last year to Kipp but he pushed back on analytic VT jobs. Leo suggested writing a separate job. Kipp wanted more time to think on this. Leo wrote up a program and brought it up last week but Kipp needs more time.
+* [ ] offline meeting with Chad, Anarya, Becca, Yun-Jing for snr maximization in composite event is the right thing to do. Anarya to schedule the meeting ([poll](https://www.when2meet.com/?18322371-Etkpw)) :arrow_forward: This hasn't happened. Let's try to make it happen this week. Please fill out the poll. 
 
 ## Agenda / Minutes
 
 * Announcements (5 minutes)
   - Please check the rota for next week's call
-  - Confirmation of next week's focus session
-* Last week's East call - [agenda](https://wiki.ligo.org/CBC/Searches/GstLALEastAgenda20230112)
+  - Confirmation of next week's focus session: Divya as chair and Cort as minutes. No focus sessions
+* Last week's East call - [agenda](https://wiki.ligo.org/CBC/Searches/GstLALEastAgenda20230112): Leo: few participants last week. Brought up Rachel’s commit that was merged. Leo pinged Kipp about the “list too long” error. We are still discussing it. 
 * Quick updates (45 minutes)
   - Operations (5 minutes)
     - LL CBC operations
+      - Leo: Rachel has been working hard on this. For MDC 10 we realized all the inspiral jobs are hanging because of iDQ job error that Rachel fixed yesterday. For Jacob, Becca has been setting up a new branch, but due to limited space on ICDS she hasn’t been able to make the build. We need more space. 
+      - Becca: what’s the status, Ron, anything that we can move to CIT to open up ICDS?
+      - Ron: I wan’t been able to get it done last week because of CIT issues. Ligo-analysis1 has limited space (4.3 TB) left. Anything that can be deleted? 
+      - Becca: we can take a look today. We found some things that can be moved to archive. 
+      - Chad: 4.3 TB seems inconsistent 
+      - Becca: when building containers the init stage failed due to lack of space
+      - Ron: ligo-software is 19GB. 
+      - Becca: I was on analysis1 partition so I’m not sure about software partition
+      - Chad: is there chance it uses tmp space during building and is there change the tmp space is full? A full trace back is needed
+      - Becca: the error came when trying to install new gstlal
+      - Chad: all of that should be <<< 4 TB though
     - Offline CBC operations
     - LL IDQ operations
   - O4 Dev (30 minutes)
     - Low latency integrated testing and Monitoring
     - Template bank
     - Likelihood ratio, background and foreground sampling
+      - Leo: I personally couldn’t attend the LR dev call so there could be sth there. I have updates on chisq stuff Prathamesh has been working on. I will update in the focus session.
     - Optimization and throughput benchmarking
     - Data format wrangling
     - offline DAG
+      - Cort: almost ready to make a commit for the offline dag regrading injections. I’m not sure how to verify that the test I’m doing is giving results we expect. 
+      - Chad: what is the 0th order thing? lalapps_inspinj is used. Production runs in O3 used injections that were generated not with lalapps_inspinj, but ones we got from R&P. Is this to make the offline dag? 
+      - Cort: the offline dag generation is failing because the injections we create ourselves using lalapps_inspinj is failing. 
+      - Becca: we should fix the CI pipeline
+      - Alex:  https://git.ligo.org/lscsoft/gstlal/-/merge_requests/381 the fix to the offline test CI pipeline is in that draft MR
+      - Chad: shio had plotted injection distributions. Maybe you can use that. 
+      - Shio: yes, happy to help
+      - Sarah: this code was reviewed and there are plotting scripts
+      - Shio: sarah, can you please share the plotting scripts in case the one i have is immature?
+      - Sarah: @shio, I will try to dig them up but probably they are in the old repo that is out of synch with current code
     - DQ dev
+      - Stefano: I’m having issues with plotting. I’m not sure but maybe with installation problems? 
+      - Becca: online plotting was seeing the same issue. Leo and Rachel were working on it. Probably due to dependency issues. We need a work around. 
+      - Leo: we still need to figure out what needs to be installed.
+      - Chad: most of the plotting comes from gstlal_plot_summary (?) and it probably imports it from somewhere. We could probably hack that. It’s nice to have TeX but may not be worth having the GB container size. 
+      - Leo: potentially this could help? https://matplotlib.org/stable/tutorials/text/usetex.html#postscript-options
+      - Chad:  I think this is a useful doc page @stefano for how we could maybe stop using latex but still have pretty figures: https://matplotlib.org/stable/tutorials/text/mathtext.html
+      -  Becca: keep it here for following weeks
     - HM search
   - Exploratory development (5 minutes)
   - Misc projects
   - Paper Updates
+    - LR: Leo has started adding theory parts and I’ll push forward. 
 * Focus session
   - [update on the new chisq statistics + signal model](https://docs.google.com/presentation/d/1oe_FsSWABZlUIPMIH-2wKr0MfmHn9dsSW3Ml2KAwugM/edit?usp=sharing) (Leo)
+    - Chad: slide 12 is showing the VT ratio results for just the new sigma model only and slide 13 is comparing with Prathamesh’s results? What are these two slides showing?
+      - Leo: slide 12 is a combination of results of slide 5 and 13. Slide 12 is the whole gain (Prathamesh’s new chisq and the new sigma model).
+      - Chad: does the two give more improvement than individually? 
+      - Leo: yes, the combination of both is the best. 
+      - Chad: looking at closed box plots there is a big outlier, which is not good unless it’s a GW event. Is it a GW event?
+      - Chad: the way in which these are independent it’s reducing DOF. in this parameter space of time and mass or time in temp id, we are choosing a very specific subset. We have a huge freedom in choosing the DOF from this parameter space. with so much machinery, it’s worth considering 
+      - Leo: given the extended covariance matrix, we have more 
+      - Chad: rather than thinking of order 1000 templates you have a matrix. You have shown that the choosing a subset of one in 1000000 reduces DOF then choosing 1 in 1000. Since we’re paying the price, we can get more out of this. 
+      - Jolien: for Gaussian noise,  you don't use chisq at all!
+      - Leo: I’m in favor of reducing the auto correlation length. To compute the time-dependent back correlation matrix takes a huge time (kind of like a bottle neck for the job right now). I have interest in reducing the time for this process.
+      - Chad: my hypothesis is that this 1 M times 1 M templates, and your goal is to choose 1000 of those at random, you will almost end up with a chisq distribution number and statistically would be independent. The overlaps would be 1000. You can benefit from a little more thought.
+      - Leo: I will think about it more
+      - Jolien: on slide 10, Cody was arguing that broader distribution isn’t necessarily a bad thing.
+      - Leo: I agree
+      - Jolien: with modified chisq, your noise gets even more separated. There are two possibilities for why your results are improving. 
+      - Jolien: for VT in slide 5, I don’t see how this is possible that the orange curve has improved VT for all FAR but worse VT for all SNR
+      - Chad: I think that SNR plot is generally wonky … it was from when Steve Privitera was around. 
+      - leo:  yeah,  tbh no one really understand wth is going on this plot????
+      - Sarah: oh, it is probably weird binning issues
+      - Jolien: it seems weird everywhere
+      - Chad: it would happen if the x axis range of left is happening outside the x axis range of the right 
+    - Sarah: Is the plan to have two chisq-ed tests in the LR?
+      - Leo: there are two (auto chisq and bank chisq). Prathamesh and I are leaning towards keeping the conventional auto chisq and replacing the bank chisq with this new chisq and keeping the info on disk.
+      - Sarah: have you looked into more complicated stuff (injections) on how to process HM etc.? 
+      - Leo: it doesn’t have to do with waveforms. The signal models can be constructed for HM waveforms as well. 
+      - Sarah: I would like to see what slide 10 would look like for our more complicated waveforms. 
+      - Leo: I haven’t tested with EW analysis yet. I will ping Surabhi and Ryan abut setting up test. 
+    - Cody: wider signal model should work better if true signals follow the wider model, right? maybe im just not appreciating something, but if VT goes up then it could just be that we're closer to the true distribution than we were before
+      - Leo: Cody, I think you are right
 * AOB
 
 ## Chat log
\ No newline at end of file