... | @@ -8,18 +8,20 @@ Link to GstLAL road map document: [roadmap](https://docs.google.com/document/d/1 |
... | @@ -8,18 +8,20 @@ Link to GstLAL road map document: [roadmap](https://docs.google.com/document/d/1 |
|
Rachael
|
|
Rachael
|
|
|
|
|
|
## Minute Taker
|
|
## Minute Taker
|
|
|
|
Shio
|
|
|
|
|
|
Char / Minute Taker / Focus Session [Rota](https://git.ligo.org/groups/gstlal/-/wikis/West-call/Rota-table)
|
|
Char / Minute Taker / Focus Session [Rota](https://git.ligo.org/groups/gstlal/-/wikis/West-call/Rota-table)
|
|
|
|
|
|
## Action Items for next week
|
|
## Action Items for next week
|
|
* [ ] have reviewers comment on review status [here](https://git.ligo.org/cbc-allsky-searches/action_items/-/issues/12)
|
|
* [x] have reviewers comment on review status [here](https://git.ligo.org/cbc-allsky-searches/action_items/-/issues/12)
|
|
* [ ] respond to the email from Francesco about the review status.
|
|
* [x] respond to the email from Francesco about the review status.
|
|
|
|
|
|
|
|
* [ ] Follow up on online analysis signal 7 and signal 11 errors. Check that the dev shm patch got applied properly and that c-layer is compiled properly.
|
|
|
|
* [ ] Surabhi: EW: look into high mass recovered inj
|
|
|
|
* [ ] Leo/Prathamesh: shoot Cody an email about extinction model
|
|
(**Leave action items here for next week.**)
|
|
(**Leave action items here for next week.**)
|
|
## Last week's call
|
|
## Last week's call
|
|
* [ ] Rachael: get horizon history MR merged
|
|
|
|
- Kipp took this over, so this can be removed from the list.
|
|
|
|
|
|
|
|
## Agenda / Minutes
|
|
## Agenda / Minutes
|
|
|
|
|
... | @@ -30,28 +32,139 @@ Char / Minute Taker / Focus Session [Rota](https://git.ligo.org/groups/gstlal/-/ |
... | @@ -30,28 +32,139 @@ Char / Minute Taker / Focus Session [Rota](https://git.ligo.org/groups/gstlal/-/ |
|
- Please check the rota for next week's call
|
|
- Please check the rota for next week's call
|
|
- Confirmation of next week's focus session
|
|
- Confirmation of next week's focus session
|
|
- GstLAL F2F at PSU scheduled for June 4th-9th.
|
|
- GstLAL F2F at PSU scheduled for June 4th-9th.
|
|
|
|
- there will be a public page. Rachel will work on it.
|
|
|
|
- we got a hotel block.
|
|
|
|
|
|
* Last week's East call
|
|
* Last week's East call
|
|
|
|
|
|
* Quick updates (45 minutes)
|
|
* Quick updates (45 minutes)
|
|
- Operations (5 minutes)
|
|
- Operations (5 minutes)
|
|
- LL CBC operations
|
|
- LL CBC operations
|
|
|
|
- Leo: we found an event last night. Been seeing grace db uploader errors. Will pull container from scratch since gracedb package will be updated by doing that. We’ve been seeing singal11 errors even tho we’ve added Rachael’s dev shm patch.
|
|
|
|
- Chad: those analyses are at PSU. Could these errors be related to the cluster?
|
|
|
|
- Becca: signal 7s too. I thought we saw them in Renee and Edward too? Maybe I’m wrong
|
|
|
|
- Chad: check in again before Ron is back
|
|
|
|
- Leo: we did see signal 11 errors on Edward and put in Rachael’s patch which fixed it but didn’t fix it for Jacob.
|
|
|
|
- Becca: I thought we had seen the errors since the patch Leo but we should check
|
|
|
|
- Leo: yea
|
|
|
|
- Rachael: we should check on CIT and ICDS again. Are these errors less frequent?
|
|
|
|
- Leo: on Edward it’s been less/I haven’t seen but I’ll check.
|
|
|
|
- Rachael: check on ICDS runs that the patch was applied properly to get the c-layer to compile properly
|
|
|
|
- ACTION ITEM: Rachael: should be added to the sprint this week.
|
|
- LL IDQ operations
|
|
- LL IDQ operations
|
|
|
|
- Rachael: not much has changed for gstlal as much. Maybe Yun-Jing is helping on why idq Kafka messages aren’t getting passed properly.
|
|
- O4 Dev (30 minutes)
|
|
- O4 Dev (30 minutes)
|
|
- Low latency integrated testing and Monitoring
|
|
- Low latency integrated testing and Monitoring
|
|
- Template bank
|
|
- IMBH Template bank
|
|
|
|
- Debnandini: working on IMBH runs
|
|
- Likelihood ratio, background and foreground sampling
|
|
- Likelihood ratio, background and foreground sampling
|
|
|
|
- Leo: presented paper at CBC call. Hope to be reviewed and send to Arxiv.
|
|
|
|
- Prathamesh: will give update on sampling issue. Chad, Leo and I looked in. Individual foreground also has the bump like in the background but when foreground is marginalized over, the bump disappears. Chad came up with a new continuous extinction scheme. After individual background of each bin is marginalized, it seems to work well. We’re still investigating how to move forward but this is promising.
|
|
|
|
- Rachael: would like to have more in depth update on this with plots
|
|
|
|
- Prathamesh: maybe I can do a focus session.
|
|
|
|
- Rachael: maybe in a week or two after the focus sessions are done
|
|
|
|
- Chad: great is a subset of us get things done in the next two weeks. Not changing for online but should be for online rerank. And have the presentation for review.
|
|
|
|
- Prathamesh: sounds good.
|
|
|
|
- Chad: few of us + Cody might work
|
|
|
|
- Cody: loop me into getting extinction model work. Shoot me an email
|
|
- Injection file format
|
|
- Injection file format
|
|
|
|
- Victoria: working on getting it into gstlal. Jolien, can you verify stuff I sent you in the email?
|
|
|
|
- Jolien: let me look at that rn.
|
|
- offline DAG
|
|
- offline DAG
|
|
|
|
- Divya: I’m trying to straighten out the inj rerank rn. That’s were I’m stuck at. Nothing much since last week.
|
|
|
|
- Rachael: you me and a younger student, can we get together this week? Let’s follow up offline
|
|
- DQ dev
|
|
- DQ dev
|
|
- HM search
|
|
- HM search
|
|
- Exploratory development (5 minutes)
|
|
- Exploratory development (5 minutes)
|
|
|
|
- Pratyusava: mdc analytics: MBTA has more triggers than GstLAL because of how MBTA groups triggers. They group triggers, drop duplicates, and takes the last trigger in H1 gps time and have far cut off.
|
|
|
|
- Rachael: MBTA takes all triggers based on H1 time and drops duplicates, put FAR cutoff
|
|
|
|
- Pratyusava: the last trigger to appear in time may not be the significant one. All pipelines would have more triggers if one uses how MBTA does it
|
|
|
|
- Chad: do you have pi chart? Do you have actual numbers?
|
|
|
|
- Pratyusava: the numbers are on my screen. This first one is for gstlal the second one is for mbta. Maybe I’ll create an issue in the mdc analytics.
|
|
|
|
- Shio: can you add that to the issue we were working on?
|
|
|
|
- Pratyusava: yes
|
|
|
|
- Chad: was that looking at mdc11?
|
|
|
|
- Pratyusava: yes
|
|
|
|
- Rachael: is the issue in mdc analytics the best place?
|
|
|
|
- Pratyusava: yes and I can ping ppl. and tmr’s EM follow call.
|
|
|
|
- Chad: if we have a test suite to show the numbers
|
|
|
|
- pastro: Shio: Anarya found that the inj NS mass cut was 2 instead of 3, so our rates and categorizations would have messed up
|
|
- Misc projects
|
|
- Misc projects
|
|
- Paper Updates
|
|
- Paper Updates
|
|
|
|
- Becca: have presentation on performance paper. Will try to get it on arxiv by Friday. Make comments ASAP.
|
|
|
|
- Rachael: did the pnp reviewer already approve it?
|
|
|
|
- Becca: yes. I’ll push changes today.
|
|
|
|
- Leo: LR paper will be submitted to arXiv soon
|
|
|
|
|
|
* Focus session
|
|
* Focus session
|
|
- Review
|
|
- Review
|
|
|
|
- Becca: pastro was gonna happen today but they postponed the mtg
|
|
|
|
- Rachael: we should follow up with them offline since that’s the last thing on the operational review
|
|
|
|
- Leo: there seems to be another review sign off page. Reached out to Anarya but haven’t heard back.
|
|
|
|
- Divya: isn’t pastro review just the mass cut?
|
|
|
|
- Rachael: I think we checked last week. But Graham (?) had a request to have pastro review done.
|
|
|
|
- Chad: idk if anyone looked at pastro results for non inj resutlts? Can someone do that?
|
|
|
|
- Becca: in the performance paper the pastro plot include pastro results.
|
|
|
|
- Chad: mdc12 pastro model is different so I would like it checked.
|
|
|
|
- Rachael: we should have a sprint task
|
|
|
|
- Shio: does that mean go into gracedb uploads and check the pastors?
|
|
|
|
- Chad: for uploads from non-inj channel. Check that non-GW events are not given low pTerr, it adds up to 1, etc.
|
|
|
|
- Shio: I’ll make an issue and send it to you for verification. [here](https://git.ligo.org/gstlal/pastro/-/issues/4)
|
|
|
|
|
|
|
|
- Becca: aggregation stuff will be done offline
|
|
|
|
|
|
|
|
- Surabhi: can we get some time for EW?
|
|
|
|
- Surabhi: using 29 Hz we tested 10 ms coinc threshold and 20 ms coinc threshold. The search sensitivity didn’t improve significantly for 20 ms. Money plots look worse. We want to keep 10 ms.
|
|
|
|
- Jolien: do you know why the money plots changed? Fewer things in the background due to coinc threshold change?
|
|
|
|
- Surabhi: that could be what happened.
|
|
|
|
- Jolien: I wouldn’t worry too much.
|
|
|
|
- Surabhi: ACTION ITEM to look into it further
|
|
|
|
- Chad: as seem from Prathamesh’s stuff, we will revisit this.
|
|
|
|
- Surabhi: sounds good. I can show money plots from online runs. I think Ryan already showed binning that was an action item from last time.
|
|
|
|
|
|
|
|
- Surabhi: last thing on review was mass model. We’re using the same code as all sky. We’re using saltpeter mass model.
|
|
|
|
- Surabhi: mdc12: we ran it for 30 days. Up time is 91 %. at 2/year FAR, we detect almost all bns inj. The ones missed are single ifo. We’re only doing coinc searches. Gracedb latency is negative latency, with some in positive.
|
|
|
|
- Jolien: I’m surprised you’re detecting some high mass things.
|
|
|
|
- Surabhi: because our bank goes up to 4 Msun or so.
|
|
|
|
- Jolien: are these detected in your bands? Like in your lowest freq band? These things are like high freq things. That’s a little weird. Shouldn’t they be filtered out by low frequency band or fail chisq test for high frequency band?
|
|
|
|
- Surabhi: ACTION ITEM: follow up on the high mass found inj.
|
|
|
|
- Jolien: are the orange ones found?
|
|
|
|
- Surabhi: no, missed
|
|
|
|
- Jolien: oh then I’m not worried
|
|
|
|
- Surabhi: anarya is working on making more plots
|
|
|
|
- Surabhi: major issue in mdc12. FAR estimation wasn’t working properly. We increased messaged that are passed through. Looks like high far threshold part is good but not low far threshold. Attaching the extinction model at snr 15 may be able to fix it.
|
|
|
|
- Jolien: by when do you hope to have results?
|
|
|
|
- Surabhi: next week?
|
|
|
|
- jolien: anything changed from mdc11 to mdc12?
|
|
|
|
- Surabhi: we used sbank in mdc11 and now manifold.
|
|
|
|
- Chad: also mass model. Main motivation to change the bank was to have mass model support.
|
|
|
|
- Surabhi: things happened in mdc11 where our analysis was uploading a lot of singles.
|
|
|
|
- Surabhi: we were using a mass model not consistent with the all sky analysis and not saltpeter. The bank sim of the manifold bank looks much better.
|
|
|
|
- Jolien: the retraction is sth only seen in EW but not seen in all sky?
|
|
|
|
- Surabhi: yes. Seems like many of these are from low bandwidths. Lower bandwidth triggers should be downwighed compared to higher bandwidth.
|
|
|
|
- Surabhi: I’ll start some offline runs to get results sooner.
|
|
|
|
- Surabhi: are two retractions we are seeing problematic?
|
|
|
|
- Jolien: does EW have requirements like the all sky?
|
|
|
|
- Surabhi: in terms of general requirements there isn’t. There are mdc success criteria which I’m not sure we’re supposed to do for review
|
|
|
|
- Jolien: we need to check what it looks like in a week.
|
|
|
|
|
|
|
|
- Rachael: should EW be monitored by the GstLAL team?
|
|
|
|
- Surabhi: that would be great if that’s ok
|
|
|
|
|
|
|
|
- Rachael: what other review items need to be done?
|
|
|
|
- Prathamesh: count tracker stuff -> offline. In the issue [here](https://git.ligo.org/gstlal/o4-code-review/-/issues/6), it shows the results of items that were requested. It would be great if reviewers can go through this.
|
|
|
|
- Prathamesh: there is combined chisq stuff. We were storing bank chisq but wasn’t using it. Combined chisq is a way to use auto chisq and bank chisq. Add the two and normalize. This comes from auto chisq not actually being a chisq distribution.
|
|
|
|
- Jolien: when are you planning for this to be used in the search?
|
|
|
|
- Prathamesh: maybe reranks. Not in online since we don’t have enough evidence. We’ll try online reranks with and without combined chisq.
|
|
|
|
- Leo: I think this has to come with the new signal model adapted to the new chisq statistics. I have a new patch but haven’t been merged to master. It’s not urgent so it won’t be online.
|
|
|
|
- Prathamesh: our preliminary results tells us that it doesn’t give crazy results. It seems to improve VT for IMBH the most.
|
|
|
|
- Prathamesh: can I mark this as done?
|
|
|
|
- Jolien: yes. Make sure it doesn’t get put into production unexpectedly. Is there a safe guard for that?
|
|
|
|
- Prathamesh: same as before, bank chisq is stored. We don’t have a explicit safe guard tho.
|
|
|
|
- Jolien: but you’re changing itacac etc.
|
|
|
|
- Prathamesh: I did a test before
|
|
|
|
- jolien: chisq isn’t replaced with combined chisq
|
|
|
|
- Prathamesh: yes. Chisq is untouched.
|
|
|
|
|
|
|
|
|
|
|
|
|
... | | ... | |