- LR Dev - ([slides](https://docs.google.com/presentation/d/1qvzh3KIqbU_gmO4uenAu4YabLv8CQ402p8KD4m6B8rM/edit?usp=sharing)) Leo, Chad
- LR Dev - ([slides](https://docs.google.com/presentation/d/1qvzh3KIqbU_gmO4uenAu4YabLv8CQ402p8KD4m6B8rM/edit?usp=sharing)) Leo, Chad
- Slide 2:
- Slide 2:
- LR is calculated by taking the ratio of data under signal and noise hypothesis
- LR is calculated by taking the ratio of data under signal and noise hypothesis
- Slide 4:
- Slide 4:
- Inspiral_lr is HARD!!
- Inspiral_lr is HARD!!
- Slide 6:
- Slide 6:
- Propose new RankingStat class (ignore P(instruments) for simplicity)
- Propose new RankingStat class (ignore P(instruments) for simplicity)
- P(chisq) and dtdphi term merget because both use histograms, so it’s easier to do it together
- P(chisq) and dtdphi term merget because both use histograms, so it’s easier to do it together
- Slide 8:
- Slide 8:
- Terms attribute added for better transparency
- Terms attribute added for better transparency
- Slide 10:
- Slide 10:
- When the class is called, all the LR contributions are shown, instead of just the final LR, for better transparency
- When the class is called, all the LR contributions are shown, instead of just the final LR, for better transparency
- Slide 12:
- Slide 12:
- In one example, if we simulate signal params and evaluate LR, we get -inf, and now we can see which term causes that. This info is not necessarily saved to disk, but we can have tools to access this info, making it easier to debug etc.
- In one example, if we simulate signal params and evaluate LR, we get -inf, and now we can see which term causes that. This info is not necessarily saved to disk, but we can have tools to access this info, making it easier to debug etc.
- Slide 13:
- Slide 13:
- In-builts tools to make plots, etc
- In-builts tools to make plots, etc
- Slide 14:
- Slide 14:
- Can create fake classes, and generate signal model using train_signal
- Can create fake classes, and generate signal model using train_signal
- Slide 21:
- Slide 21:
- We are changing coordinates from detector SNR to network snr, snr ratio, we maybe need the Jacobian term here
- We are changing coordinates from detector SNR to network snr, snr ratio, we maybe need the Jacobian term here
* AOB
* AOB
## Chat log
## Chat log
\ No newline at end of file
<09:33:55> "chad.hanna": hand up
<09:34:04> "SurabhiSachdev": we could have an ew focus session but not in one week (we should have some results)
<09:35:31> "chad.hanna": hand up for one more thing
<09:42:49> "Becca Ewing": it looks like its different at the high snr end, is that expected or explained?
<09:43:32> "stefano.schmidt": is there an online summary page for this run? I'm interested to compare this with my own offline runs, which gave weird recovery results
<09:43:38> "ryan.magee": LR wise, i don't think anything beyond prior tweaks is expected.
this is the closed box for the result that finished last week
<09:44:43> "stefano.schmidt": thanks @shio!
<09:45:28> "Shio.Sakon": @Jolien i had made mistakes in the num-banks etc. for the BBH, OTHER_BBH, and IMBH sub banks
<09:45:34> "Shio.Sakon": I'm sorry.
<09:46:00> "Shio.Sakon": I have an offline analysis with the corrected num-banks etc. that was running, but i have been stuck since Sunday
<09:46:10> "Shio.Sakon": we didn't have it
<09:46:23> "chad.hanna": will there be an east call tomorrow?
<09:46:57> "Leo Tsukada": Japan is in a week of national holidays I guess
<09:46:58> "Shio.Sakon": @Chad I am not sure. I haven't seen an email
<09:47:33> "Leo Tsukada": they have something like "Golden week" (GW)
<09:47:39> "Shio.Sakon": nothing other than what i mentioned
<09:48:46> "SurabhiSachdev": Btw I noticed for online configs we don't have different dtdphi files set for different sub banks
<09:48:52> "patrick.godwin": yeah, that's right
<09:50:30> "chad.hanna": hand up
<09:52:06> "Becca Ewing":
<09:55:32> "Leo Tsukada": I think most of the relevant codes are already in event_uploader or something?
<09:55:37> "Shio.Sakon": no
<09:55:38> "chad.hanna": hand up
<09:56:23> "jolien.creighton": hand up
<10:00:49> "jolien.creighton": you could also try simplexes in infinite time
<10:03:10> "Andre Guimaraes": My General exam is today and I've been focusing on that, so I haven't done much more progress since last Friday, but I'll be presenting something by next Monday for sure.
<10:22:06> "SurabhiSachdev": sorry I need to leave, I'll catch up with the minutes later.
<10:23:12> "Rachael Huxford": hand up
<10:23:18> "anarya": I asked bcause I have been facing this issue, for real triggers. if a trigger is found only in two detectors (say L1, V1) , I get -inf
<10:23:51> "chad.hanna": hand up at end for the -inf stuff
<10:24:25> "chad.hanna": hand up to rachaels question
<10:25:07> "Becca Ewing": that would be cool
<10:25:26> "Rachael Huxford": so it would recreate this info given like gradcedb info eg
<10:26:04> "Rachael Huxford": I gotcha. that would be awesome
<10:26:12> "anarya": yes thanks.. I just brought it up because I traced my issue to the time_phase_snr part of inspiral_extrinsics which is called by inpiral_lr. it would be helpful if we could follow this up offline.. I have to teach in 5 minutes
<10:26:31> "chad.hanna": anarya, is the dt just too big in your case?
<10:26:36> "chad.hanna": light travel time is only 30ms
<10:27:22> "anarya": no I tested that.. I sent you an email with an small example a few days agao.
<10:27:42> "chad.hanna": ah, okay
<10:28:34> "chad.hanna": hand up
<10:29:37> "anarya": @chad the problem is very simple to reproduce but its also very weird (to me) answers change depending on whther I initialize one object before the other (they are independent). if you and leo have time lets chat offline sometime? thanks..
<10:30:25> "anarya": by everyone , gotta signoff
<10:31:06> "chad.hanna": hand up on this
<10:33:35> "ryan.magee": i think it does uniformly in bin and then uniformly across the bin domain right?
<10:33:53> "ryan.magee": right
<10:34:13> "ryan.magee": just marginally closer -- i *think* i looked at this with bankchisq and it wasn't as huge of an impact as you might expect.
