00:12:55	Matthew Feickert:	so I will probably miss Wolfgang's talk (sorry), so the recording and the transcript from Giordon will be quite useful
00:17:44	Matthew Feickert:	"quarter centruy" oh man.
00:18:49	Sabine Kraml:	:-)
00:21:22	Matthew Feickert:	Do we have a GitHub Issue open asking for HS3 yet?
00:21:33	Matthew Feickert:	(for HEPData)
00:21:34	Giordon Stark (point/he/him, Deaf):	Problem is that it's not easy to implement :-/
00:22:01	Giordon Stark (point/he/him, Deaf):	HEPData doesn't know if it's HS3 or pyhf -- it just uses the "HistFactory" for any text that contains "likelihoods" or mention of it in the submission. They're not inspecting formats.
00:23:44	Sabine Kraml:	That could be changed easily. HEPData would just need to look for the keyword “HS3” or “HS3 likelihood”.
00:24:32	Giordon Stark (point/he/him, Deaf):	Replying to "That could be change..."

sure, and communicate that back to all experiments. I think it would be nicer if HEPData provided a checkbox to select instead when preparing submission.
00:25:17	Sabine Kraml:	Question to Carsten: are all the published HS3 likelihoods incompatible with pyhf? If so, can they (only) be evaluated with Roofit?
00:25:39	Giordon Stark (point/he/him, Deaf):	Replying to "Question to Carsten:..."

Most of the published ones are not HistFactory, or not only HistFactory.
00:26:40	Sabine Kraml:	Replying to "Question to Carsten:..."

Yes, but it would be useful to know (=flag) which ones are HistFactory models.
00:27:06	Giordon Stark (point/he/him, Deaf):	Replying to "Question to Carsten:..."

They're nested in the HS3 format. Some of them have multiple HistFactory models embedded. It's not so easy.
00:29:00	Matthew Feickert:	Thanks for the overview. I've fallen behind on what other groups are doing, so I think seeing the Julia work along with what Giordon has done on pyhs3 is interesting.
00:32:25	Sabine Kraml:	Replying to "Question to Carsten:..."

I was asking whether any of the published HS3 llhds can currently be evaluated with pyhf. And if yes, which. That’s an easy enough question.
00:34:47	Giordon Stark (point/he/him, Deaf):	Which online discussion? I think I missed that?
00:35:20	Giordon Stark (point/he/him, Deaf):	Replying to "Which online discuss..."

Oh, in HS3 GitHub: https://github.com/hep-statistics-serialization-standard/hep-statistics-serialization-standard/discussions/20
00:38:00	Martin Habedank:	Replying to "That could be change..."

Or HEPData tests whether the JSON adheres to the HS3 JSON schema?
00:39:57	Sabine Kraml:	Replying to "That could be change..."

That’s also a possibility. Best let Graeme decide which option he prefers?
00:40:32	Martin Habedank:	Replying to "That could be change..."

👍
00:41:23	Sabine Kraml:	Replying to "That could be change..."

Probably depends on how he identifies Combine LLHDs.
01:03:01	Matthew Feickert:	Great question. The pyhf dev team would make sure that any migrations that are needed would be super well communicated with migration docs and workflows, but we do _NOT_ want to break everyone! So this is something that won't come as a surprise.
01:03:14	Matthew Feickert:	(okay now I gotta run)
01:19:50	Sabine Kraml:	From the (re)user perspective, it would be good to stick with pyhf format for HistFactory LLHDs as long as HS3 can only be run in Root.
01:20:56	Giordon Stark (point/he/him, Deaf):	Anything without HistFactory is probably fine for HS3. But not validated with real-world examples yet.
01:31:13	Carsten Daniel Burgard:	Replying to "Question to Carsten:..."

the published HS3 files are (to my knowledge) all using HistFactory as a backend. as such, they can be converted into the pyhf "legacy" json.
however, out-of-the-box, pyhf will not read these files.
01:34:09	Sabine Kraml:	Replying to "Question to Carsten:..."

ok, that’s what i wanted to know. thanks!
01:36:00	Jack Araz:	Replying to "Question to Carsten:..."

Is there a script, perhaps in pyhs3, that converts things back and forth?
01:37:45	Sabine Kraml:	Replying to "Question to Carsten:..."

Perhaps via Root?
01:41:01	Giordon Stark (point/he/him, Deaf):	Replying to "Question to Carsten:..."

you can always do pyhf->xml->root->hs3
01:41:21	Giordon Stark (point/he/him, Deaf):	Replying to "Question to Carsten:..."

and you can also do hs3->root->xml->pyhf (if the workspace contains only a single HistFactory)
01:54:12	Carsten Daniel Burgard:	Replying to "Question to Carsten:..."

for the hs3->pyhf there's also a direct script which I wrote here:
https://gitlab.cern.ch/cburgard/RooFitUtils/-/blob/master/scripts/json-roofit2pyhf.py?ref_type=heads
01:57:14	Sabine Kraml:	So would Carsten, Giordon and Massimiliano be willing to form the task force?
01:57:55	Carsten Daniel Burgard:	I would for sure be up for it
01:57:59	Giordon Stark (point/he/him, Deaf):	Replying to "So would Carsten, Gi..."

I don't have a permanent job or funding, so I'm not sure.
02:03:38	Sabine Kraml:	i have to leave in a few minutes (sorry bout this, but I have a hard cutoff at 6 pm)
02:05:30	Sezen Sekmen:	Replying to "So would Carsten, Gi..."

maybe this helps to justify all your valuable contributions?
02:08:41	Jack Araz:	Sorry, I have to leave for another meeting. Thanks for the talks!
02:12:59	Nick Smith:	Can conda install combine now
02:14:19	Giordon Stark (point/he/him, Deaf):	Replying to "So would Carsten, Gi..."

not in the US at least. Sabine gave some pointers for non-US, but that comes with its own separate difficulties.
02:14:28	Giordon Stark (point/he/him, Deaf):	Replying to "Can conda install co..."

🥳
02:15:14	Sezen Sekmen:	Replying to "So would Carsten, Gi..."

:(
02:20:08	Giordon Stark (point/he/him, Deaf):	pyhs3 (by construction) provides gradients and autodiff'ability
02:20:22	Christopher Jackson Chang:	Thanks Wolfgang. Sentiment is the same from GAMBIT's side . We have been a bit slower than SModelS to make active use of the pyhf likelihoods (first large scale study using these should hopefully be released soon)
02:29:36	Sezen Sekmen:	I’m afraid I also have to leave.  Sorry, and have a nice day!
02:36:14	Massimiliano Galli:	I have to leave as well, bye!