09:39:01 From Alexander Held to Everyone:
	There’s a somewhat subtle point about 1d vs 2d due to the way interpolation/extrapolation works for shape variations, the way you slice the 2d space into 1d spaces will generally impact this. Ideally it shouldn’t matter though, once you depend on the inter/extrapolation, that may be a bigger problem altogether.
09:40:30 From Iason Krommydas to Everyone:
	We usually dislike Bayesian in HEP for some reason whilst astronomers love it
09:45:43 From Iason Krommydas to Everyone:
	I will ask a question here in order to not interrupt.
09:48:37 From Iason Krommydas to Everyone:
	The mean value of those Poissons is a function of the bin. However in hep it is very rare that there is an analytic function that says that lambda of each bin is given as a real function f(bin). Are these lambdas usually taken from templates of Monte Carlo where the Monte Carlo will say for instance that in the first bin you expect 100 events, in the second bin 50 and so on?
09:49:50 From Alexander Held to Everyone:
	Yes, exactly. You have a set of templates that give you the default prediction and then you obtain the effect of all parameters on the prediction per bin via the interpolation between up and down templates that Lukas is mentioning on this slide.
09:50:48 From Iason Krommydas to Everyone:
	And then you add a parameter that you want to to estimate that will multiply this template by a real number and that is the “signal strength” correct?
09:51:05 From Iason Krommydas to Everyone:
	Which should be 1 if SM is true
09:51:43 From Alexander Held to Everyone:
	Yes, that is one of these parameters of type “norm factor”. They don’t need any additional templates to define them as they just multiplicatively scale the templates they act on.
09:52:06 From Alexander Held to Everyone:
	`normfactor` (one word, but autocorrected)
10:02:18 From Iason Krommydas to Everyone:
	Are the modifiers for systematics a collaboration wide decision?
10:03:34 From Alexander Held to Everyone:
	You mean the choice of type or which you should have to consider? I can only answer for ATLAS, but there’s a mix of analysis-specific choices and more central prescriptions that analyses typically follow.
10:05:24 From Iason Krommydas to Everyone:
	The types. Like for instance who says what type of modifier to use for a particular systematic
10:06:19 From Alexander Held to Everyone:
	There is usually a clear choice depending on what you want to encode. They are all quite different from each other.
10:07:53 From Iason Krommydas to Everyone:
	For instance, by being naive, if there is a systematic that says that I have a 10% uncertainty on how many events I count, I would add a gaussian likelihood term where the mu is my counts or the sigma is 0.1*counts
10:08:15 From Iason Krommydas to Everyone:
	And the sigma*
10:08:26 From Alexander Held to Everyone:
	right and that’s a normsys modifier in pyhf
10:12:22 From Géraldine Räuber to Everyone:
	Is it possible to upload these slides on Indico?
10:13:39 From Lukas Alexander Heinrich to Everyone:
	Yes, I’ll upload them
10:13:51 From Géraldine Räuber to Everyone:
	Thanks a lot
10:38:36 From Matthew Feickert to Everyone:
	Alex designed cabinetry to work with pyhf, not just wrap it
11:01:38 From Ramdas Makhmanazarov to Everyone:
	which constrained types of params are used here?
11:06:23 From Ramdas Makhmanazarov to Everyone:
	can we change constraint term of parameter?
11:07:05 From Ramdas Makhmanazarov to Everyone:
	okay
11:07:16 From Ramdas Makhmanazarov to Everyone:
	will we talk about it today?
11:14:15 From Matthew Feickert to Everyone:
	No, There's an issue open on this
11:17:31 From Chris Ketter to Everyone:
	I had a similar problem with running many toys. I just used a "try: / except:" clause in python to the job keeps running.
11:17:35 From Matthew Feickert to Everyone:
	If you want to follow up here with explicit examples that would be good https://github.com/scikit-hep/pyhf/issues/1427
11:29:42 From Alexander Held to Everyone:
	one thing useful in that context is e.g. https://cabinetry.readthedocs.io/en/latest/api.html#cabinetry.model_utils.match_fit_results
11:32:04 From Alexander Held to Everyone:
	see https://github.com/scikit-hep/pyhf/issues/850 + https://github.com/scikit-hep/pyhf/discussions/1627
11:34:31 From Slavomira Stefkova to Everyone:
	Reacted to "see https://github.c..." with 👍
11:34:48 From Lukas Alexander Heinrich to Everyone:
	https://github.com/pyhf/pyhf-gpsys
11:37:04 From Alexander Held to Everyone:
	multi-POI: https://github.com/scikit-hep/pyhf/issues/179 (also on my slide 26)
11:39:44 From Alexander Held to Everyone:
	2d scans: https://github.com/scikit-hep/cabinetry/issues/339
11:40:17 From Alexander Held to Everyone:
	(and I agree with what Lukas said, happy for contributions!)
11:42:49 From Alexander Held to Everyone:
	staterror pruning: https://github.com/scikit-hep/pyhf/issues/662
11:54:46 From Moritz Bauer to Everyone:
	I think the feature I  tried to refer to is thiis one: https://github.com/scikit-hep/pyhf/issues/850
11:57:53 From Matthew Feickert to Everyone:
	Thanks!
11:58:00 From Michel Hernandez Villanueva to Everyone:
	Thank you!
11:58:07 From Ramdas Makhmanazarov to Everyone:
	thank you!