How to use rad_max from an IRF, v0.20.1 ?

[mdpunch]

Hello,
I'm testing out gammapy v 0.20.1, in particular trying the rad_max IRF for a variable rad_max (theta cut) as a function of reconstructed energy.
I have my IRFs, including one with RadMax2D.

print(irfs)

{'aeff': <gammapy.irf.effective_area.EffectiveAreaTable2D object at 0x7f3e56ff2e50>, 
'edisp': <gammapy.irf.edisp.core.EnergyDispersion2D object at 0x7f3e57003b80>, 
'psf': <gammapy.irf.psf.table.PSF3D object at 0x7f3e56f7d340>, 
'rad_max': <gammapy.irf.rad_max.RadMax2D object at 0x7f3e56fcb7c0>, 
'bkg': <gammapy.irf.background.Background2D object at 0x7f3e56f6ff70>}

and irfs['rad_max'].plot_rad_max_vs_energy() gives the expected result.
But, when I try to create an observation obs = Observation.create(pointing=pointing, irfs=irfs, livetime=livetime) I can access all the IRFs (e.g., obs.psf or obs.aeff) except for the rad_max:

obs.rad_max

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
/tmp/ipykernel_780266/1763588394.py in <cell line: 2>()
      1 #Tk
----> 2 obs.rad_max

/scr/punch/miniconda3/envs/alto/lib/python3.9/site-packages/gammapy/data/observations.py in rad_max(self)
    103         edisp = self.edisp
    104         if edisp is not None and edisp.is_pointlike:
--> 105             self._rad_max = RadMax2D.from_irf(self.edisp)
    106 
    107         return self._rad_max

/scr/punch/miniconda3/envs/alto/lib/python3.9/site-packages/gammapy/irf/rad_max.py in from_irf(cls, irf)
     63 
     64         if "RAD_MAX" not in irf.meta:
---> 65             raise ValueError("Irf does not contain RAD_MAX keyword")
     66 
     67         rad_max_value = irf.meta["RAD_MAX"]

ValueError: Irf does not contain RAD_MAX keyword

..,. and the rad_max IRF isn't in the list of the ones available:

obs.available_irfs

['aeff', 'edisp', 'psf', 'bkg']

Also, the background estimation later on doesn't use this information, and I have to calculate the counts_off myself for the dataset.
What am I missing?
Thanks in advance for help,
Michael.

[bkhelifi]

Hi @mdpunch,
Your code is using the function observation.create to add your custom IRFs into your 'simulated' observation. Unfortunately, the rad_max irf is added inside the create function. This is a feature to add...

[mdpunch]

Thanks for the answer, and the resulting issue: #4060

As we discussed and agreed, having the RAD_MAX as in IRF may be problematic, since it is really metadata going into the creation of aeff and edisp. It seems to be possible to add there too. So if it is in either/or/both as metadata / IRF, it needs to be handled both ways, I guess.

But that's maybe a philosophical point, whereas it might be best to go with "whatever works", which the issue will solve.