gammapy.irf is a horrible mess

Problem

gammapy.irf contains many classes to represent IRF tables with different axis. At the moment it is very difficult to use (even for me, who wrote many of these classes). I think the problem is that the class naming is not coherent at all. Sometimes classes are called SomeIRF2D and sometime EnergyDependentSomeIRF and sometimes it’s not clear at all what a certain class represent. In some cases, classes representing a certain type of IRFs are missing completely (the functionality might be hidden in other classes, but should be exposed for easier, clearer code structure, I think). If we plan to use gammapy as a tool for CTA simulation, prototyping etc. good IRF classes are a must.

Status quo

Axis names are defined as follows (should be in agreement with https://gamma-astro-data-formats.readthedocs.io/en/latest/)

• theta : Offset from pointing position, field of view offset
• energy : True energy
• ereco : Reconstruced energy
• offset : Offset from observation position (for PSF)
• migra : Energy migration ( e_true / e_reco)

At the moment the situation is the following

Energy Dispersion

Axes	Class name	Current axis names	Proposed Class Name
theta energy migra	EnergyDispersion2D	etrue offset migra	EnergyMigration3D
migra	Not implemented		EnergyMigration1D
energy ereco	EnergyDispsersion	e_true e_reco	EnergyDispersion2D
ereco	Not implemented		EnergyDispersion1D

Effective Area

Axes	Class name	Current axis names	Proposed Class Name
theta energy	EffectiveAreaTable2D	energy offset	EffectiveArea2D
energy	EffectiveAreaTable	energy	EffectiveArea1D

PSF

I list parameters as separate axis for parametrized version of the PSF

Axes	Class name	Current axis names	Proposed Class Name
energy theta sigmas, norms	EnergyDependentMultiGaussPSF	energy theta	MultiGaussPSF2D
energy offset	EnergyDependentTablePSF	energy offset	TablePSF2D
energy offset theta	PDF3D	energy offset rad	TablePSF3D
energy theta gamma, sigma	PSFKing	energy offset	KingPSF2D
offset	TablePSF	offset	TablePSF1D

Futhermore there are GaussPSF, HESSMultiGaussPSF, PositionDependentMultiGaussPSF, SherpaMultiGaussPSF . I am not always sure what their purpose is, should be made clear in the docstrings. In same cases gammapy might not be the right place for them.

Plan

I propose the following (to be discussed)

• Find good naming scheme (see proposals in the table above)
• Add overview table to the docs explaining which IRFS class has which axis and so on.

cc @cdeil @adonath @JouvinLea @registerrier @alziegler @dltiziani

[adonath]

After #3185, cleaning this up a bit more is in scope for v1.0 again. I’ll make a proposal with a naming scheme here soon.

[adonath]

Form the activity here in the issue it seems we will stick with the current names for v1.0. This is acceptable, as the current system works and code is shared under the hood with ND base classes. The recommendation here would be to relabel to v1.1.