Implementing full-enclosure IRFs

[TarekHC]

After discussion between Orel, Victor, Juan and Tarek, we came up with the following recipe:

• The current training (regression) should be good enough for the offset-dependent IRFs

• The key will at the testing stage: we need to define the event type thresholds taking into account the offset. At the testing stage we probably should:

  • Bin test gamma-cone sample (not used in the training) in offset steps (requiring enough statistics to define event types across the energy, as we do now. These bins may be smaller than those offset bins we use for computing IRFs. This is an attempt to not produce FoV effects on different types.

  • Once event types are defined properly taking into account the offset, it will just be a matter of defining broad off-axis bins (as done in ED) to compute the IRFs

  • Modifications to pyirf are stil needed, as we want to compute full-enclosure IRFs (removing the angular cut from the optimization). It should be a simple change, specially if Max is operative.

[TarekHC]

We checked current pyirf code, and it seems effective area and energy migration is already being computed for the following cuts:

masks = {
    "": gammas["selected"],
    "_NO_CUTS": slice(None),
    "_ONLY_GH": gammas["selected_gh"],
    "_ONLY_THETA": gammas["selected_theta"],
}

To test if these full-enclosure effective areas are consistent with the ones computed within ED, we could:

1. Compare ED point-like vs full-enclosure effective area.
2. Compute standard effective area with OnSource gammas with pyirf.
3. Compute _ONLY_GH effective area with a 1-deg gamma-cone file with pyirf.
4. Compare these two, and see if the difference is roughly consistent with ED.

[TarekHC]

Some of the modifications needed at the pyirf stage (either within pyirf, or scripts we need to develop on our side) are being discussed here and here.

[JBernete]

I think we can close this issue. Event-type-wise full-enclosure IRFs have already been tested with Gammapy.