SOI table and IRF linking

[kosack]

IRFs have internal and external dimensions. The internal ones define the "cube" that is stored, and the external how many cubes need to be stored separately. The external dimensions are a labeled set, not a continuous distribution.

SOI doesn't link just to a IRF Cube, but to an IRF (which is a set of cubes covering the external dimensions).

proposal for SOI table is as follows:

START	STOP	irf_name	spatial_selection	irf_component	MEMBER_NAME	MEMBER_VERSION	MEMBER_LOCATION
12.3	12.4	superclean_goodpsf	RAD_MAX	A_EFF	AEFF_PNT_SC_GP	1	fits://irfs.fits
12.3	12.4	superclean_badpsf	RAD_MAX	A_EFF	AEFF_PNT_SC_BP	1	fits://irfs.fits
12.3	12.4	superclean_badpsf	NONE	A_EFF	AEFF_PNT_SC_BP	1	fits://irfs.fits

Note that the "primary key" is then (START, STOP, EVENT_TYPE, SPATIAL_SELECTION) → set of IRF components (Aeff, PSF, etc)

Requirement: For external dimensions of the IRF, the name (and allowed values) of the column in the EventList should be identical to the name of the column in the SOI table here.

Open question:

for simple data (e.g. HESS) with no event types, do we require a mapping table (as in #21 )? If not, what do we call the "DEFAULT" (and only) irf_name?

• require then that IRF_NAME = NULL
• or allow irf_name = "*", and then by default you take all in the table.
• just always require mapping table, but hen the name of the IRF can just be whatever, and the matrix is 1

[kosack]

I think this also fundamentally is a question of how we classify each IRF. Currently in GADF there is the following hierarchy specified by the HDUCLASn headers: the question is if we want to do it similarly for VODF, or change the hierarchy. ALready there is some redundancy in the naming.

We seem to have a few things:

• spatial selection (RAD_MAX vs NONE)
• IRF Component: (A_EFF, PSF, E_DISP, BKG)
• type of IRF storage (binned 2D, binned 3D, functional 2d, functional 3d...)

Some notes:

• GADF uses R_PSF, but our PSFs may not be radially symmetic, so why not just PSF? and then the type defines if it is 2D or 1D
• Should we swap HDUCLASS3 and HDUCLASS2? (e.g. the split is more that you have multiple sets of IRFs for each spatial selection option). I.e. is the order: IRF type → IRF (more natural?), or IRF → IRF type (as in GADF)

Here is GADF's version (from the documentation):

HDUCLASS = GADF:

HDUCLAS1	HDUCLAS2	HDUCLAS3	HDUCLAS4
EVENTS
GTI
POINTING
RESPONSE	EFF_AREA	POINT-LIKE	AEFF_2D
		FULL-ENCLOSURE	AEFF_2D
	EDISP	POINT-LIKE	EDISP_2D
		FULL-ENCLOSURE	EDISP_2D
	RPSF	FULL-ENCLOSURE	PSF_TABLE
			PSF_3GAUSS
			PSF_KING
			PSF_GTPSF
	BKG	POINT-LIKE	BKG_2D
			BKG_3D
		FULL-ENCLOSURE	BKG_2D
			BKG_3D
	RAD_MAX	POINT-LIKE	RAD_MAX_2D
INDEX	OBS
	HDU