Open moralejo opened 5 months ago
Hi @moralejo
in order to make the resources request, I would like to verify the statistics:
They are a subset of the pointings in the training sample - we want to concentrate statistics in discrete pointings to get smaller stat uncertainties in IRFs.
That means the whole statistics of a band splitted between a smaller number of nodes?
Hi @moralejo , could you confirm above, so I can finalise the resource request for CSCS? Thanks
Hi @moralejo
in order to make the resources request, I would like to verify the statistics:
They are a subset of the pointings in the training sample - we want to concentrate statistics in discrete pointings to get smaller stat uncertainties in IRFs.
That means the whole statistics of a band splitted between a smaller number of nodes?
Yes, that was the idea. But see below, it may not be needed immediately.
Hi, I think the discussion here will be useful for a future production in which test MC is no longer produced in a "B_orthogonal" vs. "cos_zenith" grid of pointings (like now) but in nodes along the declination lines. It makes much more sense because it is obvious which RFs must be applied to each test node (the RFs produced along the same dec line and the neighboring ones, which will contain all nodes that can be needed for interpolation).
However, there is no immediate need for this production. We have seen that using only half of the diffuse gamma training statistics for creating the RFs has negligible impact on performance, so we are using the other half as diffuse gamma test MC.
Proposed pointings for a diffuse gamma test Production (so far in PROD2 we have only ring-wobble test gamma). (NOTE: not for production @ LP, but in the Swiss cluster)
They are a subset of the pointings in the training sample - we want to concentrate statistics in discrete pointings to get smaller stat uncertainties in IRFs. All other settings should be the same used in PROD2.