Evaluate question (related to RAD_MAX, but perhaps not only)

[mdpunch]

Hi all,

(still with gammapy 0.20.1)

I have a rad_max IRF, with just one offset, and energy bins 5/decade from 10GeV to 100TeV (so, 20 bins).

>>> rad_max.axes['energy']

MapAxis

        name       : energy    
        unit       : 'TeV'     
        nbins      : 20        
        node type  : edges     
        edges min  : 1.0e-02 TeV
        edges max  : 1.0e+02 TeV
        interp     : log       

The data (i.e., the cut applied in each of the 20 bins) is:

>>> rad_max.data.T[0]*rad_max.unit
[NaN, NaN, NaN, NaN, NaN, NaN, 1.2017647, 0.96087744, 0.75775545, 0.58407787, 0.45431667, 0.35241865, 0.28885854, 0.23838654, 0.20704, 0.17500642, 0.15643081, 0.14916256, 0.15394908, 0.13236962]∘

But, if I try to use evaluate, I get this:

>>> rad_max.evaluate(offset=18*u.deg).T[0]
[0, 0, 0, 0, 0, 0, 0, 0.96087744, 0.75775545, 0.58407787, 0.45431667, 0.35241865, 0.28885854, 0.23838654, 0.20704, 0.17500642, 0.15643081, 0.14916256, 0.15394908, 0.13236962]∘

... mostly fine, except the element [6] changes from 1.2017647 to 0 !!

Checking further:

>>> rad_max.axes['energy'].center[6]
0.19952623TeV
>>> rad_max.evaluate(offset=18*u.deg,energy=0.19952623*u.TeV)
0∘
rad_max.evaluate(offset=18*u.deg,energy=0.2*u.TeV)
1.2005242∘

In short, I did not expect evaluate to give me a zero value for rad_max just below the centre of the bin, which it happens to fall below because of rounding errors.

Of course, this only happens because I have NaN in the lower bins. If the earlier bins have zero (0) the interpolation is fine, but having zero gives errors later.

So, is it possible to change the behaviour of evaluate such that if there is an NaN in the next bin, then the interpolation sets the value to the value of the bin rather than setting to 0?

This would avoid unstable behaviour due to rounding errors...

Good Luck,
Michael.

[adonath]

Thanks @mdpunch! What happens if you use nearest neighbor interpolation instead:

rad_max.evaluate(offset=18*u.deg, method="nearest")

[mdpunch]

Nice!
'nearest' is what I was looking for:

Thanks!

[adonath]

Glad I could help!