Interpolate psf

[jsitarek]

a function to interpolate PSF tables. I made it in a similar style like the other two interpolation functions. One difference is that I added an option to interpolate over cumulative distribution. In principle this could help in the binning is fine and thus event statistics are sparse. For the tests that I did I haven't seen any significant difference, but the performance might depend on a particular use case. The default is to interpolate over differential distribution (which is slightly simpler)

@maxnoe I incorporated your comments about solid cone function, please let me know if you have any further comments

[codecov]

Codecov Report

Merging #156 (a5b0d4f) into master (c2cc679) will increase coverage by 0.29%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #156      +/-   ##
==========================================
+ Coverage   89.83%   90.13%   +0.29%     
==========================================
  Files          42       42              
  Lines        1594     1642      +48     
==========================================
+ Hits         1432     1480      +48     
  Misses        162      162              

Impacted Files	Coverage Δ
pyirf/interpolation.py	100.00% <100.00%> (ø)
pyirf/tests/test_interpolation.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update c2cc679...a5b0d4f. Read the comment docs.

[jsitarek]

Hi @maxnoe

can you approve the PR? as I wrote above, your comments are already implemented

[maxnoe]

For weights to make sense, they have to be all >= 0, not just the sum

[jsitarek]

actually there should be no negative weights, but it can happen at low energies that all bins are empty and you have 0s everywhere, this check is meant exactly for such a case

[maxnoe]

Looks good to me, sorry for the delay @jsitarek

[jsitarek]

thx for quick feedback @maxnoe, I merged the PR

[maxnoe]

I wanted to give @HealthyPear a chance to review as well, but I think it's ok

[HealthyPear]

I was about to re-start, but I understand it was open since a while...
Anyway at first sight I have nothing against - just wondering if it made sense to make 2 assertions within the same test, one for high counts and one for sparse ones, so you could check if the weights check was right

[jsitarek]

Hi, sorry @HealthyPear, since there were no comments from you before in this PR I did not think that you wanted to still have a look into this. If you have any further comments, please let me know and I can make a separate PR with them.
About what you wrote above, I'm not sure if I understand it right. When I check the normalization I do an assertion with and 'or', if the bins normalize to 1 or to 0, the second case is when there is a completely empty row (because of lack of surviving MC at some energy)