Fix hillas width 0, fixes #772

[maxnoe]

By rounding values very close to zero to exact zero, nan values are avoided.

These happend due to floating point imprecisions resulting in covariances like -8.47032947e-22,
and then sqrt returns nan.

[maxnoe]

Ah, dam. It only build the second commit on travis. I wanted to show that the first one fails and the fix makes it pass.

[maxnoe]

Fixes #772

[codecov]

Codecov Report

Merging #1240 into master will increase coverage by 0.10%.
The diff coverage is 98.61%.

@@            Coverage Diff             @@
##           master    #1240      +/-   ##
==========================================
+ Coverage   86.47%   86.57%   +0.10%     
==========================================
  Files         185      185              
  Lines       11591    11626      +35     
==========================================
+ Hits        10023    10065      +42     
+ Misses       1568     1561       -7     

Impacted Files	Coverage Δ
ctapipe/image/geometry_converter_hex.py	72.32% <ø> (ø)
ctapipe/io/containers.py	100.00% <ø> (ø)
ctapipe/io/simteleventsource.py	97.72% <ø> (-0.06%)	⬇️
ctapipe/visualization/mpl_camera.py	80.00% <85.71%> (ø)
ctapipe/calib/camera/calibrator.py	100.00% <100.00%> (ø)
ctapipe/calib/camera/tests/test_calibrator.py	100.00% <100.00%> (ø)
ctapipe/core/tests/test_tool.py	100.00% <100.00%> (ø)
ctapipe/core/tests/test_traits.py	100.00% <100.00%> (ø)
ctapipe/core/tool.py	85.29% <100.00%> (+0.10%)	⬆️
ctapipe/core/traits.py	94.52% <100.00%> (+0.07%)	⬆️
... and 13 more

---

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 5acb7c3...f824053. Read the comment docs.

[maxnoe]

Sorry @kosack, one more time. We do indeed want to test for exact equality here and I also fixed the random seed.

[vuillaut]

Hi @maxnoe
Is event weighting handled correctly when width=0 ?

[maxnoe]

Hi @maxnoe
Is event weighting handled correctly when width=0 ?

I think the hillas reconstructor raises if any telescope events have width=0 since #994

[vuillaut]

Hi @maxnoe
Is event weighting handled correctly when width=0 ?

I think the hillas reconstructor raises if any telescope events have width=0 since #994

Indeed.
Looks good to me then.