search

Professor-G / MicroLIA

Gravitational microlensing classification engine using machine learning
GNU General Public License v3.0
13 stars 6 forks source link

Fix GSkew NaN in derivative time series #26

Closed romanovzky closed 11 months ago

romanovzky commented 11 months ago

I've opend a PR here https://github.com/Professor-G/MicroLIA/pull/25 that fixes NaN appearing in GSkew arising from an Index Out of Bounds errors due to np.searchsorted returning an empty array when the q3 or q97 are "outside" the values. This happens when the errors span too many orders of magnitude, provoking a saturation of the cumulative sum.

With this PR and the previous one, I was able to reduce the NaN down to O(0.1%) of the total points, which are edge cases of too weird error distributions that can be safely dropped without impacting any statistical analysis of the dataset.

Professor-G commented 11 months ago

PR #25 has been approved.