Join GitHub today
[MRG+2] Relax assumption on the data for the SkewedChi2Sampler. #7573
What does this implement/fix? Explain your changes.
Since the Skewed-Chi^2 kernel is defined on the open interval (-self.skewedness; +\infty)^d, the transform function should not check whether X < 0 but whether X < -self.skewedness.
The line after the check does X += self.skewedness.