_UnivariateGaussian should only be constructed for scalars

probabilistic-numerics / probnum

Probabilistic Numerics in Python.

http://probnum.org

MIT License

438 stars 57 forks source link

_UnivariateGaussian should only be constructed for scalars #94

Closed marvinpfoertner closed 4 years ago

marvinpfoertner commented 4 years ago

IMHO a mean (and cov) with shape (1, 1) should construct a _MatrixVariateNormal and a mean with shape (1,) should construct a multivariate normal to be consistent with numpy. This would also make slicing much more intuitive.

https://github.com/probabilistic-numerics/probnum/blob/f13f26cf441e4182339faa2bc2b8347da77960c8/src/probnum/prob/distributions/normal.py#L271-L280

nathanaelbosch commented 4 years ago

Fixing this will break some tests in diffeq. This is a good thing! Ping me or @pnkraemer for help.

marvinpfoertner commented 4 years ago

Resolved by #135