Apparent dimension mismatch in gaussian evaluation

[curieux]

I have an example where a multilabel PointCloud has a sample with shape (7,3) but the rotation matrix ends up as (2,3) and the mean ends up as (3,3). Tracking down where the mismatch happens.

[curieux]

The above report was slightly incorrect.
The problem occurs when the number of points in the sample is less then the dimension.
It is fixed by changing line 570 of multidim/models.py from
if count <= 1: to if count < ambient_dim:

patch coming soon, with some others.