Throwing PosDefException in the corner cases

[kirill-kondrashov]

In the case when some of the components don't have enough data points throws
PosDefException: matrix is not positive definite; Cholesky factorization failed.
rather than avoiding to update those components (which seems more practical) and showing a warning.

[davidavdav]

Hello,

so you suggest that the exception should not be thrown, and instead no update of the components should take place?

[kirill-kondrashov]

Hello,

yes, I would rather give a warning or something similar. Right now it just throws and you have an inconsistently changed argument structure if I understand correctly. A similar situation takes place with deprecated ind2sub function call when the amount of data per cluster is not enough as well, but that's a separate bug, I think.

[harryscholes]

I just ran into ind2sub not defined. Possibly because all of my data was in one cluster.