Multivariate normal CDF

Questions v5
1

Does anyone have any advice on whether it is possible to evaluate the CDF of a multivariate normal from PyMC variables?

This old StackOverflow question suggests that creating an Aesara op over scipy.stats.multivariate_normal.cdf might be a potential approach.

There are some posts on the Stan forums that suggest that it might not be possible (e.g., here and here), but there seems to be some recent talk about “copulas” that I don’t quite follow but might be relevant.

For some context, I am looking into whether a multivariate ordered probit approach might be possible to model data where multiple people provide ratings for multiple items.

Thanks for any advice!

2

Yes you should be able to create the CDF of a mvnormal by wrapping the scipy Op

1 Like
3

Great, thanks Ricardo - I’ll give that a go.