GPs with Kronecker structure #3065
Conversation
|
This close #2860 right? |
|
Yup, that's the goal. |
| m = m.T | ||
| res, _ = theano.scan(kron_vector_op, sequences=[m]) | ||
| return res.T | ||
| res = kron_vector_op(m) |
There was a problem hiding this comment.
Yeah it turned out not too bad! Somehow scan returned the array in like, fortran ordering instead of c ordering. Somehow...
There was a problem hiding this comment.
I'm not sure why I did not do this in the first place. Though I wanted to avoid scan if at all possible, I feel like there was certain cases where "nicer" ways failed, such as with non-square matrices or something, but I can't remember. In any case, if the tests pass and everything still works I like this a lot!
There was a problem hiding this comment.
awesome, glad to hear it! we'll see, if they're out there maybe those issues will come up down the road, and maybe not.
|
I think this is ready for merge. I'll do the coregionalization example in its own PR. |
junpenglao
changed the title
|
Looks great! Thanks for cleaning up my mess 👍 |
|
I think this is ready to go, will merge in a couple days if everyones ok with it |
Building on @jordan-melendez's work:
gp.LatentKron, testsgp.MarginalKron,gp.LatentKronclose #2860