-
Notifications
You must be signed in to change notification settings - Fork 562
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[Bug] Fantasy Models for Multitask GPs are broken #2577
Labels
Comments
With some investigation, it seems like this line
m*d_out where d_out is the output dimension of the GP. In the next line
target and fant_mean are tensors of size m x d_out . So this seems like it would be an easy fix with respect to reshaping, but I'm not sure which is the correct shape to use. If someone can weigh in on what is correct here, I am happy to submit a PR with a new test example that covers this case. @gpleiss?
|
@ancorso we are in the middle of a reworking of the prediction strategies code (timeline tbd) for a 2.0 release. However, we'd accept a bugfix PR for the time being (as long as it's not too much work on your end!) |
cc @hvarfner |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
🐛 Bug
Getting a fantasy model for a simple multi-task GP throws an error
To reproduce
Here is a minimum working example of the bug
** Stack trace/error message **
Expected Behavior
The fantasy model with the appropriately updated cache should be returned
System information
Additional context
This has already been a topic of discussion #800 and #805 and a PR was merged that supposedly implemented this feature #2317. However, the test that was added only works because only a single additional datapoint was added to produce the fantasy model. If you switch
n_new=1
in the example I provide above, it also runs without error but I'm skeptical that the right thing is happening, if it doesn't work for more than 1 additional point.The text was updated successfully, but these errors were encountered: