-
-
Notifications
You must be signed in to change notification settings - Fork 178
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
🪛🔗 Change interactions' shape #736
Conversation
trigger ci
… into fix-performance-regression
part 2 depends on moving the users of this method to new-style, which may require a re-check of performance implications. Opened a new issue in #739 |
trigger ci
trigger c
trigger ci
trigger ci
trigger ci
I verified that the BoxE repro config on WN18RR still succeeds. |
@PyKEEN-bot test |
trigger ci
… into fix-performance-regression
trigger ci
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
either look into changing the score_hrt()/self() dichotomy, or just
This PR is a draft to change the format from the "more canonical" shape
(batch_size, num_heads, num_relations, num_tails, *dims)
to an arbitrary number of batch dimensions(*batch_dims, *dims)
, This should avoid some of the unnecessary dimension expansion and reductions, which may be the cause of the performance issue observed in #735 .In particular it allows
score_hrt
from shape(batch_size, dim)
instead of the expansion to(batch_size, 1, 1, 1, dim)
.