-
Notifications
You must be signed in to change notification settings - Fork 149
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
Lightgcn fix #602
Lightgcn fix #602
Conversation
Removed normalization for layers, not used for lgcn. Fixed sum weight constant to num layers instead of cur layer index. Allow lgcn to take blocks. Fixed requirement error caused by newer dgl versions. Moved edge normalization to graph for easier use.
@tqtg as per #600 (comment), I've created a pull request for LightGCN only. |
thanks @theisjendal for clarifying all the changes. They all make sense to me after revising the authors' implementation of LightGCN. Could you also try to compare the updated version and the old one using this example with the same hyper-parameters? I'm curious about the changes in accuracy and speed of convergence. |
@tqtg, here are the results:
NEW
VALIDATION:
TEST:
|
LGTM. Feel free to merge this PR. |
Description
LightGCN is a simplified version of NGCF, however, some of the simplifications made in LightGCN where not included.
Therefore:
1/(L+1)
and instead of1/(l+1)
whereL
is the number of layers andl
is the current layer index.Checklist:
README.md
(if you are adding a new model).examples/README.md
(if you are adding a new example).datasets/README.md
(if you are adding a new dataset).