search

malllabiisc / CompGCN

ICLR 2020: Composition-Based Multi-Relational Graph Convolutional Networks
Apache License 2.0
597 stars 107 forks source link

Regarding number of GCN updates #26

Closed Megh-Thakkar closed 3 years ago

Megh-Thakkar commented 3 years ago

Hi, thank you for the easy-to-understand code repository. As per my understanding, there is only one CompGCN update at every step, which would be similar to aggregating information from the first-hop neighbors. Did you experiment with more hops/iterations?

Thank you.

soumyasanyal commented 3 years ago

Yes, we can always add multiple layers of CompGCN to increase the neighborhood influence as required.

Megh-Thakkar commented 3 years ago

Thanks for the very quick reply. Just wanted to confirm if it had any effects that you might have experimented with. Closing this for now.

soumyasanyal commented 3 years ago

Well, we observed varied benefits of using higher-order neighborhoods for encoding knowledge graphs based on the score function. Usually, 1 layer is quite good but sometimes multiple layers can show better performance. You can check out the commands for the different score functions we have in the Readme, in some of them, we used multiple layers.

Megh-Thakkar commented 3 years ago

Got it! Thanks again.