search

awslabs / graphstorm

Enterprise graph machine learning framework for billion-scale graphs for ML scientists and data scientists.
Apache License 2.0
367 stars 59 forks source link

[Feature request] support inductive link-prediction splits in `partition_graph_lp.py` #642

Closed wangz10 closed 11 months ago

wangz10 commented 11 months ago

Currently the link-prediction split is done by randomly splitting all the edges. We need an option to hold out nodes and their links in test set for inductive setting.

classicsong commented 11 months ago

Can you be more specific of what does inductive setting mean?

wangz10 commented 11 months ago

@classicsong In inference time of inductive setting, there are new/unseen nodes from the training graph. The link-prediction model is tasked to predict links 1) between an unseen node and a seen node; or 2) between two unseen nodes.

classicsong commented 11 months ago

Can we close this @wangz10