[Question] Running Inference on New Nodes

facebookresearch / poincare-embeddings

PyTorch implementation of the NIPS-17 paper "Poincaré Embeddings for Learning Hierarchical Representations"

Other

1.68k stars 231 forks source link

[Question] Running Inference on New Nodes #67

Open kheyer opened 4 years ago

kheyer commented 4 years ago

Lets say I have a graph that I've embedded in a Poincare space and I have trained embeddings for all nodes in the graph.

Now I want to predict the embedding of a new node. In this scenario, I know links/edges between the new node and existing embedded nodes.

Is it possible to do this?

lematt1991 commented 4 years ago

It's a good question and not something that we've experimented with. One idea is to use the Einstein midpoint as is done in this paper to aggregate points in hyperbolic space. You'd just do this aggregation for all nodes that the new node is connected to and treat this aggregation as the new node's embedding.