Node embeddings in one line: new gds.embedding endpoints for Aura Graph Analytics (preview)

Getting embeddings for your graph no longer means choosing an algorithm or tuning hyperparameters. The GDS Python client now includes two new

preview

endpoints, available only on Aura Graph Analytics:

gds.embedding.create

, which generates node embeddings with sensible defaults, and

gds.embedding.train

, which trains a model on your node properties and known classes.

Use it to:

Get node embeddings without picking a model, tuning parameters, or knowing graph ML
Power similarity search, clustering, classification, and recommendations from the resulting vectors
Train an embedding model on nodes you have already classified, such as accounts marked by risk level, to get embeddings tuned to that task, and store the model to reuse it in later sessions

How to use it 🛠️

create

adds the embeddings to your in-memory graph as a node property (

embedding

by default). From there, they are yours to use: stream them into pandas, feed them to other algorithms in the same session, or write them back to your database.

gds.embedding.create(G)

That is the whole call: with no encoder specified, it runs FastRP with sensible settings chosen for you. For more control, you can set options such as

embedding_dimension

random_seed

When the default isn't enough,

gds.embedding.train

trains a model that works like any other trained model in Aura Graph Analytics: pass its name to

create

, and optionally store it with

gds.model.store

to reuse it from later sessions through your project's model catalog.

Both endpoints are in preview and may change as we gather feedback. They require GDS Python client 2.2 or later. See the Embeddings API docs, and let us know what you think.