GraphML Quick Start

Predict missing labels and uncover hidden patterns in your graph with machine learning, no data science pipeline required

Prerequisites

Access to the platform with GraphML services enabled.
A database with graph data (node and edge collections).
For Node Classification: an attribute with known values on some nodes, marked with Predict this during featurization. For Node Embeddings: no labels required.

Train a model and predict

Create a project

In the Arango Contextual Data Platform, go to Agentic AI Suite > GraphML and click Add new project.

Featurize your graph

Select your graph, choose the node collections and attributes to use as features, configure the job, and click Begin featurization. To use Node Classification later, click Predict this on the attribute you want to predict.

Train a model

Choose the task type - Node Classification or Node Embeddings - set the parameters, and click Begin training.

Select the best model

Review the training metrics, pick the best-performing model, and click Select model for prediction.

Run a prediction

Configure the prediction job (optionally enable scheduling) and click Run Prediction.

Enable ArangoML

arangoml = %enable_arangoml
project = arangoml.get_or_create_project("my_project")

Featurize

job = arangoml.jobs.featurize(featurization_spec)
arangoml.wait_for_featurization(job.job_id)

Train

job = arangoml.jobs.train(training_spec)
arangoml.wait_for_training(job.job_id)

Select the best model

model = arangoml.get_best_model(
    "my_project", training_job_id, sort_parent_key, sort_child_key
)

Predict

job = arangoml.jobs.predict(prediction_spec)
arangoml.wait_for_prediction(job.job_id)

You now have predicted labels (Node Classification) or vector embeddings (Node Embeddings) written to your target collection, ready for clustering, similarity search, anomaly detection, or link prediction.

Next steps

Setup: The full walkthrough.
UI: The web interface in detail.
Notebooks API: All arangoml methods and specs.