CartoBoost GraphNeuralOperator

Use GraphNeuralOperator for advanced experimental field-to-field mapping on regional or gridded panels. It consumes field values, coordinates, graph edges, and optional exogenous fields, then returns future, residual, and uncertainty fields.

Python Example

from cartoboost.deep import GraphNeuralOperator

operator = GraphNeuralOperator(smoothing=0.25, coordinate_scale=0.1)
prediction = operator.predict(
    field_values=[[42, 35, 18], [44, 36, 19], [51, 40, 24]],
    coordinates=[[0.0, 0.0], [0.5, 0.4], [1.0, 0.1]],
    edges=[{"source": 0, "target": 1, "weight": 0.7}],
    exogenous_fields=[[0.1, 0.2, 0.0], [0.1, 0.3, 0.1], [0.2, 0.2, 0.1]],
)
benchmark = GraphNeuralOperator.synthetic_benchmark()

FourierGeoOperator and SpatioTemporalOperator are aliases.

Use When

Use this experimental operator for field-to-field prediction when coordinates and graph structure define the output domain. Start with kriging or a pointwise model for ordinary interpolation tasks.

Browser WASM Example

GraphNeuralOperator Wasm example

Runs the Rust-backed browser export on a substantial multi-route taxi panel with a held-out prediction window.

Ready to run in this page.

Validation

Report the maintained synthetic benchmark and compare against a pointwise MLP proxy before making any field-transfer claim.

Limitations

• Current evidence is synthetic and mechanism-oriented.
• Coordinate scaling and graph construction strongly affect learned fields.
• Real-data use needs comparison with pointwise and spatial baselines.