CartoBoost NeuralEmbeddingRegressor

Use NeuralEmbeddingRegressor when you want one wrapper that learns ID embeddings, appends them to the dense row features, and fits a CartoBoost regressor.

When To Use

• Stable IDs carry repeated residual signal.
• You want a single object for the embedding fit and final regressor fit.
• You need a direct comparison against the same model without embedding columns.
• You can validate repeated-ID and cold-ID behavior separately.

Interactive Example

Neural embedding regressor browser model

Runs embedding in the browser with the bundled CartoBoost Wasm model.

Ready to run in this page.

Python Example

from cartoboost.neural import NeuralEmbeddingRegressor

model = NeuralEmbeddingRegressor(
    dim=8,
    n_estimators=200,
    random_state=7,
)
model.fit(X_train, y_train, ids=train_ids)
pred = model.predict(X_valid, ids=valid_ids)

Inputs

Input	Meaning
X	Dense row features for the final regressor.
y	Numeric target.
ids	Stable high-cardinality IDs used to learn embedding vectors.

Use When

Need	Better surface
Serve the supervised ID model directly.	NeuralEmbeddingStandaloneRegressor
Generate reusable embedding columns.	NeuralEmbeddingFeatures
Fit embeddings and a CartoBoost regressor together.	NeuralEmbeddingRegressor

Validation

Compare against a non-neural CartoBoostRegressor on the same dense features. If the gain disappears under cold-ID validation, describe the model as capturing repeated-ID signal rather than unseen-entity structure.

Limitations

• The wrapper adds training cost and tuning choices beyond the base regressor.
• Unseen IDs use documented fallback behavior rather than a learned identity effect.
• Any comparison must give the non-neural baseline the same dense features and split.