Neural Embedding Models

Use neural models when stable identifiers carry repeated residual signal. These guides are for learned ID structure, not for graph relationships or forecasting panels.

Use CartoBoost Forecasting Model Guides for NeuralPanelForecaster and LaneNeuralPanelForecaster, and CartoBoost Graph Model Guides when the relationship network itself is the thing being modeled. Use CartoBoost Deep Model Guides for ordered pairs, response curves, event probabilities, residual correction, graph sequences, and constrained candidate selection under cartoboost.deep.

Choose A Guide

Guide	Best when
CartoBoost Neural Embedding Regressor	The supervised ID embedding model is the artifact you train, score, save, and serve.
CartoBoost NeuralEmbeddingFeatures	Learned ID vectors should become columns for another estimator or ablation.
CartoBoost NeuralEmbeddingRegressor	One wrapper should learn ID embeddings and fit the downstream CartoBoost regressor.

See Neural Features for input shapes, saving and loading, unseen-ID behavior, and common failure modes.