grf_gp.model module

class grf_gp.model.ExactGraphGP(x_train, y_train, likelihood, kernel: BaseGRFKernel | BaseExactKernel)

Bases: GraphGP

Exact graph GP model using the full kernel matrix.

predict(x_test, **kwargs)

Evaluate the exact predictive distribution.

Parameters:

  • x_test – Test node indices.

  • kwargs – Unused prediction options.

Returns:

Predictive distribution evaluated at x_test.

class grf_gp.model.GRFGP(x_train, y_train, likelihood, kernel: BaseGRFKernel | BaseExactKernel)

Bases: GraphGP

Graph GP model that predicts by pathwise posterior sampling.

predict(x_test, batch_size=64)

Approximate the predictive distribution from posterior samples.

Parameters:

  • x_test – Test node indices.

  • batch_size – Number of posterior samples used for the estimate.

Returns:

Approximate predictive multivariate normal distribution.

predict_sample(x_test, n_samples=64)

Draw posterior samples at test nodes.

Parameters:

  • x_test – Test node indices.

  • n_samples – Number of posterior samples to draw.

Returns:

Posterior samples with shape (n_samples, len(x_test)).

class grf_gp.model.GraphGP(x_train, y_train, likelihood, kernel: BaseGRFKernel | BaseExactKernel)

Bases: ExactGP, ABC

Abstract base class for graph Gaussian process models.

forward(x)

Construct the GP prior at the requested inputs.

Parameters:

x – Input node indices.

Returns:

Multivariate normal prior distribution at x.

abstractmethod predict(x_test, **kwargs)

Predict at test inputs.

Parameters:

  • x_test – Test node indices.

  • kwargs – Model-specific prediction options.

Returns:

Predictive distribution or samples at x_test.

class grf_gp.model.LowRankGRFGP(x_train, y_train, likelihood, kernel: LowRankGRFKernel)

Bases: GRFGP

Low-rank graph GP model using Woodbury pathwise conditioning.

predict_sample(x_test, n_samples=64)

Draw posterior samples using the low-rank feature representation.

Parameters:

  • x_test – Test node indices.

  • n_samples – Number of posterior samples to draw.

Returns:

Posterior samples with shape (n_samples, len(x_test)).