rgpycrumbs.surfaces.standard

Classes

FastTPS	Thin Plate Spline (TPS) surface implementation.
FastMatern	Matérn 5/2 surface implementation.
FastIMQ	Inverse Multi-Quadratic (IMQ) surface implementation.

Functions

negative_mll_tps(log_params, x, y)
_tps_solve(x, y, sm)
_tps_predict(x_query, x_obs, w, v)
_tps_var(x_query, x_obs, lhs_inv)
negative_mll_matern_std(log_params, x, y)	Negative marginal log-likelihood for the standard Matern 5/2 model.
_matern_solve(x, y, sm, length_scale)
_matern_predict(x_query, x_obs, alpha, length_scale)
_matern_var(x_query, x_obs, K_inv, length_scale)
negative_mll_imq_std(log_params, x, y)
_imq_solve(x, y, sm, epsilon)
_imq_predict(x_query, x_obs, alpha, epsilon)
_imq_var(x_query, x_obs, K_inv, epsilon)

Module Contents

rgpycrumbs.surfaces.standard.negative_mll_tps(log_params, x, y)

rgpycrumbs.surfaces.standard._tps_solve(x, y, sm)

rgpycrumbs.surfaces.standard._tps_predict(x_query, x_obs, w, v)

rgpycrumbs.surfaces.standard._tps_var(x_query, x_obs, lhs_inv)

class rgpycrumbs.surfaces.standard.FastTPS(x_obs, y_obs, smoothing=0.001, optimize=True, **_kwargs)

Thin Plate Spline (TPS) surface implementation. Includes a polynomial mean function and supports smoothing optimization.

Added in version 1.0.0.

x_obs

y_obs

y_mean

__call__(x_query, chunk_size=500)

Predict values at query points using chunking.

Args:

x_query: Query inputs (M, D). chunk_size: Processing batch size.

Returns:

jnp.ndarray: Predicted values (M,).

predict_var(x_query, chunk_size=500)

Predict posterior variance at query points.

Added in version 1.1.0.

Args:

x_query: Query inputs (M, D). chunk_size: Processing batch size.

Returns:

jnp.ndarray: Predicted variances (M,).

rgpycrumbs.surfaces.standard.negative_mll_matern_std(log_params, x, y)

Negative marginal log-likelihood for the standard Matern 5/2 model.

Used as the objective function during hyperparameter optimization. The parameters are passed in log-space for unconstrained optimization.

Args:

log_params: Array of [log(length_scale), log(noise)]. x: Training inputs, shape (N, D). y: Training observations (centered), shape (N,).

Returns:

Scalar negative MLL value.

Added in version 1.0.0.

rgpycrumbs.surfaces.standard._matern_solve(x, y, sm, length_scale)

rgpycrumbs.surfaces.standard._matern_predict(x_query, x_obs, alpha, length_scale)

rgpycrumbs.surfaces.standard._matern_var(x_query, x_obs, K_inv, length_scale)

class rgpycrumbs.surfaces.standard.FastMatern(x_obs, y_obs, smoothing=0.001, length_scale=None, optimize=True, **_kwargs)

Bases: rgpycrumbs.surfaces._base.BaseSurface

Matérn 5/2 surface implementation.

Added in version 1.0.0.

_fit(smoothing, length_scale, optimize)

Internal method to perform parameter optimization.

_solve()

Internal method to solve the linear system for weights.

_predict_chunk(chunk)

Internal method for batch prediction.

_var_chunk(chunk)

Internal method for batch variance.

rgpycrumbs.surfaces.standard.negative_mll_imq_std(log_params, x, y)

rgpycrumbs.surfaces.standard._imq_solve(x, y, sm, epsilon)

rgpycrumbs.surfaces.standard._imq_predict(x_query, x_obs, alpha, epsilon)

rgpycrumbs.surfaces.standard._imq_var(x_query, x_obs, K_inv, epsilon)

class rgpycrumbs.surfaces.standard.FastIMQ(x_obs, y_obs, smoothing=0.001, length_scale=None, optimize=True, **_kwargs)

Bases: rgpycrumbs.surfaces._base.BaseSurface

Inverse Multi-Quadratic (IMQ) surface implementation.

Added in version 1.0.0.

_fit(smoothing, length_scale, optimize)

Internal method to perform parameter optimization.

_solve()

Internal method to solve the linear system for weights.

_predict_chunk(chunk)

Internal method for batch prediction.

_var_chunk(chunk)

Internal method for batch variance.