Provided Matérn container classes¶

The gpmpcontrib.models package contains preconfigured Matérn model classes. They configure ModelContainer and gpmp.core.Model for common Matérn GP models. Each class installs adapters that build gpmp.parameter.Param objects from raw meanparam and covparam vectors.

Classes using REMAP selection build prior-regularized criteria when select_params is called. For classes with priors, use get_prior and set_prior on the model container to inspect or set per-output prior values. Use model[k].get_param() to inspect selected parameters in named form.

Constructor contracts¶

Model_ConstantMean_Maternp_ML(name, output_dim, covariance_specification=None): Constant mean with one explicit mean parameter per output. Parameters are selected by maximum likelihood. covariance_specification must provide {"p": int} for the Matérn regularity.
Model_ConstantMean_Maternp_REML(name, output_dim, mean_specification, covariance_specification): Linear-predictor mean with covariance parameters selected by restricted likelihood. mean_specification accepts {"type": "constant"} or {"type": "linear"}. covariance_specification must provide {"p": int}.
Model_ConstantMean_Maternp_REMAP_logsigma2(name, output_dim, mean_specification, covariance_specification, gamma=None, sigma2_coverage=None): Matérn class using REMAP selection with a Gaussian prior on log(sigma^2). Missing prior hyperparameters are resolved from gpmp.kernel.prior_defaults when select_params builds the criterion.
Model_ConstantMean_Maternp_REMAP_logsigma2_and_logrho_prior(name, output_dim, mean_specification, covariance_specification, gamma=None, sigma2_coverage=None, alpha=None, rho_min_range_factor=None, logrho_min=None, covparam0_prior=None, logsigma2_0_prior=None, logrho_0_prior=None): Matérn class using REMAP selection with priors on log(sigma^2) and logrho. Missing prior anchors are resolved from covparam0_prior when provided, then from the anisotropic initial guess computed on the current observations. Direct anchors logsigma2_0_prior and logrho_0_prior take priority over covparam0_prior.
Model_ConstantMean_Maternp_REMAP: Alias of Model_ConstantMean_Maternp_REMAP_logsigma2_and_logrho_prior.

Prior access¶

set_prior(*, gamma, sigma2_coverage, alpha, rho_min_range_factor, logrho_min, covparam0_prior, logsigma2_0_prior, logrho_0_prior, output_idx=None): Set prior values on classes using REMAP selection with priors. output_idx=None applies the supplied values to every output. covparam0_prior is in covariance coordinates [log(sigma2), -log(rho_0), ...]. logrho_0_prior is in logrho coordinates, so it has the opposite sign from the stored lengthscale coordinates.

get_prior(output_idx=None, resolved=True): Return the resolved prior object for one output or a list of resolved prior objects for all outputs. Raises ValueError before select_params when required defaults or data-dependent anchors have not yet been resolved.

Matern-based GP model classes.

class gpmpcontrib.models.Model_ConstantMean_Maternp_ML(name, output_dim, covariance_specification=None)[source]¶

Bases: ModelContainer

GP model with constant mean and Matern covariance, parameters by ML.

build_covariance(output_idx, param)[source]¶

Build and return a covariance callable for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the covariance function

Returns:

A covariance function.

Return type:

callable

build_mean_function(output_idx, param)[source]¶

Build and return (mean_callable, n_mean_params) for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the mean function

Returns:

The corresponding mean function and the number of parameters.

Return type:

(callable, int)

build_param_procedures(output_idx, **kwargs)[source]¶

Return Param procedures

Parameters:: output_idx (int)

build_parameters_initial_guess_procedure(output_idx, **build_param)[source]¶

Build an initial guess procedure for anisotropic parameters.

Override in subclass.

Returns:: A function to compute initial guesses for anisotropic parameters.
Return type:: function
Parameters:: output_idx (int)

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

gpmpcontrib.models.Model_ConstantMean_Maternp_REMAP¶: alias of Model_ConstantMean_Maternp_REMAP_logsigma2_and_logrho_prior

class gpmpcontrib.models.Model_ConstantMean_Maternp_REMAP_logsigma2(name, output_dim, mean_specification, covariance_specification, gamma=None, sigma2_coverage=None)[source]¶

Bases: Model_ConstantMean_Maternp_REML

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

get_prior(output_idx=None, resolved=True)[source]¶

Return type:: LogSigma2Prior | list[LogSigma2Prior]

set_prior(*, gamma=<object object>, sigma2_coverage=<object object>, covparam0_prior=<object object>, logsigma2_0_prior=<object object>, output_idx=None)[source]¶

class gpmpcontrib.models.Model_ConstantMean_Maternp_REMAP_logsigma2_and_logrho_prior(name, output_dim, mean_specification, covariance_specification, gamma=None, sigma2_coverage=None, alpha=None, rho_min_range_factor=None, logrho_min=None, covparam0_prior=None, logsigma2_0_prior=None, logrho_0_prior=None)[source]¶

Bases: Model_ConstantMean_Maternp_REML

REMAP Matern model with Gaussian prior on log(sigma^2) and barrier prior on logrho.

Parameters:

name (str) – Model name.
output_dim (int) – Number of outputs.
mean_specification (dict | list[dict]) – Mean-function specification forwarded to ModelContainer.
covariance_specification (dict | list[dict]) – Covariance-function specification forwarded to ModelContainer.
gamma (float, optional) – Multiplicative calibration factor for the Gaussian prior on log(sigma^2).
sigma2_coverage (float, optional) – Coverage probability used to calibrate the Gaussian prior on log(sigma^2).
alpha (float, optional) – Right-tail slope parameter for the logrho barrier-linear prior.
rho_min_range_factor (float, optional) – Safeguard factor used when inferring logrho_min from observation points.
logrho_min (array_like, optional) – Optional fixed lower bound for logrho. If None, it is inferred from data.
covparam0_prior (array_like, optional) – Optional prior anchor in covariance-parameter space [log(sigma^2), loginvrho_1, ...] used to derive missing prior centers.
logsigma2_0_prior (float, optional) – Optional direct prior center for log(sigma^2).
logrho_0_prior (array_like, optional) – Optional direct prior center for logrho.

Notes

Precedence for prior centers: 1. logsigma2_0_prior / logrho_0_prior when provided. 2. Otherwise derived from covparam0_prior when provided. 3. Otherwise derived from anisotropic initial guess on current data.

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

get_prior(output_idx=None, resolved=True)[source]¶

Return type:: LogSigma2AndLogRhoPrior | list[LogSigma2AndLogRhoPrior]

set_prior(*, gamma=<object object>, sigma2_coverage=<object object>, alpha=<object object>, rho_min_range_factor=<object object>, logrho_min=<object object>, covparam0_prior=<object object>, logsigma2_0_prior=<object object>, logrho_0_prior=<object object>, output_idx=None)[source]¶

class gpmpcontrib.models.Model_ConstantMean_Maternp_REML(name, output_dim, mean_specification, covariance_specification)[source]¶

Bases: ModelContainer

build_covariance(output_idx, covariance_build_param)[source]¶

Build and return a covariance callable for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the covariance function
covariance_build_param (dict)

Returns:

A covariance function.

Return type:

callable

build_mean_function(output_idx, mean_build_param)[source]¶

Build and return (mean_callable, n_mean_params) for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the mean function
mean_build_param (dict)

Returns:

The corresponding mean function and the number of parameters.

Return type:

(callable, int)

build_param_procedures(output_idx, **kwargs)[source]¶

Return Param procedures

Parameters:: output_idx (int)

build_parameters_initial_guess_procedure(output_idx, **build_param)[source]¶

Build an initial guess procedure for anisotropic parameters.

Override in subclass.

Returns:: A function to compute initial guesses for anisotropic parameters.
Return type:: function
Parameters:: output_idx (int)

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

class gpmpcontrib.models.Model_Noisy_ConstantMean_Maternp_REML(name, output_dim, mean_specification, covariance_specification)[source]¶

Bases: ModelContainer

build_covariance(output_idx, covariance_build_param)[source]¶

Build and return a covariance callable for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the covariance function
covariance_build_param (dict)

Returns:

A covariance function.

Return type:

callable

build_mean_function(output_idx, mean_build_param)[source]¶

Build and return (mean_callable, n_mean_params) for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the mean function
mean_build_param (dict)

Returns:

The corresponding mean function and the number of parameters.

Return type:

(callable, int)

build_param_procedures(output_idx, **kwargs)[source]¶

Return Param procedures

Parameters:: output_idx (int)

build_parameters_initial_guess_procedure(output_idx, **build_param)[source]¶

Build an initial guess procedure for anisotropic parameters.

Override in subclass.

Returns:: A function to compute initial guesses for anisotropic parameters.
Return type:: function
Parameters:: output_idx (int)

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

Maximum likelihood models¶

ML Matern model classes.

class gpmpcontrib.models.models_ML.Model_ConstantMean_Maternp_ML(name, output_dim, covariance_specification=None)[source]¶

Bases: ModelContainer

GP model with constant mean and Matern covariance, parameters by ML.

build_covariance(output_idx, param)[source]¶

Build and return a covariance callable for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the covariance function

Returns:

A covariance function.

Return type:

callable

build_mean_function(output_idx, param)[source]¶

Build and return (mean_callable, n_mean_params) for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the mean function

Returns:

The corresponding mean function and the number of parameters.

Return type:

(callable, int)

build_param_procedures(output_idx, **kwargs)[source]¶

Return Param procedures

Parameters:: output_idx (int)

build_parameters_initial_guess_procedure(output_idx, **build_param)[source]¶

Build an initial guess procedure for anisotropic parameters.

Override in subclass.

Returns:: A function to compute initial guesses for anisotropic parameters.
Return type:: function
Parameters:: output_idx (int)

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

Classes with restricted likelihood selection¶

REML Matern model classes.

class gpmpcontrib.models.models_REML.Model_ConstantMean_Maternp_REML(name, output_dim, mean_specification, covariance_specification)[source]¶

Bases: ModelContainer

build_covariance(output_idx, covariance_build_param)[source]¶

Build and return a covariance callable for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the covariance function
covariance_build_param (dict)

Returns:

A covariance function.

Return type:

callable

build_mean_function(output_idx, mean_build_param)[source]¶

Build and return (mean_callable, n_mean_params) for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the mean function
mean_build_param (dict)

Returns:

The corresponding mean function and the number of parameters.

Return type:

(callable, int)

build_param_procedures(output_idx, **kwargs)[source]¶

Return Param procedures

Parameters:: output_idx (int)

build_parameters_initial_guess_procedure(output_idx, **build_param)[source]¶

Build an initial guess procedure for anisotropic parameters.

Override in subclass.

Returns:: A function to compute initial guesses for anisotropic parameters.
Return type:: function
Parameters:: output_idx (int)

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

Classes with REMAP selection¶

Matern classes using REMAP parameter selection.

This module provides classes using REMAP selection, built on top of Model_ConstantMean_Maternp_REML.

Public model classes¶

Model_ConstantMean_Maternp_REMAP_power_laws: REMAP criterion with the power-law prior from gpmp.kernel.
Model_ConstantMean_Maternp_REMAP_logsigma2: REMAP criterion with a Gaussian prior on log(sigma^2).
Model_ConstantMean_Maternp_REMAP_logsigma2_and_logrho_prior: REMAP criterion with a Gaussian prior on log(sigma^2) and a barrier-linear prior on logrho.
Model_ConstantMean_Maternp_REMAP: Alias of Model_ConstantMean_Maternp_REMAP_logsigma2_and_logrho_prior.

Prior containers¶

LogSigma2Prior
LogSigma2AndLogRhoPrior

These classes hold resolved prior values (per output) and expose dictionary-like field access through __getitem__.

Prior resolution policy¶

For classes with configurable REMAP priors, values are stored per output and resolved when selection criteria are built from the current (model, xi, zi) context. Explicit user anchors (for example logsigma2_0_prior and logrho_0_prior) take precedence. Otherwise, anchors are inferred from covparam0_prior or from anisotropic initial guesses.

class gpmpcontrib.models.models_REMAP.LogSigma2AndLogRhoPrior(gamma=None, sigma2_coverage=None, alpha=None, rho_min_range_factor=None, logrho_min=None, covparam0=None, logsigma2_0_prior=None, logrho_0_prior=None, log_sigma2_0=None, logrho_0=None, covparam0_param_object=None)[source]¶

Bases: _PriorAccess

Parameters:: covparam0_param_object (Any | None)

class gpmpcontrib.models.models_REMAP.LogSigma2Prior(gamma=None, sigma2_coverage=None, covparam0=None, logsigma2_0_prior=None, log_sigma2_0=None, covparam0_param_object=None)[source]¶

Bases: _PriorAccess

Parameters:: covparam0_param_object (Any | None)

gpmpcontrib.models.models_REMAP.Model_ConstantMean_Maternp_REMAP¶: alias of Model_ConstantMean_Maternp_REMAP_logsigma2_and_logrho_prior

class gpmpcontrib.models.models_REMAP.Model_ConstantMean_Maternp_REMAP_logsigma2(name, output_dim, mean_specification, covariance_specification, gamma=None, sigma2_coverage=None)[source]¶

Bases: Model_ConstantMean_Maternp_REML

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

get_prior(output_idx=None, resolved=True)[source]¶

Return type:: LogSigma2Prior | list[LogSigma2Prior]

set_prior(*, gamma=<object object>, sigma2_coverage=<object object>, covparam0_prior=<object object>, logsigma2_0_prior=<object object>, output_idx=None)[source]¶

class gpmpcontrib.models.models_REMAP.Model_ConstantMean_Maternp_REMAP_logsigma2_and_logrho_prior(name, output_dim, mean_specification, covariance_specification, gamma=None, sigma2_coverage=None, alpha=None, rho_min_range_factor=None, logrho_min=None, covparam0_prior=None, logsigma2_0_prior=None, logrho_0_prior=None)[source]¶

Bases: Model_ConstantMean_Maternp_REML

REMAP Matern model with Gaussian prior on log(sigma^2) and barrier prior on logrho.

Parameters:

name (str) – Model name.
output_dim (int) – Number of outputs.
mean_specification (dict | list[dict]) – Mean-function specification forwarded to ModelContainer.
covariance_specification (dict | list[dict]) – Covariance-function specification forwarded to ModelContainer.
gamma (float, optional) – Multiplicative calibration factor for the Gaussian prior on log(sigma^2).
sigma2_coverage (float, optional) – Coverage probability used to calibrate the Gaussian prior on log(sigma^2).
alpha (float, optional) – Right-tail slope parameter for the logrho barrier-linear prior.
rho_min_range_factor (float, optional) – Safeguard factor used when inferring logrho_min from observation points.
logrho_min (array_like, optional) – Optional fixed lower bound for logrho. If None, it is inferred from data.
covparam0_prior (array_like, optional) – Optional prior anchor in covariance-parameter space [log(sigma^2), loginvrho_1, ...] used to derive missing prior centers.
logsigma2_0_prior (float, optional) – Optional direct prior center for log(sigma^2).
logrho_0_prior (array_like, optional) – Optional direct prior center for logrho.

Notes

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

get_prior(output_idx=None, resolved=True)[source]¶

Return type:: LogSigma2AndLogRhoPrior | list[LogSigma2AndLogRhoPrior]

set_prior(*, gamma=<object object>, sigma2_coverage=<object object>, alpha=<object object>, rho_min_range_factor=<object object>, logrho_min=<object object>, covparam0_prior=<object object>, logsigma2_0_prior=<object object>, logrho_0_prior=<object object>, output_idx=None)[source]¶

class gpmpcontrib.models.models_REMAP.Model_ConstantMean_Maternp_REMAP_power_laws(name, output_dim, mean_specification, covariance_specification)[source]¶

Bases: Model_ConstantMean_Maternp_REML

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

Noisy classes with restricted likelihood selection¶

Noisy-output REML Matern model classes.

class gpmpcontrib.models.models_noisy_REML.Model_Noisy_ConstantMean_Maternp_REML(name, output_dim, mean_specification, covariance_specification)[source]¶

Bases: ModelContainer

build_covariance(output_idx, covariance_build_param)[source]¶

Build and return a covariance callable for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the covariance function
covariance_build_param (dict)

Returns:

A covariance function.

Return type:

callable

build_mean_function(output_idx, mean_build_param)[source]¶

Build and return (mean_callable, n_mean_params) for the given output.

Override in subclasses.

Parameters:

output_idx (int) – The index of the output for which the covariance function is being created.
param (dict) – Additional parameters for the mean function
mean_build_param (dict)

Returns:

The corresponding mean function and the number of parameters.

Return type:

(callable, int)

build_param_procedures(output_idx, **kwargs)[source]¶

Return Param procedures

Parameters:: output_idx (int)

build_parameters_initial_guess_procedure(output_idx, **build_param)[source]¶

Build an initial guess procedure for anisotropic parameters.

Override in subclass.

Returns:: A function to compute initial guesses for anisotropic parameters.
Return type:: function
Parameters:: output_idx (int)

build_selection_criterion(output_idx, context=None, **build_params)[source]¶

Override in subclass to construct a selection criterion.

Parameters:: output_idx (int)

gpmpcontrib.models.models_noisy_REML.build_mown_kernel(output_idx, **params)[source]¶

Build Matern kernel handling per-output noise dimensions.

Parameters:: output_idx (int)

gpmpcontrib.models.models_noisy_REML.noisy_outputs_parameters_initial_guess(model, xi, zi, output_dim)[source]¶: Initial guess for noisy multi-output models, using spatial dimensions only.