Package index • gaussianprocesses

Stable interfaces

From version 1.0.0 these change incompatibly only in a major release, after a deprecation period. Results and options that concern an experimental model class, kernel, or argument follow that interface’s tier (see the package help page, ?gaussianprocesses).

Exact Gaussian-process regression

fit_gp(): Fit an exact Gaussian-process regression model
predict_gp(): Predict from an exact Gaussian-process regression model
log_marginal_likelihood(): Gaussian-process log marginal likelihood
log_marginal_likelihood_gradient(): Gradient of the Gaussian-process log marginal likelihood
optimize_gp(): Optimize Gaussian-process hyperparameters
gaussianprocesses_model gaussianprocesses_sparse_model gaussianprocesses_heteroscedastic_model gaussianprocesses_latent_model gaussianprocesses_state_space_model: Fitted model objects
predict(<gaussianprocesses_model>) predict(<gaussianprocesses_sparse_model>) predict(<gaussianprocesses_state_space_model>) predict(<gaussianprocesses_heteroscedastic_model>) predict(<gaussianprocesses_latent_model>) logLik(<gaussianprocesses_model>) logLik(<gaussianprocesses_sparse_model>) logLik(<gaussianprocesses_state_space_model>) logLik(<gaussianprocesses_heteroscedastic_model>) logLik(<gaussianprocesses_latent_model>) nobs(<gaussianprocesses_model>) nobs(<gaussianprocesses_sparse_model>) nobs(<gaussianprocesses_state_space_model>) nobs(<gaussianprocesses_heteroscedastic_model>) nobs(<gaussianprocesses_latent_model>) fitted(<gaussianprocesses_model>) fitted(<gaussianprocesses_sparse_model>) fitted(<gaussianprocesses_state_space_model>) fitted(<gaussianprocesses_heteroscedastic_model>) fitted(<gaussianprocesses_latent_model>) residuals(<gaussianprocesses_model>) residuals(<gaussianprocesses_sparse_model>) residuals(<gaussianprocesses_state_space_model>) residuals(<gaussianprocesses_heteroscedastic_model>) summary(<gaussianprocesses_model>) summary(<gaussianprocesses_sparse_model>) summary(<gaussianprocesses_state_space_model>) summary(<gaussianprocesses_heteroscedastic_model>) summary(<gaussianprocesses_latent_model>) print(<summary.gaussianprocesses_model>): Base R methods for fitted Gaussian-process models

Sampling and simulation

Reproducible function draws from priors and posteriors.

sample_gp_prior(): Draw functions from a Gaussian-process prior
sample_gp_posterior(): Draw functions from a Gaussian-process posterior
simulate_gp_data(): Simulate observations from a Gaussian process

Kernel specifications

Reusable kernel objects. Combine them, read and update their parameters, and pass them to models; evaluate_kernel() computes their covariance.

rbf_kernel(): Create an RBF kernel specification
matern12_kernel() matern32_kernel() matern52_kernel(): Create Matérn kernel specifications
rational_quadratic_kernel(): Create a rational-quadratic kernel specification
periodic_kernel(): Create a periodic kernel specification
linear_kernel(): Create a linear kernel specification
white_noise_kernel(): Create a white-noise kernel specification
sum_kernel(): Add covariance kernels
product_kernel(): Multiply covariance kernels
scale_kernel(): Scale a covariance kernel
select_dimensions(): Restrict a kernel to selected input columns
evaluate_kernel(): Evaluate a kernel specification
kernel_diagonal(): Evaluate only the diagonal of a kernel matrix
kernel_parameters(): Extract kernel parameters
update_kernel_parameters(): Update parameters in a kernel specification
kernel_gradient(): Derivatives of a covariance matrix with respect to log hyperparameters

Covariance matrices

Functions that compute one covariance matrix directly from parameter values, without a kernel object. They use the same formulas as the kernel specifications.

kernel_rbf(): Squared-exponential covariance kernel
kernel_matern12() kernel_matern32() kernel_matern52(): Matérn covariance kernels
kernel_rational_quadratic(): Rational-quadratic covariance kernel
kernel_periodic(): Periodic covariance kernel
kernel_linear(): Linear covariance kernel
kernel_white_noise(): White-noise covariance kernel

Mean functions

Zero, constant, and parametric means. The coefficients of parametric means are fixed, estimated by generalized least squares, or marginalized under a Gaussian prior.

zero_mean() constant_mean(): Gaussian-process mean functions
linear_mean() polynomial_mean() basis_mean(): Parametric mean functions
estimate_coefficients() coefficient_prior(): Treatments of mean-function coefficients
coef(<gaussianprocesses_latent_model>) coef(<gaussianprocesses_model>) vcov(<gaussianprocesses_model>): Coefficients of a Gaussian-process mean
evaluate_mean(): Evaluate a Gaussian-process mean function

Leave-one-out prediction and scores

loo_gp(): Exact leave-one-out Gaussian-process diagnostics
gp_diagnostics(): Comprehensive diagnostics for an exact Gaussian-process model
gp_scores() gp_holdout_scores() gp_loo_scores(): Proper scoring rules for Gaussian predictions

Time series

gp_time_index(): Build an explicit Gaussian-process time index
time_series_kernel(): Create a trend-periodic-local time-series kernel
fit_time_series_gp(): Fit a GP to ordered time-series observations
forecast_gp(): Forecast or interpolate from a time-aware GP model
rolling_origin_gp(): Rolling-origin evaluation for time-aware Gaussian processes
forecast_metrics(): Uncertainty-aware forecast metrics

Experimental interfaces

These may change in a minor release; every change is listed in NEWS.

Sparse approximations (FITC and VFE)

fit_sparse_gp(): Fit a sparse Gaussian process (FITC or VFE)
optimize_sparse_gp(): Optimize the hyperparameters of a sparse Gaussian process
predict_sparse_gp(): Predict from a sparse Gaussian-process model
select_inducing_points(): Select inducing points deterministically

Latent models and likelihoods

Gaussian processes observed through a likelihood, such as binary responses or counts, with the Laplace approximation, and the likelihoods and scores they use.

fit_latent_gp(): Fit a latent Gaussian-process model
optimize_latent_gp(): Optimize the hyperparameters of a latent Gaussian-process model
predict_latent_gp(): Predict from a latent Gaussian-process model
gaussian_likelihood() bernoulli_likelihood() poisson_likelihood(): Likelihoods for Gaussian-process models
evaluate_likelihood(): Evaluate a likelihood and its derivatives
likelihood_predictive(): Predictive distribution of new responses
gp_classification_scores(): Scores of binary probability forecasts
gp_count_scores(): Scores of count forecasts

Heteroscedastic noise

fit_heteroscedastic_gp(): Fit an approximate heteroscedastic Gaussian process
predict_heteroscedastic_gp(): Predict from an approximate heteroscedastic GP

Recent kernels and methods

State-space inference for time series, spectral-mixture and changepoint kernels, and derivatives of Gaussian processes.

optimize_time_series_gp(): Optimize the hyperparameters of a time-series GP
spectral_mixture_kernel(): Spectral-mixture kernels
initialize_spectral_mixture(): Initialise a spectral-mixture kernel from the data's spectrum
changepoint_kernel(): Changepoint kernels
predict_gradient_gp(): Predict the gradient of the latent function
kernel_input_gradient(): Derivatives of a covariance matrix with respect to the inputs

Hyperparameter uncertainty and diagnostics

gp_hyperparameter_uncertainty(): Uncertainty of estimated hyperparameters
gp_profile_likelihood(): Profile likelihood of one hyperparameter
gp_numerical_diagnostics(): Summarize numerical conditioning of a fitted GP
gp_calibration_diagnostics(): Summarize Gaussian-process uncertainty calibration

Multi-output Gaussian processes

Coregionalization kernels for several correlated outputs, observed at the same or different inputs.

coregionalization_kernel(): Create a coregionalization kernel for several outputs
coregionalization_matrix(): Output covariance matrices of coregionalization kernels
initialize_coregionalization(): Initialize a coregionalization kernel from the data
gp_stack_outputs(): Stack several outputs into one input matrix

Benchmarks and simulation scenarios

gp_simulation_scenario(): Generate standard GP simulation scenarios
run_gp_benchmark_suite(): Run the standard GP benchmark suite
benchmark_exact_gp_scaling(): Benchmark exact-GP scaling
benchmark_sparse_gp(): Benchmark sparse FITC against exact GP regression
compare_gp_predictions(): Compare two GP posterior predictions numerically