User guide

The guide starts after the Hartmann4 run in Getting started. It explains how gpmp-contrib stores models, parameters, diagnostics, priors, and sequential-procedure state.

Main path

Start with these pages if the goal is to build and inspect a Gaussian-process model from observations:

1. Concepts gives the package map and states which objects belong to gpmp and which objects are added by gpmp-contrib.

2. Models and computer experiments explains ComputerExperiment, ModelContainer, the provided Matérn classes, and the shape conventions for xi and zi.

3. Parameter selection explains what select_params does, how optimizer starts and bounds are chosen, and when ML, REML, or REMAP is used.

4. Model state and parameter objects shows where selected parameters, Param objects, optimizer reports, and resolved prior objects are stored.

5. Diagnostics and inspection explains the printed diagnosis and prediction-performance summaries.

Specialized procedures

Read these pages when the model is already clear and the task needs one of these procedures:

• Priors for REMAP selection for REMAP models with explicit prior anchors or hyperparameters.

• Sequential design, optimization, and set estimation for expected improvement, excursion sets, set inversion, SMC particles, and BSS particles.

• Relaxed Gaussian processes for relaxed Gaussian-process modeling.

Relation to examples

The guide describes objects and stored state. The example pages show complete scripts, plots, and representative results. After reading a guide page, use Examples to find the closest executable script.

• Concepts
  • Object map
  • Common notation
  • Relation to gpmp
  • When to use gpmp-contrib
• Models and computer experiments
  • Shapes
  • ComputerExperiment
  • ModelContainer
  • Matérn model used by the provided classes
  • Posterior prediction
  • Mean and covariance specifications
  • Provided Matérn container classes
  • Constructor arguments
  • Minimal construction
  • Where to continue
• Parameter selection
  • Covariance parameter vector
  • Which criterion is used
  • Maximum likelihood
  • Restricted likelihood
  • REMAP
  • What select_params does
  • Optimizer starts
  • Bounds and SciPy options
  • Stored results
  • Failure diagnosis
• Model state and parameter objects
  • Per-output state
  • Raw covariance parameters
  • Parameter objects
  • Bounds
  • Optimizer information
  • Prior objects
• Diagnostics and inspection
  • Model diagnosis
  • Prediction performance
  • Model state checks
  • Selection-criterion inspection
  • Posterior parameter sampling
• Priors for REMAP selection
  • Inspect and set priors
  • Available prior objects
  • Log-variance prior
  • Log-lengthscale prior
  • Resolution policy
  • Checks
• Sequential design, optimization, and set estimation
  • Base objects
  • Predictive quantities used by criteria
  • Expected improvement
  • Excursion sets
  • Set inversion
  • SMC and BSS particles
  • Which strategy to use
• Relaxed Gaussian processes
  • Problem setup
  • Optimization problem
  • Main functions
  • Threshold selection
  • Shapes
  • Example
  • Returned state
  • Failure conditions