CHANGELOG.md

CHANGELOG

[2.0.0] (2025-02-05)

Requirements

• Set matplotlib version back to 3.7.3 to ensure stability in combination with seaborn

Added

Features

• caculate_moments method in Meta_Model class for a general calculation of the moments
• PCEGPR class for combining polynomial chaos and gaussian processes

Tests

• Tests for GPE class
• Tests for calculate_moments method in Meta_Model and GPESkl classes
• Tests for PCEGPR class
• Tests for PostProcessingclass

Examples

• Example analytical_function_pcegpr.py to show application of the combined class PCEGPR

Changed

• Split surrogate classes into template class MetaModel and child classes PCE and GPESklearn
• Changed Exception format in GPE class to more specific Exceptions
• Extend Inputs.add_marginals to allow direct setting of marginal properties
• Move Sobol' calculation to PCE.calculate_sobol
• Clean PostProcessing class and update for general MetaModel objects
• Split BayesInferenceclass into BayesInference, template class PostSampler and child class RejectionSampler
• Class MCMC child class of PostSampler
• Split BayesInference.create_inference() into BayesInference.run_inference()and BayesInference.run_validation()
• Plot in PCE.adaptive_regression is saved automatically
• Moved adaptPlot() to be Engine.plot_adapt()

Renaming

• Renamed attributes of classes BayesInference, BMC, MCMC, SeqDesign and Engine following from structure reworking
• Renamed attribues and functions to snake-style naming:
  • *.Discrepancy to *.discrepancy
  • surrogate_models.py to meta_model.py
  • *.MetaModel to *.meta_model
  • *.ExpDesign to *.exp_design
  • *.Model to *.model
  • *LCerror to *lc_error
  • *.InputSpace to *.input_space
  • *.Marginals to *.marginals
  • *.Rosenblatt to *.rosenblatt
  • *.add_InputSpace to *.add_input_space
  • *.InputObj to *.input_object
  • *.JDist to *.j_dist
  • ._autoSelect to *._auto_select
  • ExpDesigns.generate_ED to ExpDesigns.generate_ed
  • ExpDesigns.X to ExpDesigns.x
  • ExpDesigns.Y to ExpDesigns.y
  • PyLinkForwardModel.Output to PyLinkForwardModel.output
• Class auto_vivification to AutoVivification

Bug fixes

• ExpDesign always uses user-defined samples when given
• ExpDesign n_init_samples can be user-defined
• Fixed pce-loocv calculation for few training samples
• Two-step coefficient estimation for LARS PCE training options

Removed

• Input n_samples for ExpDesign.generate_ED()
• Reduced Discrepancy options to only total_sigma2

[1.1.0]

Requirements

• numpy >= 1.23.5

Added

Features

• class SeqDesign for sequential training
• Engine can be built without a surrogate
• BayesInference and BayesModelComparison can be performed on an Engine object without a surrogate

Examples

• Example user_guide to go along with the user guide on the website
• Example principal_component_analysis to show application of pca on metamodel outputs
• Example 'only_model' for use of inference and model comparison without a metamodel

Changed

• Moved functions for sequential training from Engine to SeqDesign
• Moved hellinger_distance, logpdf, subdomain into surrogate_models/seq_design
• Early stop in BayesInf for improved performance of BayesModelComp
• Allow singular matrices in exploitation with BayesActDesign
• ExpDesign.generate_ED no longer needs transform

Bug fixes

• Import of ExpDesign allowed
• Images in PostProcessing only saved, not opened
• Fixed option MetaModel.dim_red_method = 'pca'

Removed

• Disabled exploration with voronoi
• BayesModelComp.just_n_meas

[1.0.0]

Requirements

• numpy now at 1.23.3
• ....

Added

Features

• PyLinkForwardModel has new link_type 'umbridge' for UM-Bridge type models
• class InputSpace as parent class to ExpDesigns
• MetaModel has option to train with OLS+constraints

Examples

• Example umbridge_tsunamitutorial to show two options of using UM-Bridge type models
• Example convergence_tests for the constraints

Changed

General

• Independent functions moved out of classes
• Constructors of PostProcessing, BayesInference, BayesModelComp are to be given the engine instead of the metamodel
• Constructor of Exploration to be given ExpDesigns instead of MetaModel

Engine

• MetaModelEngine renamed to Engine
• Split run() into train_normal() and train_sequential()
• opt_SeqDesign() renamed to choose_next_sample()

MetaModel

• MetaModel fully independent of model
• MetaModel uses InputSpace instead of ExpDesigns
• create_metamodel() split into build_metamodel() and fit()
• fit() renamed to regression()
• eval_metamodel() only runs on given samples, full functionality retained in Engine.eval_metamodel()
• Polynomial related functions moved from MetaModelEngine to MetaModel

PyLinkForwardModel

• Read-in of MC-references now performed by PyLinkForwardModel.read_observation()

ExpDesigns

• Parameters related to sequential training were moved from MetaModel to ExpDesigns (Seq*, valid_model_runs)

Removed

• Class SequentialDesign
• ExpDesigns.method
• MetaModel.create_basis_indices()