UQLab Release Notes

UQ with UQLab Developers Corner
1

In this post you can find all the detailed release notes for every UQLab release.

Table of contents

UQLab Modules 1.4.0 (01.02.2021)

UQLabModules V1.3.0 ➜ UQLabModules V1.4.0

Stable release of UQLabModules.

New features

  • Reliability analysis module:​​
    • Introduced a new framework for active learning reliability (Developed and documented by @moustapha)
    • Introduced asynchronous learning, a feature that allows users to interrupt an active learning analysis, run the computational model outside of UQLab, and then resume the analysis with a new evaluation
  • High-performance computing (HPC) dispatcher module:​
    • A new module to dispatch UQLab computations from local computing resources (e.g., laptops, desktops) to distributed computing resources (Developed and documented by @damarginal)
  • PCE module:
    • Introduced two new sparse solvers: Subspace pursuit (SP) and Bayesian compressive sensing (BCS) (Developed and documented by @nluethen)​
  • UQLib:​
    • Introduced uq_map, a new dispatcher-aware command to dispatch generic functions evaluations to distributed computing resources​​

Enhancements

  • UQLink​ module:
    • Unique IDs based on a timestamp for different runs of the same UQLink model is now supported ​​​
  • Bayesian inversion module:​
    • Simultaneous estimation of multiple point estimators (mean, map, and custom)​ is now allowed
  • Documentation​:
    • Major restructuring of the PCE module user manual​​, new sections on SP and BCS solvers,
      and a new instruction on how to add a custom sparse regression method
    • A new section on data groups in the Bayesian inversion user manual

Changes

  • MATLAB R2015b is now a minimum requirement for UQLab
  • PCE: The OMP solver always adds the constant regressor first
  • UQLink: Auxiliary files are now saved in a different folder for each run
  • Bayesian inversion: Predictive distributions are now computed on data groups not forward models

Bug fixes

  • Fixed problem in q-norm adaptivity for PCE and the displayed best q-norm
  • Fixed problem related to multiple soft constraints in the RBDO module
  • Fixed problem when a single forward model was explicitly supplied in the Bayesian inversion module

:arrow_up_small: Back to top

UQLab Modules 1.3.0 (19.09.2019)

UQLabModules V1.2.1 ➜ UQLabModules V1.3.0

Stable release of UQLabModules.

New features

  • Input module:
    • Major update on the module; it now includes new types of copulas (CVine and DVine) and supports independent sets of random variables as well as statistical inference for both marginals and copulas (Developed and documented by @torree)
  • Reliability-based design optimization (RBDO) module:​
    • A new module to conduct reliability-based design optimization is now available (Developed and documented by @moustapha)
  • Kriging module:
    • Gaussian process (GP) regression for noisy observations is now available
      (Developed and documented by @damarginal)​
  • Sensitivity analysis module:​​
    • New sample-based estimator for the Kucherenko indices​ (compatible with non-Gaussian copulas)
    • Borgonovo indices can now be computed from pre-existing samples​
  • UQLib:​
    • A collection of standard UQLab plotting and plot formatting functions is now consolidated in uq_graphics inside the lib folder (Developed and documented by @philippe, @paulremo, and @damarginal)​

Enhancements

  • Bayesian inversion module:
    • A sample generated by any MCMC sampler is automatically post-processed using the uq_postProcessInversion function at the end of an inverse analysis
    • Posterior covariance and correlation matrices are now estimated from the MCMC sample by the uq_postProcessInversion function​
  • UQLink​ module:
    • Mathematical expressions with input variables can now be entered in the template file​​​
  • Documentation​:
    • Kriging module:​ Add elaboration on the ​cross-validation estimation
  • Sensitivity module:
    • Chunk-allocation now used for models with high-dimensional inputs to avoid out-of-memory issues​

Changes

  • Added warnings when using sensitivity analysis methods that don’t support dependence for inputs with dependent inputs​

Bug fixes

  • Fixed LRA-based Sobol’ indices not working for multiple-output models
  • Fixed optimization bound issues when using Kriging in certain situations
  • Fixed inconsistent images used in the documentation w.r.t. the actual examples
  • Bugfixes and improvements across the board

:arrow_up_small: Back to top

UQLab Modules 1.2.1 (07.03.2019)

UQLabModules V1.2.0 ➜ UQLabModules V1.2.1

Stable release of UQLabModules.

Bug fixes

  • Addressed a number of compatibility issues with versions of MATLAB older than R2016a

:arrow_up_small: Back to top

UQLab Modules 1.2.0 (22.02.2019)

UQLabModules V1.1.0 ➜ UQLabModules V1.2.0

Stable release of UQLabModules.

New features

  • Bayesian inversion module:
    • A new module for solving Bayesian inverse problems is now available
      (developed and documented by @paulremo)​
  • Sensitivity analysis module:
    • Kucherenko and ANCOVA indices for global sensitivity analysis with dependent inputs are now available (developed and documented by @philippe)​
  • Polynomial chaos expansion module:​
    • Added adaptive q-norm truncation for the regression-based PCE​
    • Improved the leave-one-out calculation for the LARS regression method
  • UQLib​​:
    • A collection of general-purpose open-source libraries (including differentiation, optimization, kernel, and input/output processing) is now available and accessible in the lib folder
      (developed and documented by @moustapha, @c.lataniotis, @philippe, and @damarginal)​
      ​

Enhancements

  • Kriging, SVR, and SVC modules:
    • Evaluation of the kernel is now based on the general-purpose kernel evaluation function provided by UQLib (uq_eval_Kernel)​​​
  • Documentation:
    • Sensitivity analysis module:​
      • Added statements on each method whether the method is applicable for dependent input variables​
  • General:​​
    • The uq_gradient function is now vectorized and part of UQLib​​ differentiation library
    • Removed dependence from Optimization and Global Optimization toolboxes by defaulting to optimization algorithms available in UQLib

Changes

  • Documentation:
    • Sensitivity analysis module:​
      • One theory section for all Sobol’ indices​
      • New section on the usage chapter to showcase the sensitivity analysis methods that support dependent inputs (Kucherenko and ANCOVA indices)
  • Kriging:
    • Updated default optimization parameters to provide more accurate results

​​Bug fixes

  • UQLink:
    • UQLink can now handle cases where a command line is given using the full path to the executable that contains white spaces

:arrow_up_small: Back to top

UQLab Modules 1.1.0 (05.07.2018)

UQLabModules V1.0.0 ➜ UQLabModules V1.1.0

Stable release of UQLabModules.

New features

  • Metamodelling tools:
    • Support vector machines for classification (SVC) and regression (SVR) now available (developed and documented by @moustapha)
  • UQLink:
    • seamless connection of third-party software to UQLab now available by using universal “wrapping” of external codes through templates and a markup system (developed and documented by @moustapha)
  • Sensitivity analysis:
    • Borgonovo moment-independent indices are now available (developed and documented by C. Mylonas from ETH Zurich)
  • General:
    • New subsampling, one-hot-encoding and cobweb plot functions now available in the lib/ folder

Enhancements

  • General:
    • Standardized the examples for improved readability
  • Documentation:
    • Added the outputs of uq_print to all manuals
    • Added comments on the default values used in the minimal working examples
    • General readability and consistency improvements
  • Reliability analysis module:
    • AKMCS:
      • Added a convergence criterion on beta
    • IS:
      • One instrumental density function can now be specified for each model output
  • Sensitivity analysis module:
    • Removed the requirement for an input object for SRC / Correlation-based sensitivity analyses when a sample is provided

Changes

  • General:
    • Changes in uq_display for many modules to optimize readability
  • Metamodeling modules:
    • Polynomial chaos expansions:
      • Default degree for Quadrature set equal to 3, for degree-adaptive methods to 1:3
      • Fixed issues that broke the evaluation of a quadrature PCE for multiple outputs models
      • Initialization sets the maximum degree either from the provided degree or custom truncation. If both are provided, the custom truncation will be used.
      • Stability fixes for arbitrary polynomials (fix for integration waypoints)
    • Kriging:
      • Specification ExpDesign.Sampling = 'user' or 'data' is no longer necessary when providing the samples manually
      • Removed ExpDesign.time from results
      • Moved ExpDesign.muX and .sigmaX from Results to Internal

Bug fixes

  • Reliability analysis module:
    • SORM: can now be run on a pre-existing FORM analysis
    • IS: Removed warning in initialization if no instrumental density distribution is provided
  • Sensitivity analysis module:
    • Small stability fixes to sensitivity- and PCE- related calculations
    • Fixed the assembling of the PCE-based Sobol’ indices to avoid problems when using constant variables
    • Fixed LRA-based Sobol’ indices to prevent failing for models with multiple outputs
    • Sobol’ indices can be plotted as a pie-diagram

:arrow_up_small: Back to top

UQLab 1.0.0 (28.04.2017)

UQLabBeta V0.92 ➜ UQLabModules V1.0.0 stable

Stable release of UQLabModules.

New features

  • Metamodeling tool:
    • Canonical Low-Rank approximations now available (developed and documented by Dr. K. Konakli and C. Mylonas from ETH Zurich)
    • Polynomial-Chaos-Kriging now available (developed and documented by @rschoebi)
  • Open source release of the scientific code with extensive command-line help (UQLab Dev Team)

Enhancements

  • General:
    • Constant variables are now supported throughout UQLab modules. Most algorithms are now aware of constant variables and will exclude them to improve computational efficiency (UQLab Dev Team)
  • Input module:
    • Several input marginals added to the existing ones (E. Dodoula and @c.lataniotis)
  • Polynomial Chaos Expansions module:
    • Orthogonal Matching Pursuit added to the regression methods (M. Berchier)
    • Polynomials orthogonal to arbitrary distributions now available (C. Mylonas)
  • Reliability analysis module:
    • Polynomial Chaos-Kriging can now be used as a metamodel in AK-MCS
  • Documentation:
    • Now available in pdf and HTML formats in the Doc/Manuals folder,
      accessible via the uq_doc function

Changes

  • Kriging module:
    • Default correlation family changed to matern-5_2
    • Covariance matrix of the predictor is now available as the third output of uq_evalModel
  • Polynomial Chaos Expansions module:
    • Default quadrature scheme changed to Full when input dimension < 4 (cheaper)
  • Input module:
    • Changed handling of custom distributions

Bug fixes

  • General bug fixes and performance improvements across modules w.r.t. V0.92

:arrow_up_small: Back to top

1 Like