This changelog follows the format defined at: https://keepachangelog.com/en/1.0.0/
Adjusted Saltelli sampling to follow recommendation of Owen (2020) (http://arxiv.org/abs/2008.08051; https://github.com/scipy/scipy/pull/10844#issuecomment-672186615)
Initial support for parallel analysis
Updated Sobol’ G-function analytic results (PR #464, Issues #335 #461)
Sobol’ analysis: Optional storage of intermediate resample results to allow analysis of variation (PR #462)
Updated Salteli sampling examples to use powers of 2 following recommendations
Upgrade PyScaffold to v4
Replaced recommonmark with MyST (PR #466)
Shortlist of changes since v1.3.x
High Dimensional Model Representation (HDMR) method (PR #275)
PAWN method (PR #415)
Support for sampling/analysis method chaining (PR #339)
Support for truncated normal distribution (PR #383)
Confidence Interval estimation for FAST-based methods (PR #375)
Initial support for parallel model evaluation
Defining non-uniform sampling now explicitly documented
Many general documentation improvements
Generalized support for non-uniform sampling methods (PR #346)
Many documentation improvements
Explicitly mention extended FAST in documentation
Saltelli sampling: Warnings displayed when selected samples do not meet requirements (PR #416).
Group sampling and analysis enabled for Sobol’ and morris
Enhanced DataFrame support for groups
Conversion to DataFrame when groups are defined with Sobol’ results (PR #413 and Issue #387)
Various minor performance enhancements (PR #253 #264)
Added some visualisation methods (PR #259)
Tidying up of the Command Line Interfance, and num samples (PR #260 #291)
Improved efficiency of summing distances in local optimization (PR #246)
Revamped fast method for consistency (PR #239)
Updated Sobol-G function to modified G-function (#269)
Method of morris didn’t adjust with levels above 4 (PR #252)
Add missing seed option for morris sampling
Handle singular value matrix cases (PR #251)
Fixed typo (#205)
Updated import of scipy comb function (PR #243)
Update documentation for Morris and DSGM methods (#261 #266)
Updated PyScaffold to version 3.2.2 (#267)
Updated Travis and package config (#285)
Refactored Method of Morris so the Ruano et al. local approach is default
Inputs to morris.analyze are provided as floats
Removed calls to standard random library as inconsistent between Python 2 & 3
First row in Sobol sequences should be zero, not empty
Added a Code of Conduct
Added DAETools, BCMD and others to citations - thanks for using SALib!
Removed misleading keyword arguments in docs and readme examples
Updated documentation for Method of Morris following refactor
Improved existing documentation where lacking e.g. for fractional factorial method
Implemented automatic deployment to PyPi
Fixed a bug preventing automatic deployment to PyPi upon tagging a branch
Removed postgres from travis config
Release of our stable version of SALIB to coincide with an submission to JOSS:
Added a paper for submission to the Journal of Open-source Software
Updated back-end for documentation on read-the-docs
Updated the back-end for version introspection using PyScaffold, rather than versioneer
Updated the Travis-CI scripts
Moved the tests out of the SALib package and migrated to using pytest
Improvements to Morris sampling and Sobol groups/distributions
Adds improved sampling for the Morris method (thanks to @JoerivanEngelen) and group sampling/analysis for the Sobol method (thanks to @calvinwhealton).
@calvinwhealton has also added non-uniform distributions to the Sobol sampling. This will be a baseline for adding this to the other methods in the future.
Also includes several minor bug fixes.
New documentation, doc strings and installation requirements
@dhadka has kindly contributed a wealth of documentation to the project, including doc strings in every module
no longer test for numpy <1.8.0 and matplotlib < 1.4.3, and these requirements are implemented in a new setup script.
Parallel option for Sobol method
New option to run analyze.sobol function in parallel using multiprocessing
This release does not contain any new functionality, but SALib now is citable using a Digital Object Identifier (DOI), which can be found in the readme.
Some minor updates are included:
morris: sigma has been removed from the grouped-morris results and plots, replaced by mu_star_conf - a bootstrapped confidence interval. Mu_star_conf is not equivalent to sigma when used in the non-grouped method of morris, but its all we have when using groups.
some minor updates to the tests in the plotting module
Set up to include and test plotting functions
Specific plotting functions for Morris
Fractional Factorial SA from Saltelli et al.
Repo transferred to SALib organization, update setup and URLs
Vectorized bootstrap calculations for Morris and Sobol
Optional trajectory optimization with Gurobi, and tests for it
Several minor bugfixes
Starting with v0.5, SALib is released under the MIT license.
Better Python API without requiring file read/write to the OS. Consistent functional API to sampling methods so that they return numpy matrices. Analysis methods now accept numpy matrices instead of data file names. This does not change the CLI at all, but makes it much easier to use from native Python.
Also expanded tests for regression and the Sobol method.
Improvements to Morris sampling and analysis methods, some bugfixes to make consistent with previous versions of the methods.
Improvements to Morris sampling methods (support for group sampling, and optimized trajectories). Much better test coverage, and fixed Python 3 compatibility.
First numbered release. Contains reasonably well-tested versions of the Sobol, Morris, and FAST methods. Also contains newer additions of DGSM and delta methods which are not as well-tested yet. Contains setup.py for installation.