Description:

libcmaes is a multithreaded C++11 (with Python bindings) implementation of the CMA-ES algorithm for stochastic optimization of nonlinear 'blackbox' functions. The implemented algorithms have a wide range of applications in various disciplines, ranging from pure function optimization, optimization in industrial and scientific applications, to the solving of reinforcement and machine learning problems.

Current features include: high-level API for simple use in external applications, implementatio of several flavors of CMA-ES, IPOP-CMA-ES, BIPOP-CMA-ES, active CMA-ES, active IPOP and BIPOP restart strategies, sep-CMA-ES and VD-CMA (linear time & space complexity) along with support for IPOP and BIPOP flavors as well.

Some operations benefit from multicores, and there's support for objective function gradient, when available. A control exe in the command line is provided for running the algorithm over a range of classical single-objective optimization problems.

Full documentation is available from https://github.com/beniz/libcmaes/wiki

Developer API documentation is available from http://beniz.github.io/libcmaes/doc/html/index.html

Changes to previous version:

This is a major release, with several novelties, improvements and fixes, among which:

• step-size two-point adaptaion scheme for improved performances in some settings, ref #88

• important bug fixes to the ACM surrogate scheme, ref #57, #106

• simple high-level workflow under Python, ref #116

• improved performances in high dimensions, ref #97

• improved profile likelihood and contour computations, including under geno/pheno transforms, ref #30, #31, #48

• elitist mechanism for forcing best solutions during evolution, ref 103

• new legacy plotting function, ref #110

• optional initial function value, ref #100

• improved C++ API, ref #89

• Python bindings support with Anaconda, ref #111

• configure script now tries to detect numpy when building Python bindings, ref #113

• Python bindings now have embedded documentation, ref #114

• support for Travis continuous integration, ref #122

• lower resolution random seed initialization

BibTeX Entry: Download

Supported Operating Systems: Linux, Windows, Mac Os X

Data Formats: Any

Tags: Black Box Optimization, Evolution Strategy, Stochastic Optimization

Archive: download here

Other available revisons

Version	Changelog	Date
0.9.5	This is a major release, with several novelties, improvements and fixes, among which: step-size two-point adaptaion scheme for improved performances in some settings, ref #88 important bug fixes to the ACM surrogate scheme, ref #57, #106 simple high-level workflow under Python, ref #116 improved performances in high dimensions, ref #97 improved profile likelihood and contour computations, including under geno/pheno transforms, ref #30, #31, #48 elitist mechanism for forcing best solutions during evolution, ref 103 new legacy plotting function, ref #110 optional initial function value, ref #100 improved C++ API, ref #89 Python bindings support with Anaconda, ref #111 configure script now tries to detect numpy when building Python bindings, ref #113 Python bindings now have embedded documentation, ref #114 support for Travis continuous integration, ref #122 lower resolution random seed initialization	March 9, 2015, 09:05:22
0.9.4	Update works around clang bug (e.g. for OSX) and implements uncertainty handling scheme. Main changes: work around clang bug, now working with clang, ref #19 easier build on OSX added uncertainty handling scheme for noisy objective functions, ref #65 optional support for surrogates at compile time, reducing the overal lib size, ref #90 fixed uninstall of python bindings	January 8, 2015, 11:09:02
0.9.3	This is an important update: full support for surrogates, allowing optimization of costly objective functions, ref #57 integrated rankign SVM default surrogate, ref #83 Python bindings for surrogates, ref #75 more informed optimization status and error messages, ref #85 API for computing confidence intervals around optima, ref #30 API for computing 2D contour around optima, ref #31 new 'elitist' scheme for improved restart strategy useful on some rather difficult functions, ref #77 fixed Eigen namespace import, ref #62 fixed and added new parameter vector getter in Candidate, ref #84	November 17, 2014, 14:04:10
0.9.2	Main changes: new VD-CMA algorithm with linear time and space complexity for black-box optimization API control of stopping criteria, with individual activation scheme improved memory control when tackling large-scale optimization problems solutions now support printing out in pheno space improved API of solutions object fixed compilation error with gcc 4.7	October 30, 2014, 14:08:34
0.9.1	Small release with two bug fixes and tiny changes otherwise: small API improvements fixed bug in tolX stopping criteria when using 'sep' algorithm fixed bug to the natural gradient with genotype /phenotype transforms file stream now outputs parameter's mean in phenotype tiny wrapper to simplify maximization of objective function (default is minimization)	October 9, 2014, 10:08:18
0.9.0	Python bindings, ref #26 Cleaned up setters / getters interface, ref #64 Lib is now quiet by default, ref #61 Support for pkg-config, ref #58 Improved make uninstall, ref #66 API improvements (e.g. new parameters constructor from vector, ref #60) Stopping criteria with explicit control of in-memory history size for large-scale optimization	September 10, 2014, 10:13:53
0.8.1	Added customization of data to file streaming function, ref #51 Added configure control for compiling the library alone without examples or tools, ref #11 Fixed code in order to avoid various compiler warnings Fixed sample code in README, ref #54 Fixed get_max_iter and set_mt_feval in Parameters object New CMAParameters constructor, from x0 as a vector of double Updated building instructions for Mac OSX New set_str_algo in Parameters object	August 12, 2014, 16:18:31
0.8	Initial Announcement on mloss.org.	July 15, 2014, 11:20:02

Comments

No one has posted any comments yet. Perhaps you'd like to be the first?

Leave a comment

You must be logged in to post comments.