June 20, 2014
MACROS 3.0 Release
DATADVANCE development team is pleased to announce the release of MACROS 3.0 Final, a new stable version of the MACROS library and featured plug-ins.
This release brings many upgrades and new features compared to the previous major release, MACROS 2.0. In version 3.0 you will find the improved Surrogate Based Optimization, new Tensor Gaussian Processes approximation technique and updated model validation, dimension reduction model export, the support for fractional factorial design of experiments, faster input/output data processing and better NumPy support.
Surrogate Based Optimization method (SBO):
- Supports mixed integer optimization problems, allowing to solve single- and multi-objective problems with a mix of integer and continuous variables.
- Improved quality of internal approximation models used in SBO makes it even more efficient in terms of reducing the required number of objective and constraint function evaluations.
- Increased multi-objective SBO method robustness in case of badly scaled search space.
We have also added an option to extend GTOpt's result data set so it includes additional optimal solutions that violate problem constraints to a certain (controllable) extent. Previously, such data was not included in the result and could be found only by manually searching the problem evaluation history added in MACROS 1.11.0. This feature already proved useful in optimization tasks where there is a possibility of incorrectly specified constraints due to a model or human error.
Moreover, GTOpt is now able to analyze behavior of linear and quadratic objectives and constraints and restore analytical functions so they are evaluated internally by solver. This feature potentially increases optimization speed since it allows less communication between the solver and the optimized model. It is available in all GTOpt modes and is enabled by default.
Tensor Gaussian Processes approximation technique (TGP):
- TGP is a further development of the methods first introduced in the Tensor Approximation (TA) technique added in MACROS 1.7.
- This technique modifies the Gaussian Processes (GP) algorithm for big data sets, provided they were obtained using a Cartesian product DoE.
- A distinctive feature of the TGP technique is that it provides model accuracy evaluation support not available in the original TA or incomplete Tensor Approximation (iTA) techniques.
Major GTApprox updates also include:
- Improved internal validation with the support for the Mixture of Approximators (MoA) technique added in MACROS 1.10.1, and allows to save model values calculated during the validation process.
- Ability to use incomplete noise variance data when applying the output noise variance feature (implemented in MACROS 1.8) to treat a noisy training sample.
- Increased Gaussian Processes technique robustness in certain special cases.
GTDR dimension reduction models now support the full range of export methods, making them available in Octave, MEX and C.
MACROS DoE technique selection has been extended with yet another classical DoE, the 2-level fractional factorial. The new Fractional Factorial technique also supports discrete variables added in MACROS 1.9.1.
Finally, we have further improved MACROS support for NumPy, the fundamental package for scientific computing with Python. MACROS for Python interfaces now use NumPy arrays wherever possible. When communicating with the high-performance C++ MACROS core, NumPy interface is noticeably faster, making it especially efficient in iterative processes such as optimization or adaptive DoE.
Since this is an interface change, it may require certain updates for your code if it uses GTOpt's advanced optimization problem class or blackbox-based techniques. You can find more information on required updates, as well as details on new features in bugfixes, in the MACROS 3.0 Final release changelog.
Please contact us to receive more information and MACROS updates!