October 4, 2013
MACROS 1.11.0 Released
DATADVANCE development team is pleased to announce the next major release of MACROS, version 1.11. This release finalizes the updates done since the previous major version 1.10 and will serve as the development basis in MACROS 1.11 series.
Compared to MACROS 1.10, the most important improvements and updates are:
- The Mixture of Approximators tool previously found in MACROS for Python Extras has finally moved to GTApprox and is now directly available from the GTApprox interface as one of the approximation techniques.
- An additional statistical utilities module has been added. This module implements common statistical methods and currently allows to calculate various elementary statistics, detect outliers in data samples and perform some distribution tests.
- User blackbox classes were redesigned to make the concept more clear and easier to use. This update also makes the support for analytical gradients common for all blackboxes and adds methods to work with the blackbox evaluation history. Note that new blackboxes are not compatible with previous versions; more details on compatibility are found in the MACROS for Python documentation.
- Also note that MACROS for Python now requires NumPy 1.6.1 or newer to run.
- Generic Tool for Approximation (GTApprox):
- Added the Mixture of Approximators (MoA) technique (see above).
- Implemented the support for Gaussian processes with additive kernel. Adds a new covariance function type to the Gaussian Processes (GP) modeling technique and results in improved GP models quality in high-dimensional cases or when the functional dependence contains interaction terms.
- Lowered the training sample size requirements after reviewing the GTApprox techniques.
- Generic Tool for Data Fusion (GTDF):
- Added the low-fidelity sample bias compensation feature. This optional algorithm will benefit the cases when the function generating the low-fidelity training sample is actually a biased high-fidelity sample generation function.
- Added the ability to manually select the approximation algorithm when using the High Fidelity Approximation (HFA) technique.
- Generic Tool for Design of Experiments (GTDoE):
- New techniques added: parametric study and Box-Behnken design generation. Both are also available for generating an initial sample when using adaptive DoE.
- Generic Tool for Important Variable Extraction (GTIVE):
- The SMBFAST technique was improved, also now it may be selected automatically if the input data satisfies the technique requirements.
- More robust score variance calculation and new options giving user the control over variance estimation.
- Better handling of constant blackbox inputs in blackbox-based techniques.
- Generic Tool for Optimization (GTOpt):
- Added the support for multi-objective Surrogate Based Optimization. Earlier, GTOpt allowed to apply surrogate model-based methods to only one objective even in multi-objective problems. Now any number of problem objectives and/or constraints may be defined "expensive" (taking a long time to evaluate). When solving the problem, GTOpt internally and automatically replaces such functions with fast surrogate models, dramatically reducing the time to solve.
- Improved global optimum search functionality. GTOpt now allows to control the complexity of applied global methods. As a result, relatively simple global optimization problems (where, for example, objective functions are nearly unimodal) may be solved significantly faster by selecting global algorithms of a lesser complexity which should suffice in such case. On the other hand, most complex and time-consuming global algorithms are as well available for solving hard multimodal cases.
Important bugfixes include resolving the rare cases of incorrect behavior in GTApprox (the Mixture of Approximators technique), GTDoE (adaptive DoE generator) and GTIVE (the FAST blackbox-based technique), fixing the compatibility with old GP-based models in GTApprox, and a fix for an internal GTOpt bug which could cause a fatal error if the GTOpt/TimeLimit option is set.
The full details on version 1.10 update, including all changes and bugfixes, are found in the release 1.11.0 summary changelog in the MACROS for Python documentation.
Please contact us to receive more information and MACROS updates!