November 20, 2015
MACROS 6.3 Release
DATADVANCE development team is pleased to announce the release of MACROS 6.3, a new stable version of the MACROS library and featured plug-ins.
This release contains several updates to GT Approx and a new kind of adaptive DoE generation method:
- New piecewise-linear approximation technique (PLA) added to GT Approx. PLA is a simple and fast technique based on Delaunay triangulation which is suitable for low-dimensional models that do not require smooth approximation.
- GT Approx can now be configured to tolerate deviations of sample inputs from a grid-like DoE by applying controllable input rounding at the sample preprocessing step. Such preprocessing allows using noisy samples with tensor approximation techniques that require "almost factorial" DoE.
- The support for categorical variables has been added to more approximation techniques, including Gaussian processes-based techniques and incomplete tensor approximation.
- GT DoE can now perform adaptive DoE generation based on a user-defined grid. This method produces an optimal DoE sample that includes only the points that are found on the grid.
This release also enables componentwise model training by default, since this mode works better for multidimensional models with independent outputs. GTApprox continues to support models with dependent outputs, but this approximation mode is now optional.
MACROS 6.3 also fixes a few bugs discovered since the previous release. For full details on recent updates and bugfixes, you can refer to the latest changelog, or contact us to receive more information and MACROS updates!