mlpack 1.0.11
·
17724 commits
to master
since this release
Released December 11th, 2014.
- Proper handling of dimension calculation in PCA.
- Load parameter vectors properly for LinearRegression models.
- Linker fixes for AugLagrangian specializations under Visual Studio.
- Add support for observation weights to LinearRegression.
- MahalanobisDistance<> now takes root of the distance by default and therefore satisfies the triangle inequality (TakeRoot now defaults to true).
- Better handling of optional Armadillo HDF5 dependency.
- Fixes for numerous intermittent test failures.
- math::RandomSeed() now sets the seed for recent (>= 3.930) Armadillo versions.
- Handle Newton method convergence better for SparseCoding::OptimizeDictionary() and make maximum iterations a parameter.
- Known bug: CosineTree construction may fail in some cases on i386 systems (376).