Milk

The most important change is the inclusion of eigen in the source distribution, which makes milk easier to compile. In addition, this release adds subspace projection k-nearest neighbours and mds_dists functionality.

Interfaces are more consistent (learners ignore arguments they cannot use and the default model supports the apply_many method). There are many improvements and bugfixes.

New features: parallel processing, perceptron, and error correcting output codes. Enhancements: setting the random seed in random forests, a 'multi_strategy' parameter for defaultlearner(), a return value from gridminimise, faster dot-kernel SVMs, and sigmoidal fitting. A bugfix in randomforest.

The new milk.ext.jugparallel module was added to interface with jug (http://luispedro.org/software/jug). This makes it easy to parallelize things such as n-fold cross validation (each fold runs on its own processor) or multiple kmeans random starts. Some new functions were added: measures.curves.precision_recall, milk.unsupervised.kmeans.select_best.kmeans. A tricky bug in SDA and a few minor issues elsewhere were fixed.

Many speed improvements. Some bugfixes (to gridminimize and tree learning). A few new utility functions.