v0.8.1

The biggest change introduced by this release moves the responsibility of filtering points from the KdTree to the search visitor. Previously, the KdTree would use a visitor's max() method to filter both the nodes of the tree as well as the points residing in the non-filtered nodes. Moving this responsibility makes it possible for the visitor to inspect all the points of the non-filtered nodes instead of just a sub-set, giving the visitor more control and information during a query. This allows, for example, the implementation of the following features:

• An approximate radius search. This was impossible with an older version (equal to setting a smaller search radius).
• Knowing the exact amount of points encountered during a query (see kd_tree_custom_search_visitor.cpp).

Filtering in the visitor also improved the existing approximate k nearest neighbor query. Of all the points that are visited during a query, it will now be guaranteed to return k the nearest.

WARNING! To those who implemented a custom search visitor: Make sure to add extra filtering to avoid having the best nearest neighbor candidate being overwritten by a worse candidate!

Other changes:

• The LInf metric has been added for both C++ and Python.
• Updated the search_box interface for Python to obtain a small performance improvement.
• Removed an unsupported SpaceMap constructor.
• Interfacing with an Eigen::Map of const Eigen::Matrix added. There was only support for non-const matrices.
• Added the approximate version of SearchNn (C++), SearchRadius (C++), and search_radius (Python).
• Updated various examples and added KdTree serialization and MNIST examples.