v0.6.1

• Adds support for non-x86 CPUs, including ARM-based Macs
• Removes accommodations for pre-C++11 compilers
• Formally ends support for Python 2.7

v0.6

• Adds vectorized, multi-threaded calculation of many shortest path routes at once
• Restores usability of network.plot() by eliminating usage of Matplotlib's deprecated Basemap toolkit

v0.5.1

• Fixes a performance bug in network.get_node_ids()

v0.5

• Adds support for calculating shortest path lengths between arbitrary origins and destinations, with vectorization and multi-threading
• Restores alternate names for aggregation types, which were inadvertently removed in v0.4
• Fixes a bug with matplotlib backends
• Improves compilation in MacOS 10.15 Catalina
• Eliminates the scikit-learn dependency
• Makes matplotlib and osmnet dependencies optional
• Revises the documentation and demo notebook

v0.4.4

This was released on 2019-09-04, but was not tagged in GitHub at that time

• Restores support for pre-C++11 compilers

v0.4.3

• Improved compiler support

v0.4.2

• Speed of network aggregations is improved
• Support for aggregating integer values is restored
• Thread count and contraction hierarchy status messages are restored
• All code written for v0.3 now runs in v0.4, raising deprecation warnings when appropriate
• Compilation improvements for Mac

Improvements for pre-C++11 compilers

• Documentation fixes.
• Replaced uses of std::map::at() since it's not supported in pre-C++11 compilers.
• Replaced initialization lists due to the same reason as above.

Rewrite of the layer between Python and C++ to remove global memory constraints.

• Major rewrite of the layer between Python and C++, which was previously written using the numpy C++ API, and now is written in cython.
• The C++ that remains has been cleaned up a bit and formatted.
• The major functionality change is that global memory is no longer used, so reserve_num_graphs no longer needs to be called and Network objects can be created and destroyed at the user's pleasure.
• The change in global memory made the calls to init_pois no longer necessary. Then, that method has been removed and the max_items and max_distance parameters were relocated in the set_pois call.
• The nearest neighbor queries are now resolved with Scipy instead of libANN. That removed additional global memory.

Python3 support and general improvements

• Python 3 compatibility.
• The “network.nearest_pois()” method can now return the labels of the pois rather than just the distances
• OSM data loading is now done via the osmnet package.
• Changes to support multiple graphs.
• Added reindex functions.
• Updated documentation.
• Switched code style checker in Travis CI to “pycodestyle”, which has replaced the “pep8” package.