hdbscan

Boruvka scales poorly with dimension; by approximating the minimal spanning tree we can achieve far better scaling with dimension at the cost of slightly less accurate clustering. In testing the loss of accuracy seems very small, so I'm pushing it out with that defaulted to on.

Fix bugs introduced when working around numpy performance regression.

Setting upper condo building

Fix issues with 32 bit Windows installs
Fix issues caused by a performance regression in bumpy structured array accesses

Better performance
Fixes for Robust Single Linkage
Faster cluster extraction from trees
New Notebooks

Further optimization of routines that became bottlenecks due to the new faster minimum spanning tree algorithm. Reworking the calling of query on trees to optimize core distance computation in higher dimensional spaces.

Bug fixes for pip installing 0.4

New algorithms bring dramatic speed improvements, particularly for low dimensional data.

Support low memory; Robust single linkage.