Peppermint Patty

• Adding Umap to minHashClustering
• Adding some test cases
• make azure testing available
• enable macOS (x86) again

Linus van Pelt

• Add modulo to support hashing of values > 2^32
• add structures for a potential AVX support. However, a few operations require AVX512 which is implemented on a minority of CPUs
• Fixing a bug that occured if less than four features of an instances were present
• Adding the option for minHashClustering.py to apply an additional PCA on the computed k-nn
• Dropping cuda support in setup.py, restructuring setup.py. There is some bug if I want to compile it, however, I don't know how to fix this atm.

Snoopy

This release fixes the partial fitting. It is now possible to transfer only parts of the matrix at once and save with this memory. Users can now define large shares of rows of the raw matrix shall be transferred to C++ at a time. This only applies for the clustering methods, for the standalone MinHash the user needs to loop over the data.

Lucy van Pelt

• Update to Python 3 and drop of Python 2 support.
• Update to Cuda compile level sm_60
• Reduced memory usage for transfer of data Python -> C++

Charlie Brown

Support for SSE4.1. This version improves the speed of the algorithm. It can be used if and only if your cpu supports SSE4.1 which should be the case for an Intel cpu of the generation 'Penryn' from 2007/2008 or newer. For an AMD cpu it is the generation 'Bulldozer', 'Jaguar' and 'Piledriver' or newer.

Gilbert

Release of the approximate nearest neighbor search algorithms 'MinHash' and 'WtaHash' for sparse datasets.

Publication release

The current release is for submitting a publication. The parts with known bugs are deactivated. Examples and utils are removed.

Masterthesis level

This release is to preserve the development level of minHash as it was at the hand in date of my master thesis.