Approximate nearest neighbor: Faster and Better
ann-fab is an approximate nearest neighbor search built around the NearPy (http://nearpy.io/) framework. It was started as a project for Zalando's HackWeek #4 in December 2015.
One of the primary motivators for its development was finding (near) duplicates in large datasets. In this case, a large number of hashes are used, and the matrix-vector product contributes significantly to the computational requirement.
The extensions to the NearPy framwork aim to address this, but allowing for parallelisation and also GPU acceleration.
To ensure that the C++ backend is built, run:
python setup.py build_lib
This should build and copy the file _annfab.so
to the annfab
module folder.
python setup.py tests