SparseCluster

sparsecluster is a python package to perform hierarchical agglomerative clustering on approximate nearest neighbor graphs. it's basically a wrapper around pynndescent and gbbs

installation

clone this repo and load all associated submodules. installing python bindings for gbbs can be a bit tricky, below is a hacky workaround. This will probably be resolved in the separate [ParHAC](https://github.com/ParAlg/ParHAC) repo.

git clone git@github.com:bobermayer/sparsecluster.git
cd sparsecluster
git submodule update --init
cd gbbs
git submodule update --init
bazel build //...
bazel build //pybindings:gbbs_lib.so
cd ..
pip install .
cp gbbs/bazel-bin/pybindings/gbbs_lib.so /path/to/python/site-packages/gbbs_lib.cpython-??-x86_64-linux-gnu.so

usage

call signature is essentially similar to fastcluster:

Z = sparsecluster.linkage(X, metric='euclidean', method='single', n_neighbors=50, n_backup=10, n_jobs=2, ...)

where X is a (possibly sparse) matrix of dimensions (n_samples, n_features).

dependencies

• numpy
• scipy
• scikit-learn >= 0.22
• numba >= 0.51
• bazel
• libprotobuf

all of which should be pip or conda installable.

in addition, a custom fork of pynndescent is required:

git clone git@github.com:bobermayer/pynndescentSC.git
cd pynndescentSC
pip install .