- Wraps Eigen3's eigsy function (C++) in Python
- Spherical Kmeans with kmeans++ init
This project uses the PEP517 pyproject.toml file, and should be compatible with the new python build standards.
It is recommended to use pip to install into the current environment:
# At the git root
pip install .pyproject.toml is configured to install a version of pybind11 in the compilation environnement, instead of pre-including it like before v0.3.
During tests, this failed when an outdated system install of pybind11 was present in addition to the temporary one ; maybe remove it if not used.
If the installation shows UNKNOWN as package name, it seems to be a similar problem but with an outdated system setuptools (which does not retrieve the metadata in pyproject.toml).
The easiest solution seems to be to use a virtualenv to isolate from system libs, and use updated tools (pip install --upgrade pip setuptools).
After an install in the current environment, the modules can be loaded in any python interpreter or script:
import apt.eigen_wrapper # For eigsy
import apt.kmeans # For kmeans
# Print module info, including list of functions
help(apt.eigen_wrapper)
help(apt.kmeans)
# Usage
import numpy
m = numpy.eye(10) # Id_10
(values, vectors) = apt.eigen_wrapper.eigsy(m)
# Kmeans
apt.kmeans.spherical_kmeans(points, nb_clusters, max_iter)Dependencies :
- pybind11 : automatically installed by pip if needed
- Eigen, v3.3.7 : header only, local copy