High dimensional data clustering with COPAC

We implement COPAC (Correlation Partition Clustering), which

1. computes the local correlation dimensionality based on the largest eigenvalues
2. partitions the data set based on this dimension
3. calculates a Euclidean distance variant weighted with the correlation dimension, called correlation distance
4. further clusters objects within each partition with Generalized DBSCAN, requiring a minimum number of objects to be within eps range for each core point.

Installation

Clone this repository and pip install like so:

git clone https://github.com/VarIr/copac.git
cd copac
python3 -m pip install .

COPAC is then available through the copac package.

Example

COPAC usage follows scikit-learn's cluster API.

from copac import COPAC
# load some X here ...
copac = COPAC(k=10, mu=5, eps=.5, alpha=.85)
y_pred = copac.fit_predict(X)

Implementation

Published in GitHub: https://github.com/VarIr/copac

Citation

The original publication of COPAC.

@article{Achtert2007,
         author = {Achtert, E and Bohm, C and Kriegel, H P and Kroger, P and Zimek, A},
         title = {{Robust, Complete, and Efficient Correlation Clustering}},
         journal = {Proceedings of the Seventh Siam International Conference on Data Mining},
         year = {2007},
         pages = {413--418}
}

License

This work is free open source software licensed under GPLv3.