/fsfc

Feature Selection for Clustering

Primary LanguagePythonMIT LicenseMIT

Feature Selection for Clustering

mit Documentation Status

FSFC is a library with algorithms of feature selection for clustering.

It's based on the article "Feature Selection for Clustering: A Review." by S. Alelyani, J. Tang and H. Liu

Algorithms are covered with tests that check their correctness and compute some clustering metrics. For testing we use open datasets:

Project documentation is available on Read the Docs

Implemented algorithms:

  • Generic Data:
    • SPEC family - NormalizedCut, ArbitraryClustering, FixedClustering
    • Sparse clustering - Lasso
    • Localised feature selection - LFSBSS algorithm
    • Multi-Cluster Feature Selection
    • Weighted K-means
  • Text Data:
    • Text clustering - Chi-R algorithm, Feature Set-Based Clustering (FTC)
    • Frequent itemset extraction - Apriori

Dependencies:

  • numpy
  • scikit-learn
  • scipy

How to use:

Now the project is in the early alpha stage, so it isn't publish to pip.

Because of it, installation of the project is a bit complicated. To use FSFC you should:

  1. Clone repository to your computer.
  2. Run make init to install dependencies.
  3. Copy content of the folder fsfc to the source root of your project.

After it you can use feature selectors as follows:

import numpy as np
from fsfc.generic import NormalizedCut
from sklearn.pipeline import Pipeline
from sklearn.cluster import KMeans

data = np.array([...])

pipeline = Pipeline([
    ('select', NormalizedCut(3)),
    ('cluster', KMeans())
])
pipeline.fit_predict(data)

How to support:

You can support development by testing and reporting of bugs or opening pull-requests.

Project has tests, they can be run with the command make test

Also code there is a Sphinx documentation for code, it can be built with the command make html. Documentation uses numpydoc, so it should be installed on the system. To do it, run pip install numpydoc.

References: