Karate Club

Karate Club is an open-source Python framework with 2,200+ stars on GitHub for unsupervised learning on graphs — a single, consistent API for network embedding and community detection methods, published at CIKM 2020.

Overview

Graph-structured data — social networks, citation graphs, molecular structures — does not fit standard tabular machine learning, and research implementations of graph algorithms are usually scattered, inconsistent and hard to compare. Karate Club, built as an extension library for NetworkX, collects state-of-the-art unsupervised methods behind one scikit-learn-style API. It provides network embedding techniques at both the node and the whole-graph level, including DeepWalk, Node2Vec and Graph2Vec-style methods, plus a variety of overlapping and non-overlapping community detection algorithms such as Louvain and label propagation, and graph clustering and classification building blocks. Every method follows the same fit-and-extract pattern, so swapping algorithms in an experiment takes a line of code. Last updated in 2024, it remains a widely cited Swiss Army knife for small-scale graph mining research and teaching. Its Read the Docs site carries installation guides, example notebooks and a complete API reference for every implemented method.

Key Features

• Node embeddings – DeepWalk, Node2Vec and related random-walk embedding methods.
• Graph embeddings – whole-graph representations such as Graph2Vec for graph classification.
• Community detection – overlapping and non-overlapping methods including Louvain and label propagation.
• Unified API – consistent, scikit-learn-style interface across dozens of algorithms.
• NetworkX integration – works directly with the standard Python network analysis library.
• Research grade – implements the methods from the CIKM 2020 paper with documentation and examples.

💳 Pricing

Karate Club is completely free and open source under the GPL-3.0 license. Install it from PyPI and run it locally; there is no hosted version, paid tier or usage fee of any kind.

Who Is It For

Researchers, data scientists and students working with network data who need reliable implementations of graph embedding and community detection algorithms without reimplementing papers themselves.