Ex-Fuzzy

A Python library for explainable AI through fuzzy logic programming

Ex-Fuzzy is a modern, explainable fuzzy logic library for Python that I developed together with Prof. Javier Andreu-Perez. It provides comprehensive capabilities for explainable artificial intelligence through fuzzy logic programming, with a focus on accessibility and visualization.

Key Features

  • Explainable AI: Create interpretable models using fuzzy association rules that humans can understand and verify
  • Rich Visualizations: Built-in tools for visualizing fuzzy sets, membership functions, and rule bases
  • Scikit-learn Compatible: Follows the scikit-learn API for seamless integration with existing ML pipelines
  • High Performance: Optional GPU support for computationally intensive tasks
  • Comprehensive Support: Classification, regression, and rule mining capabilities
  • Multiple Fuzzy Set Types: Support for Type-1, Interval Type-2, and General Type-2 fuzzy sets

Installation

pip install ex-fuzzy

Quick Example

from ex_fuzzy import evolutionary_fit as gfs
from ex_fuzzy import utils

# Load your data
X_train, y_train, X_test, y_test = utils.load_data()

# Create and train a fuzzy classifier
fuzzy_clf = gfs.FuzzyRulesClassifier()
fuzzy_clf.fit(X_train, y_train)

# Get interpretable rules
print(fuzzy_clf.print_rules())

# Make predictions
predictions = fuzzy_clf.predict(X_test)
GitHub Repository
Documentation
PyPI Package

Papers Using Ex-Fuzzy

Core Library Paper

Ex-Fuzzy: A Library for Symbolic Explainable AI Through Fuzzy Logic Programming (Neurocomputing, 2024) Fumanal-Idocin & Andreu-Perez The foundational paper describing the library’s architecture, capabilities, and design philosophy. Introduces the scikit-learn compatible API and visualization tools.

New Methods Implemented in Ex-Fuzzy

A Fast Interpretable Fuzzy Tree Learner (arXiv, 2025) Fumanal-Idocin, Fernandez-Peralta & Andreu-Perez Introduces a novel fuzzy decision tree algorithm optimized for speed and interpretability. The implementation is available in Ex-Fuzzy as an alternative to rule-based classifiers.

Compact Rule-Based Classifier Learning via Gradient Descent (arXiv, 2025) Fumanal-Idocin, Fernandez-Peralta & Andreu-Perez Implements gradient-based optimization for learning compact fuzzy rule bases, extending Ex-Fuzzy beyond genetic algorithms to differentiable learning.

Reliable Classification with Conformal Learning and Interval-Type 2 Fuzzy Sets (FUZZ-IEEE, 2025) Fumanal-Idocin & Andreu-Perez Combines conformal prediction with Interval Type-2 fuzzy sets for uncertainty-aware classification. Extends Ex-Fuzzy’s IT2 capabilities with reliability guarantees.

Crisp Complexity of Fuzzy Classifiers (FUZZ-IEEE, 2025) Fernandez-Peralta, Fumanal-Idocin & Andreu-Perez Proposes new complexity metrics for fuzzy classifiers. Uses Ex-Fuzzy to empirically study the relationship between rule complexity and model performance.

Applications Using Ex-Fuzzy

ArtXAI: Explainable Artificial Intelligence Curates Deep Representation Learning for Artistic Images Using Fuzzy Techniques (IEEE Trans. Fuzzy Systems, 2023) Fumanal-Idocin, Andreu-Perez, Cordón, Hagras & Bustince Uses Ex-Fuzzy to provide explainable classification of artistic images by learning fuzzy rules over deep learning embeddings. Demonstrates how fuzzy rules can interpret neural network representations.

Interpreting Contrastive Embeddings in Specific Domains with Fuzzy Rules (FUZZ-IEEE, 2024) Fumanal-Idocin, Jamalifard & Andreu-Perez Applies Ex-Fuzzy to interpret contrastive learning embeddings (e.g., CLIP) using domain-specific fuzzy rules, bridging black-box representations with human-understandable concepts.

Comparison and Benchmarking

Efficient Online Generation of Fuzzy Measures via Aggregation Functions (Information Fusion, 2026) Gonzalez-Garcia, Horanská, Beliakov & Bustince Uses Ex-Fuzzy as a baseline for comparing fuzzy measure generation methods, demonstrating the library’s utility as a benchmarking tool in the fuzzy systems community.

News and Updates

  • 2025: New fuzzy tree learner and gradient-based optimization methods added
  • 2025: Three papers accepted at FUZZ-IEEE 2025 using Ex-Fuzzy
  • 2024: Ex-Fuzzy 2.0 released with improved performance and new features
  • 2024: Core library paper published in Neurocomputing
  • 2023: ArtXAI paper published in IEEE Trans. Fuzzy Systems
  • 2023: Initial public release on PyPI

Citation

If you use Ex-Fuzzy in your research, please cite:

@article{FUMANALIDOCIN2024128048,
  title = {Ex-Fuzzy: A library for symbolic explainable AI through fuzzy logic programming},
  journal = {Neurocomputing},
  pages = {128048},
  year = {2024},
  author = {Javier Fumanal-Idocin and Javier Andreu-Perez},
  doi = {https://doi.org/10.1016/j.neucom.2024.128048}
}

Contributing

Ex-Fuzzy is open source under the AGPL v3 license. Contributions, bug reports, and feature requests are welcome on GitHub.

If you are a BSc/MSc student looking for thesis projects, I am actively supervising work on extending Ex-Fuzzy functionalities. Feel free to contact me.