FuzzyCocoPython documentation

FuzzyCocoPython

Python bindings and scikit-learn style estimators for fuzzycoco, an evolutionary fuzzy rule learning engine written in C++. This package wraps the C++ core as a Python module and exposes FuzzyCocoClassifier and FuzzyCocoRegressor with a familiar fit/predict API.

Features

Train fuzzy rule-based classifiers and regressors using fit, predict, and score
Inspect learned linguistic variables, rules, and activation statistics from Python
Persist trained estimators with save/load helpers based on joblib

Installation

This package is not yet on PyPI; install it from source. Make sure the following prerequisites are available:

A C++17 compiler toolchain (GCC/Clang on Linux & macOS, MSVC on Windows)
CMake ≥ 3.21 and Ninja on your PATH
uv is recommended for dependency management, but pip works just as well

Clone the repository, initialise the fuzzycoco submodule, then build and install in editable mode with uv:

git clone https://github.com/Lonza-RND-Data-Science/fuzzycocopython.git
cd fuzzycocopython
git submodule update --init --recursive

uv venv
source .venv/bin/activate
uv pip install -e .

If you prefer the standard tooling, create a virtual environment manually and install with pip:

python -m venv .venv
source .venv/bin/activate

pip install -e .

When you only need the published bindings and not the repository, install straight from GitHub:

pip install git+https://github.com/Lonza-RND-Data-Science/fuzzycocopython.git

For development tasks (tests, linting, docs) install the optional toolchain:

uv pip install -e '.[dev]'

The build compiles the bundled C++ bindings. Refer to the fuzzycoco project for background on the engine itself.

Quick start

import pandas as pd
from sklearn.datasets import load_iris
from fuzzycocopython import FuzzyCocoClassifier

data = load_iris(as_frame=True)
clf = FuzzyCocoClassifier(random_state=0)
clf.fit(data.data, data.target)

preds = clf.predict(data.data)
score = clf.score(data.data, data.target)

print(f"Accuracy: {score:.3f}")
print(clf.rules_df_.head())

The estimators now expose scikit-learn style hyper-parameters for direct tuning:

from fuzzycocopython import FuzzyCocoClassifier

model = FuzzyCocoClassifier(nb_rules=10, nb_sets_in=3, random_state=42)
model.fit(X, y)
preds = model.predict(X)

Bit widths for variable and set indices are automatically derived from the training data. Override them with nb_bits_vars_in, nb_bits_sets_in, nb_bits_vars_out, or nb_bits_sets_out when you need full manual control.

For a guided tour, open demo.ipynb. Additional usage examples live in tests/test_fuzzycocopython.py.

Documentation

Full API documentation is available at Lonza-RND-Data-Science.github.io/fuzzycocopython. To build the docs locally run:

uv pip install -e '.[docs]'
uv run sphinx-build -W -b html docs docs/_build/html

Pre-commit hooks

Install and activate the provided hooks (Ruff lint/format, mypy, general hygiene) once per clone:

uv run pre-commit install

Run them against the full tree when needed:

uv run pre-commit run --all-files

License

This fuzzycoco software is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0).

Contents

Guide

API Reference

API Reference