English | 简体中文

tbls

Tree-based Broad Learning System (TBLS) for classification, with a fully scikit-learn-compatible API.

tbls packages a Tree-based Broad Learning System classifier (TBLS) built from small regression trees arranged in the Broad Learning System's two-stage "mapping → enhancement" architecture, trained with a closed-form ridge solve, plus the building blocks it is composed from: a classic random-weight BroadLearningSystem, and two-view kernel CCA feature extractors (PairwiseKCCA, GraphFuzzyKCCA) with optional Intuitionistic Fuzzy Set sample weighting and graph-Laplacian regularization.

Installation

pip install tbls

The published package depends only on numpy, scipy, and scikit-learn.

Quickstart

from sklearn.datasets import make_classification
from sklearn.model_selection import cross_val_score

from tbls import TBLS

X, y = make_classification(n_samples=200, n_features=20, random_state=0)

model = TBLS(n_map_trees=10, n_enhance_trees=10, random_state=0)
model.fit(X, y)
print(model.predict(X[:5]))
print(model.predict_proba(X[:5]))

print(cross_val_score(model, X, y, cv=3))

For runnable worked examples against the real biomedical dataset (a single TBLS run, and a TBLS --grid hyperparameter sweep), see examples/. They call the same experiment internals as the training CLI and finish in seconds.

What's in the box

All classifiers (TBLS, BroadLearningSystem) are standard sklearn.base.BaseEstimator/ClassifierMixin implementations and work with cross_val_score, GridSearchCV, Pipeline, etc.

Documentation

Development

uv sync --group dev --group experiments   # install everything
ruff check .                              # lint
ruff format --check .                     # format check
mypy src/tbls                             # type check
pytest                                     # tests

See docs/development.md for the full guide.

Contributing

Issues and pull requests are welcome. Please read docs/development.md first for project conventions (Conventional Commits, the estimator contract, docs/translation structure).

References

The GEIB intuitionistic-fuzzy-set formulation used by TBLS follows Chen et al., IEEE Transactions on Fuzzy Systems, 2025. The Broad Learning System architecture follows the original BLS literature.

License

Apache License 2.0 — see LICENSE.