faster-rule

faster-rule is the distribution name of the NumPy-first additive rule ensemble package imported as fastogb. It provides configurable query-selection objectives, search algorithms, losses and weight-update methods, with mandatory Numba CPU acceleration and optional CUDA acceleration for supported NVIDIA systems.

Install the current checkout from the project root with pip install -e .. An editable installation immediately uses subsequent source changes, although a running Python or Jupyter process must be restarted after imports change.

The main public estimator is GeneralRuleBoostingEstimator. Its complete constructor, nested configuration, methods and fitted attributes are described in the GeneralRuleBoostingEstimator interface.

from fastogb import (FullyCorrective, GeneralRuleBoostingEstimator, OrthogonalBoostingObjective,
                     load_csv)

data, target, feature_names, categorical = load_csv(
    'data.csv', target_name='target', target_map={'negative': -1.0, 'positive': 1.0})
model = GeneralRuleBoostingEstimator(
    num_rules=10, objective_function=OrthogonalBoostingObjective,
    weight_update_method=FullyCorrective(), loss='logistic', search='greedy', n_jobs=4,
    search_params={'feature_names': feature_names, 'categorical': categorical},
    objective_params={'epsilon': 1e-4})
model.fit(data, target)

print(model.rules_)
predictions = model.predict(data)
probabilities = model.predict_proba(data)