gpr-algorithm 1.0.0

Gene Programming Rules (GPR) implementation

An implementation of an extremely simple classifier (GPR) that consists of highly interpretable fuzzy metarules and is suitable for many applications. GPR is effective in accuracy and area under the receiver operating characteristic (ROC) curve. We provide a Python implementation of the GPR algorithm to enable the use of the algorithm without using commercial software tools and open access to the research community. We also added enhancements to facilitate the reading and interpretation of the rules.

Example usage

import numpy as np
from gpr_algorithm import GPR

feature_names = ['weight', 'height']
target_names = ['sick', 'healthy']

cls = GPR(
    feature_names=feature_names,
    target_names=target_names,
    max_n_of_rules=2, max_n_of_ands=2, n_generations=10, n_populations=10,
    verbose=False
)

attributes = np.array([
    [.9, .1],  # sick
    [1., .9],  # sick
    [0., .9],
    [.1, .1]
])
labels = np.array([
    0,  # sick
    0,  # sick
    1,
    1
])
cls.fit(attributes, labels)

pred_labels = cls.predict(attributes)

assert np.all(labels == pred_labels)
rules = cls.rules
assert rules == ['IF weight is Low THEN healthy | Support: 0.9500', 'ELSE sick']