Stochastic Gradient Trees - Python

Stochastic Gradient Trees¹ by Henry Gouk, Bernhard Pfahringer, and Eibe Frank implementation in Python. Based on the parer's accompanied repository code.

Python Version 3.7 or later

Used Python libraries:

• numpy>=1.20.2
• scipy>=1.6.2
• pandas>=1.3.3
• scikit-learn>=0.24.2

Usage:

    from pysgt.StochasticGradientTree import StochasticGradientTreeClassifier

    from sklearn.model_selection import train_test_split
    from sklearn.datasets import load_breast_cancer
    from sklearn.metrics import confusion_matrix, accuracy_score, log_loss

    def train(X, y):

        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.34)
        
        tree = StochasticGradientTreeClassifier()

        tree.fit(X_train, y_train)
    
        y_pred = tree.predict(X_test)

        proba = tree.predict_proba(X_test)        

        acc_test = accuracy_score(y_test, y_pred)
        print(confusion_matrix(y_test, y_pred))
        print('Acc test: ', acc_test)
        print('Cross entropy loss: ', log_loss(y_test, proba))

        return tree, acc_test

    if __name__ == "__main__":

        breast = load_breast_cancer(as_frame=True)

        X = breast.frame.copy()
        y = breast.frame.target
        
        X.drop(['target'], axis=1, inplace=True) 

        tree, _ = train(X, y)

Binary classification example:

python classification_breast.py

Multiclass classification (using the One-vs-the-rest multiclass strategy):

python classification_iris.py

Regression example:

python regression_diabetes.py

Footnotes

1. Gouk, H., Pfahringer, B., and Frank, E. Stochastic gradient trees. In Proceedings of The Eleventh Asian Conference on Machine Learning, volume 101 of Proceedings of Machine Learning Research, pp. 1094–1109. PMLR, 2019. ↩