Parallel Boosting

This repository provides an unofficial implementation of the paper "Optimal Parallelization of Boosting".

Features

• Built on top of scikit-learn 1.3.2.
• Implements ParallelBoostingClassifier, a parallelized version of scikit-learn's AdaBoostClassifier.
• Parallel computation is performed using multi-threading.
• Includes two test cases for the algorithm: the noisy spiral dataset and MNIST.

Basic Usage

Here is a simple example demonstrating the usage of ParallelBoostingClassifier:

from parallel_boosting import ParallelBoostingClassifier
from sklearn.tree import DecisionTreeClassifier
boost = ParallelBoostClassifier(
    estimator=DecisionTreeClassifier(max_depth=3),
    n_rounds=200,
    n_queries=20,
    n_boosting_steps=10,
    learning_rate=0.5,
    random_state=seed,
    algorithm='SAMME',
    gamma=0.05,
    subsample_constant=1.0
)
boost.fit(X_train, y_train)
y_pred = boost.predict(X_test)