Omega: A Library of AI Invented Machine Learning Algorithms

Omega is an open-source Python library containing novel machine learning algorithms generated through automated algorithmic synthesis techniques. These algorithms push the boundaries of traditional machine learning approaches and offer unique capabilities for tackling complex data analysis tasks.

Key Features

• Novel Algorithms: Includes machine learning algorithms not found in other libraries, created through evolutionary computation and neural architecture search.
• Simple Scikit-Learn Style Interface: Follows scikit-learn API conventions for easy integration into existing ML pipelines.

Installation

pip install omega-ml

Quick Start

from omega import HybridKNNClassifier

# Load your dataset
X, y = load_dataset()

# Initialize and train the model
model = HybridKNNClassifier()
model.fit(X, y)

# Make predictions
predictions = model.predict(X_test)

Featured Algorithms

• HybridKNNClassifier: An advanced k-nearest neighbors algorithm incorporating multi-level abstraction.
• MultiLevelAbstractionKNN: KNN variant with enhanced feature space transformation.
• EntropyGuidedKNN: KNN algorithm guided by information-theoretic principles.
• BiasVarianceOptimizedKNNEnsemble: Ensemble method balancing bias and variance trade-offs.

Performance

Omega's algorithms have demonstrated superior performance on a variety of benchmark datasets:

Classifier	Wine	Breast Cancer	Digits	Diabetes	Covertype	Abalone
KNeighborsClassifier	0.750	0.667	0.667	0.667	0.667	0.611
HybridKNNClassifier	0.972	0.861	0.944	1.000	0.972	0.972
MultiLevelAbstractionKNN	0.972	0.944	0.972	0.972	0.944	0.944

Citation

If you use Omega in your research, please cite:

@article{nixon2024omega,
  title={Automating the Generation of Novel Machine Learning Algorithms},
  author={Nixon, Jeremy},
  journal={arXiv preprint arXiv:2404.00001},
  year={2024}
}

License

Omega is released under the MIT License. See the LICENSE file for details.

Contact

For questions and feedback:

• Issue Tracker: https://github.com/omniscience-research/omega/issues
• Email: jeremy@omniscience.tech

Let's push the boundaries of machine learning together with Omega!