MLZero: Machine Learning from Scratch

Overview

MLZero is a Python library providing a collection of machine learning algorithms implemented from scratch. The goal is to offer a clear, educational codebase for understanding the fundamentals of machine learning, with practical driver scripts and a modular design for easy extension.

Features

• Classifiers: Perceptron, AdaLine, Logistic Regression, k-Nearest Neighbors (kNN), Naive Bayes, Softmax Regression
• Clusterers: K-Means clustering algorithm
• Regressors: Linear regression (closed-form and gradient descent), L1 (Lasso) and L2 (Ridge) regularization, ElasticNet, polynomial regression, multiple variable regression
• Small Neural Nets: Basic implementation of a multi-neuron layer
• Decomposers: Principal Component Analysis (PCA)
• Metrics: Regression and classification metrics (MAE, MSE, R², accuracy, precision, recall, F1, etc.)

Requirements

To run this project, you need the following Python libraries:

• numpy
• matplotlib for plotting purpose not necessary otherwise, recommended to have installed

Install the dependencies using:

bash pip install -r requirements.txt

Directory Structure

mlzero/ ├── classifiers/ # Classification algorithms and driversC/ ├── clusterers/ # Clustering algorithms and driversK/ ├── regressors/ # Regression algorithms and driversR/ ├── decomposers/ # Dimensionality reduction and driversD/ ├── metrics/ # Regression and classification metrics ├── small_neural_net/ # Multi-neuron layer implementations └── requirements.txt # Dependencies list

Usage

Each algorithm has a corresponding driver script in its drivers subdirectory. For example:

Run the ElasticNet regressor:

bash python regressors/driversR/driverElasticNet.py

Run the kNN classifier:

bash python classifiers/driversC/driverKNNClassifier.py

Development Status

MLZero is under active development. The codebase is modular and designed for educational purposes. Contributions for new algorithms, bug fixes, and documentation improvements are welcome.

Contributing

Contributions are welcome! Feel free to fork the repo, submit issues, or open pull requests. Please ensure your code is well-documented and tested before submitting.

License

This project is licensed under the MIT License.