SomeB1oody/RustyML

What it solves

RustyML is a pure Rust library that provides a complete machine learning and deep learning workflow. It eliminates the need for C or C++ dependencies (FFI), offering a memory-safe, portable, and high-performance alternative for implementing ML models in Rust.

How it works

The library is organized into modular feature flags, allowing users to include only the necessary components. It leverages the ndarray crate for tensor operations and Rayon for multi-threaded parallelization of heavy kernels. It supports two types of neural network architectures: a Sequential model for linear stacks of layers and a Graph model for complex architectures like residual connections or multi-input/output networks.

Who it’s for

Developers who want to build and deploy machine learning models using a pure Rust ecosystem, avoiding the complexities of linking against external C/C++ libraries while maintaining memory safety and concurrency.

Highlights

• Pure Rust Implementation: No FFI or external C/C++ dependencies for easier portability and memory safety.
• Broad Algorithm Coverage: Includes classical supervised/unsupervised learning, anomaly detection, and a full neural network framework.
• Flexible NN Architectures: Supports both sequential and graph-based models (allowing for residual blocks and shared towers).
• Deterministic Results: Provides global and per-component seed management for reproducible experiments.
• Model Persistence: Uses Serde and postcard to save and load trained models as compact binary files.
• Scikit-learn Compatible Metrics: Includes a comprehensive suite of evaluation metrics for regression, classification, and clustering.