Crate iron_learn 

Iron Learn - A Rust Machine Learning Library

A pure Rust, type-safe machine learning library in pure Rust with GPU acceleration. Provides tensor operations, gradient-based optimization, and support for complex-valued computations.

Key Features

• GPU-Accelerated Training: CUDA kernels for efficient large-scale optimization
• Type Safety: All operations return Result types; no unwrap-induced panics
• Zero-Copy Borrowing: Compute without unnecessary allocations
• Generic Numerics: Works with f64, f32, i32, i64, Complex, and custom types
• Automatic Preprocessing: Feature normalization and bias handling built-in

Modules

• tensor: Core multi-dimensional array operations
• complex: Complex number arithmetic
• numeric: Generic type system for computations
• gradient_descent: CPU optimization algorithms
• gpu_regression: CUDA-accelerated training
• regression: High-level training interface
• app_context: Global application state

Quick Example

use iron_learn::{Tensor, linear_regression};

let x = Tensor::new(vec![10, 3], vec![/* data */])?;
let y = Tensor::new(vec![10, 1], vec![/* labels */])?;
let mut w = Tensor::new(vec![4, 1], vec![0.0; 4])?;

for _ in 0..1000 {
    w = linear_regression(&x, &y, &w, 0.01)?;
}

Version

0.5.0 - GPU-Accelerated Release

Structs

AppContext

Global application context with training configuration and GPU capabilities

Complex

A complex number represented by real and imaginary components

Data

Container for complete dataset with both linear and logistic regression splits

Tensor

The Tensor structure is the cornerstone of this library, providing a comprehensive suite of mathematical operations for the manipulation of multidimensional data.

XY

Training and test data container for regression tasks

Statics

GLOBAL_CONTEXT

Global singleton instance of application context

Traits

Numeric

The Numeric trait defines a set of operations that numeric types must support. It includes basic arithmetic operations and the ability to return special values like zero and one.

Functions

gradient_descent

Core gradient descent optimization algorithm

init_context

Initialize the global application context

linear_regression

Linear regression with automatic preprocessing

logistic_regression

Logistic regression with automatic preprocessing

predict_linear

Prediction function for linear regression

predict_logistic

Prediction function for logistic regression

run_linear

Trains and evaluates linear regression model

run_logistic

Trains and evaluates logistic regression model

run_logistics_cuda

Main GPU training function for logistic regression Implements gradient descent using CUDA kernels for efficient computation Returns accuracy metrics on test set upon completion