Experimental Machine Learning Library

minimum_ml is a lightweight, experimental machine learning library written in Rust. It provides basic building blocks for neural networks, including:

• Tensor operations and automatic differentiation
• Common activation functions (ReLU, Sigmoid, Softmax, etc.)
• Neural network layers (Linear, Convolution, etc.)
• Optimizers (SGD, Adam)
• Quantization support for model compression
• Dataset utilities for batching and iteration

Features

This crate supports optional features to minimize dependencies:

• serialization: Enable model saving/loading with serde (adds serde, serde_json, anyhow)
• logging: Enable TensorBoard logging (adds tensorboard-rs, chrono)
• progress: Enable progress bars during training (adds indicatif)
• full: Enable all features

By default, no optional features are enabled, keeping the dependency footprint minimal.

Examples

Minimal install (only core dependencies):

[dependencies]
minimum_ml = "0.1.1"

With serialization support:

[dependencies]
minimum_ml = { version = "0.1.1", features = ["serialization"] }

With all features:

[dependencies]
minimum_ml = { version = "0.1.1", features = ["full"] }

Usage Example

use minimum_ml::ml::{Tensor, Graph};
use minimum_ml::ml::params::MM;

// Create a computational graph
let graph = Graph::new();

// Create tensors and perform operations
let x = Tensor::new(vec![1.0, 2.0, 3.0], vec![3]);
let linear = MM::new(3, 2);
let output = linear.forward(&graph, &x);