Crate neurox 

Neurox — minimalist numerical and ML building blocks in Rust.

This crate provides:

• A Tensor type for numeric data (row-major)
• Core tensor ops (matrix multiplication, element-wise add/mul, broadcasting)
• Activation functions (ReLU, Sigmoid, Tanh, Softmax)
• A Dense layer with backprop and cached activations
• A sequential Model with forward pass and training via SGD/Adam
• Losses (MSE, Cross-Entropy), data utilities, and error types

All operations currently run on CPU and are designed for clarity and extensibility.

Example: small MLP with ReLU

use neurox::{Model, Tensor, Activation};

// 3 -> 4 -> 2 MLP
let mut model = Model::new(&[3, 4, 2], Activation::ReLU);

// batch of 8 samples with 3 features
let x = Tensor::random(8, 3);
let logits = model.forward(&x).unwrap();
let probs = neurox::activations::softmax(&logits);
println!("probs: {:?}", probs);

Re-exports

pub use crate::model::Model;

pub use crate::tensor::Tensor;

pub use crate::layers::Dense;

pub use crate::layers::Activation;

pub use crate::optimizer::SGD;

pub use crate::optimizer::Adam;

pub use crate::errors::NeuroxError;

pub use crate::errors::NeuroxResult;

Modules

activations

Provides activation functions and their derivatives for neural networks.

data

Provides utilities for data loading and manipulation.

errors

layers

Defines the layers of a neural network, such as the Dense layer.

loss

model

Defines the main Model struct, its training loops, and evaluation utilities.

ops

Provides basic mathematical operations for Tensors.

optimizer

Provides optimization algorithms for updating model parameters.

prelude

Prelude with the most commonly used items.

tensor

Defines the core Tensor struct and its associated methods.

utils