Struct Mlp 

pub struct Mlp { /* private fields */ }

A fully-connected network with tanh hidden activations and a linear output layer.

Implementations

impl Mlp

pub fn new(sizes: &[usize], seed: u64) -> Self

A network with the given layer sizes, e.g. &[1, 16, 16, 1] (one input, two hidden layers of 16, one output). Weights are seeded-Xavier initialized; biases start at zero.

pub fn n_params(&self) -> usize

Number of trainable parameters.

pub fn forward(&self, x: &[f64]) -> Vec<f64>

Forward inference in plain f64 (no tape): map an input vector to the output vector.

pub fn train_step(&mut self, xs: &[Vec<f64>], ys: &[Vec<f64>], lr: f64) -> f64

One Adam optimizer step over the full batch. Returns the MSE before the step.

pub fn train( &mut self, xs: &[Vec<f64>], ys: &[Vec<f64>], epochs: usize, lr: f64, ) -> f64

Train for epochs full-batch Adam steps; returns the final MSE.