Struct NeuralOde 

pub struct NeuralOde { /* private fields */ }

A Neural ODE for a 2-D system: an MLP field f_θ : ℝ² → ℝ², trained by trajectory matching.

Implementations

impl NeuralOde

pub fn new(hidden: usize, seed: u64) -> Self

A Neural ODE with the given hidden width.

pub fn field(&self, x: &[f64]) -> [f64; 2]

The learned field f_θ(x) in plain f64 (for inspection).

pub fn rollout(&self, x0: &[f64], n: usize, dt: f64) -> Vec<[f64; 2]>

Roll out the learned dynamics from x0 for n RK4 steps of size dt (plain f64).

pub fn train_step(&mut self, obs: &[[f64; 2]], dt: f64, lr: f64) -> f64

One Adam step of trajectory matching: integrate f_θ from obs[0] and match the observed states, backpropagating through the whole rollout. Returns the loss.

pub fn train( &mut self, obs: &[[f64; 2]], dt: f64, epochs: usize, lr: f64, ) -> f64

Train for epochs trajectory-matching steps; returns the final loss.