Struct TimeSeriesOde 

pub struct TimeSeriesOde {
    pub times: Vec<f64>,
    pub observations: Vec<Vec<f64>>,
    pub solver: NeuralOdeSolver,
    pub learning_rate: f64,
    pub n_iter: usize,
    pub loss_history: Vec<f64>,
}

A convenience wrapper that fits a Neural ODE to observed time-series data using gradient descent on the MSE loss.

Fields

times: Vec<f64>

Observed time points (sorted ascending).

observations: Vec<Vec<f64>>

Corresponding observations, one per time point.

solver: NeuralOdeSolver

The underlying Neural ODE solver.

learning_rate: f64

Learning rate for gradient descent.

n_iter: usize

Number of fitting iterations.

loss_history: Vec<f64>

MSE loss history across iterations.

Implementations

impl TimeSeriesOde

pub fn new( times: Vec<f64>, observations: Vec<Vec<f64>>, solver: NeuralOdeSolver, learning_rate: f64, n_iter: usize, ) -> Self

Construct a TimeSeriesOde from observed data and an initial solver.

pub fn fit(&mut self)

Run gradient descent to fit the Neural ODE parameters to the observations.

Uses finite-difference parameter gradients (one perturbation per weight). This is intentionally simple — a production implementation would use the adjoint method for efficiency.

pub fn compute_loss(&self, dt: f64) -> f64

Compute MSE between predicted trajectory and observations.

pub fn predict(&self, t: f64) -> Vec<f64>

Predict the state at time t by integrating from the first observation.

Returns the predicted observation vector.

Trait Implementations

impl Clone for TimeSeriesOde

fn clone(&self) -> TimeSeriesOde

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for TimeSeriesOde

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.