Struct StateTrajectory 

pub struct StateTrajectory { /* private fields */ }

Records and analyses the trajectory of hidden states over time.

Implementations

impl StateTrajectory

pub fn new(dim: usize) -> Self

Create a new trajectory recorder for the given state dimension.

pub fn push(&mut self, state: Array1<f32>) -> ModelResult<()>

Append a state vector; returns an error on dimension mismatch.

pub fn len(&self) -> usize

Number of states recorded.

pub fn is_empty(&self) -> bool

Whether no states have been recorded.

pub fn velocities(&self) -> ModelResult<Vec<f32>>

Compute per-step velocities: ||s_{t+1} - s_t||_2 for t in 0..len-1.

Returns an error if fewer than 2 states have been recorded.

pub fn participation_ratio(&self) -> ModelResult<f32>

Effective dimensionality via participation ratio.

PR = (Σ λ_i)² / Σ λ_i² where λ_i are per-dimension variances. A high value means the state uses many dimensions; a low value means information is concentrated in a few dimensions.

pub fn most_variable_dims(&self, k: usize) -> ModelResult<Vec<usize>>

Return indices of the k dimensions with highest variance, sorted descending.

pub fn autocorrelation(&self, lag: usize) -> ModelResult<f32>

Compute state autocorrelation at lag lag.

Returns the mean Pearson correlation of s_t and s_{t+lag} across all valid pairs. A lag-0 autocorrelation is always 1.0.

pub fn to_matrix(&self) -> ModelResult<Array2<f32>>

Flatten all states into an (num_steps, dim) matrix.