Struct ActivationHistogram 

pub struct ActivationHistogram {
    pub bins: Array2<f32>,
    pub edges: Array1<f32>,
    pub num_dims: usize,
    /* private fields */
}

2D histogram of activation distributions across sequence steps.

Bins are computed along the value axis; each column represents one dimension of the activation vector.

Fields

bins: Array2<f32>

Bin counts — shape (num_bins, num_dims)

edges: Array1<f32>

Bin edges — shape (num_bins + 1,)

num_dims: usize

Number of activation dimensions

Implementations

impl ActivationHistogram

pub fn new(num_bins: usize, min_val: f32, max_val: f32, num_dims: usize) -> Self

Create a new histogram with num_bins evenly spaced between min_val and max_val for num_dims dimensions.

pub fn update(&mut self, x: &Array1<f32>) -> ModelResult<()>

Accumulate one activation vector into the histogram.

pub fn density(&self) -> Array2<f32>

Return normalized histogram (density) per dimension.

Each column sums to ≈ 1.0 (or 0.0 if no data was accumulated).

pub fn per_dim_entropy(&self) -> Array1<f32>

Per-dimension Shannon entropy of the distribution (in nats).

pub fn most_active_dims(&self, top_k: usize) -> Vec<usize>

Return indices of the top_k dimensions with highest entropy.

Trait Implementations

impl Clone for ActivationHistogram

fn clone(&self) -> ActivationHistogram

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ActivationHistogram

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

Formats the value using the given formatter.