Struct TensorStats 

pub struct TensorStats {
    pub count: usize,
    pub min: f32,
    pub max: f32,
    pub mean: f32,
    pub std: f32,
    pub nan_count: usize,
    pub inf_count: usize,
    pub l2_norm: f32,
}

Statistics for a tensor without storing the tensor itself.

Computes min, max, mean, std, L2 norm, NaN/Inf counts in a single pass. Used for anomaly detection (explosion, vanishing gradients, NaN propagation).

Example

let stats = TensorStats::from_slice(&tensor_data);
if stats.has_anomaly() {
    log::warn!("Anomaly detected: {}", stats.anomaly_description());
}

Fields

count: usize

Number of elements analyzed

min: f32

Minimum value (ignoring NaN/Inf)

max: f32

Maximum value (ignoring NaN/Inf)

mean: f32

Mean value (ignoring NaN/Inf)

std: f32

Standard deviation (ignoring NaN/Inf)

nan_count: usize

Count of NaN values

inf_count: usize

Count of Inf values

l2_norm: f32

L2 norm (sqrt of sum of squares)

Implementations

impl TensorStats

pub fn from_slice(data: &[f32]) -> Self

Compute statistics from a slice in a single pass.

Uses Welford’s algorithm for numerically stable mean/variance.

pub fn has_anomaly(&self) -> bool

Check if this tensor has any anomalies.

Anomaly detection rules (from E.11.2):

• NaN detected: nan_count > 0
• Explosion: max.abs() > 1e6 or std > 1e4
• Vanishing: std < 1e-6 (should check after first few layers)

pub fn is_vanishing(&self) -> bool

Check if values are vanishing (for layers past warmup).

pub fn anomaly_description(&self) -> Option<String>

Get a description of any anomaly detected.

Trait Implementations

impl Clone for TensorStats

fn clone(&self) -> TensorStats

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for TensorStats

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

Formats the value using the given formatter.

impl Default for TensorStats

fn default() -> TensorStats

Returns the “default value” for a type.

impl PartialEq for TensorStats

fn eq(&self, other: &TensorStats) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for TensorStats