Struct DeltaEncoder 

pub struct DeltaEncoder;

Stateless delta encoding utilities

Implementations

impl DeltaEncoder

pub fn encode(old: &[f32], new: &[f32]) -> VectorDelta

Encode the difference between old and new as a VectorDelta.

scale is set to the maximum absolute value in the delta (useful for quantization). If old and new have different lengths, the shorter is padded with 0.0.

pub fn apply(base: &[f32], delta: &VectorDelta) -> Vec<f32>

Apply a single delta to base, returning the updated vector.

If base and delta.delta have different lengths, the shorter is padded.

pub fn apply_sequence(base: &[f32], deltas: &[VectorDelta]) -> Vec<f32>

Apply a sequence of deltas in order to base.

pub fn compress(delta: &VectorDelta, threshold: f32) -> VectorDelta

Zero out delta components whose absolute value is below threshold.

pub fn merge(deltas: &[VectorDelta]) -> Option<VectorDelta>

Merge multiple deltas into a single combined delta by summing components.

Returns None if deltas is empty. The merged delta gets id = 0, timestamp = 0, and the scale of the sum.

pub fn stats(deltas: &[VectorDelta]) -> DeltaStats

Compute statistics about a sequence of deltas.

pub fn magnitude(delta: &VectorDelta) -> f32

Compute the L2 magnitude of a single delta (Euclidean norm).

pub fn quantize_to_i8(delta: &VectorDelta) -> Vec<i8>

Quantize the delta to int8 range [-127, 127].

Uses the stored scale as the normalisation factor. A scale of 0.0 results in all-zero output.

pub fn from_i8( quantized: &[i8], scale: f32, id: usize, timestamp: u64, ) -> VectorDelta

Reconstruct a VectorDelta from int8 quantized values. scale must be the same scale used during quantize_to_i8.