Module quantization_error 

Quantization error tracking for INT8/binary quantization.

This module measures and bounds quantization error introduced by INT8 or binary quantization, enabling monitoring of search quality degradation over time.

Overview

Quantization reduces memory usage and speeds up distance computations, but at the cost of precision. The QuantizationErrorTracker accumulates per-vector error measurements and exposes rolling statistics (MSE, MAE, p99 MSE, SNR) to help operators decide when re-indexing or parameter tuning is necessary.

Example

use ipfrs_semantic::quantization_error::{QuantizationErrorTracker, QErrorError};

let mut tracker = QuantizationErrorTracker::new();

let original  = vec![0.1_f32, 0.2, 0.3, 0.4];
let quantized = vec![0.1_f32, 0.2, 0.3, 0.4]; // perfect quantization

let err = tracker.compute_error(&original, &quantized).unwrap();
assert!(err.mse < 1e-9);

tracker.record(err);
assert_eq!(tracker.history_len(), 1);

Structs

QuantizationError

Quantization error metrics for a single vector pair.

QuantizationErrorTracker

Tracks quantization error across a rolling window of vectors.

Enums

QErrorError

Errors that can be returned from QuantizationErrorTracker operations.