fingerprint

Sampling, Validation and ML Diagnostics

Model fingerprinting for linear, non-linear, and pairwise feature effects.

Why This Module Exists

Quantifies behavior of fitted models beyond scalar accuracy metrics.

Key Public APIs

RegressionModelFingerprint
ClassificationModelFingerprint
Effect
PairwiseEffect

Core Math

Partial Effect

\[f_j(x_j)=E_{X_{-j}}[f(X)|X_j=x_j]\]

Pairwise Interaction

\[I_{ij}=f(x_i,x_j)-f_i(x_i)-f_j(x_j)\]

Code Examples

Create regression fingerprint

use openquant::fingerprint::RegressionModelFingerprint;

let fp = RegressionModelFingerprint::new(&model, &x);
let effects = fp.linear_effects()?;

Implementation Notes

Compare fingerprints across retrains for drift detection.
Use pairwise effects to detect hidden interaction risk.