pub fn nmse(a: &[f32], b: &[f32]) -> f64Expand description
Normalized mean-squared error.
NMSE = mse(a, b) / mse(a, 0). Returns the raw mse(a, b) when the
reference is degenerate (all zeros) — falls back gracefully so tests
for ops that legitimately output zero don’t divide by zero.
§Panics
Panics if a.len() != b.len().