divergence_coefficient(): Lyubishchev’s standardised
separation coefficient D for two groups on continuous features.scatter_ellipse(): fits per-class covariance ellipses
(mean, covariance, sample size).transgression(): detects ellipse overlap via squared
Mahalanobis distance against a chi-squared threshold.classify(): Edgeworth-Pearson multivariate posterior
classification of a new specimen with a numerically stable softmax.