Diagnostics¶
Robust distance diagnostics are central to the package. They help users decide whether a covariance estimate is meaningful for anomaly detection or whether the tail behavior is too far from Gaussian assumptions.
Diagnostic report¶
report = rc.diagnostic_report(est)
print(report.summary())
The report includes radial kurtosis, QQ tail deviation, condition number, detected fraction, and recommendations.
Distance profile / proline plot¶
rc.plot_robust_distance_profile(est, output_path="profile.png", show=False)
This plot shows sorted robust distances and the threshold. It makes the tail and threshold crossing easy to inspect.
Distance panel¶
rc.plot_robust_distance_panel(est, output_path="panel.png", show=False)
The panel combines a distance profile, histogram, and chi-square QQ plot.
2D anomaly diagnostics¶
rc.plot_anomaly_scatter_2d(est, X[:, :2], labels=y, output_path="scatter.png", show=False)
For real high-dimensional data, use PCA or domain features before plotting.