NEWS

---

lookout 2.0.2.9000

• The exact density estimates (fast = FALSE) now use a fixed-radius neighbour search via dbscan::frNN(), so memory no longer grows with NROW(X)^2. Results are unchanged up to rounding.
• Nearest neighbour searches now use dbscan rather than RANN, which is no longer imported.
• The default is now fast = NROW(X) > 10000.
• lookout() gives a more helpful error when the GPD cannot be fitted because too many observations have no neighbours within the bandwidth.
• mvscale() now matches weird::mvscale(), with a robust MCD covariance estimate by default. This changes the scaled data used by lookout() and persisting_outliers() when scale = TRUE.
• Fixed the Epanechnikov kernel scaling for multivariate data.
• The fast argument of find_tda_bw() is deprecated and ignored.
• Now requires R >= 4.1.0.
• Bug fixes and documentation improvements.

lookout 2.0.2 (2026-07-21)

• Compatibility with mlpack 4.8.0, which changed the value returned by mlpack::emst().

lookout 2.0.1 (2026-03-26)

• find_tda_bw() now much faster, especially for large data sets, by using MST instead of Rips complex to compute persistent homology.

lookout 2.0.0 (2026-01-19)

• Updated lookout algorithm as per Hyndman, Kandanaarachchi and Turner (2026).
• Added mvscale() to do robust multivariate scaling.
• Exported find_tda_bw().

lookout 0.1.4 (2022-10-13)

• First CRAN version, based on Kandanaarachchi and Hyndman (JCGS, 2022).