Time-Series Models¶

Public objects from aberrant.model.timeseries:

• XLagDAMP

XLagDAMP implements the original authors' X-Lag Amnesic DAMP backward processing with lookahead=0, their pure-online mode. It scores the subsequence ending at each scalar event by its nearest preceding subsequence.

Notes: - Input must contain exactly one consistently named numeric feature. - subsequence_length should be roughly 50% to 90% of a typical period. - Memory is bounded by x_lag + subsequence_length - 1 learned values. - Early-abandoned scores are approximate; the highest peaks identify discord candidates. - Constant subsequences are unsupported by the reference algorithm and rejected.