expectedarc(i,n) this polynomial indicates the expected number of NN-arcs that would pass over index i for a series of lengthn if the series was completely random. This is used to normalize the `segmentprofile` to account for bias towards the edges.

profile = matrix_profile(A, T, m, dist=ZEuclidean(); showprogress=true)

Mutual matrix profile between A and T.

profile = matrix_profile(T, m, [dist = ZEuclidean()]; showprogress=true)

Return the matrix profile and the profile indices of time series T with window length m. See fields profile.P, profile.I. You can also plot the profile. If dist = ZEuclidean() the STOMP algorithm will be used.

Reference: Matrix profile II.

motifs(p::Profile, k, found_motifs = Motif[]; r=2, th = p.m, dist = ZEuclidean())

• k is the number of motifs to extract
• r controls how similar two windows must be to belong to the same motif. A higher value leads to more windows being grouped together.
• th is a threshold on how nearby in time two motifs are allowed to be.

Also see the function anomalies(profile) to find anomalies in the data, sometimes called discords.

mpdist(A, B, m, d = ZEuclidean, k = ((length(A) + length(B)) - 2m) ÷ 20)

The MP distance between A and B using window length M and returning the kth smallest value.

mpdist(MP::AbstractArray, th::Real, N::Int)

The MP distance given a precomputed matrix profile, calculating k as th*N where N is the length of the data used to create MP.

mpdist_profile(A::AbstractVector, B::AbstractVector, m::Int, [d = ZEuclidean()])

MP distance profile between two time series using window length m.

mpdist_profile(T::AbstractVector, S::Int, m::Int, [d])

All MP distance profiles between subsequences of length S in T using internal window length m.

i = segment(p::Profile)

Returns an index i indicating the most likely segmentation point of profile p, i.e., the point which the fewst nearest-neighbor arcs passes over.

Ref: Matrix Profile VIII: Domain Agnostic Online Semantic Segmentation at Superhuman Performance Levels

segment_profile(p::Profile)

Calculate the MP-index of a matrix profile. This index has the same length as the profile, and tells you how many nearest-neighbor arcs passes over index i. It's normalized by the expected number of arcs for a completely random time series, so that a value of 1 indicates a poor segmentation point, and a value close to 0 indicates a likely segmentation point.

snippets(T, k, S; m = max(S ÷ 10, 4))

Summarize time series T by extracting k snippets of length S The parameter m controls the window length used internally.