Struct MatrixProfile 

pub struct MatrixProfile { /* private fields */ }

Computed matrix profile for a fixed (series, window) pair.

The profile array is always in 1-to-1 correspondence with the n − m + 1 candidate subsequences; profile[i] is the nearest non-trivial-neighbour distance for subsequence T[i..i+m], and index[i] is that neighbour’s starting offset.

Examples

use anomstream_core::MatrixProfile;

// Synthetic: smooth cosine with one injected spike near i=48.
let mut series: Vec<f64> = (0..128)
    .map(|i| (f64::from(i as i32) * 0.3).cos())
    .collect();
for v in &mut series[48..56] {
    *v += 5.0;
}
let mp = MatrixProfile::compute(&series, 8, None).expect("mp");
let (pos, _score) = mp.discord();
assert!((40..=56).contains(&pos));

Implementations

impl MatrixProfile

pub fn compute( series: &[f64], window: usize, exclusion_zone: Option<usize>, ) -> RcfResult<Self>

Run STOMP over series with subsequence length window.

exclusion_zone is the half-width of the trivial-match band around each query (|i − j| < exclusion_zone is skipped). Pass None for the conventional ceil(window / 4) default (Keogh / Mueen matrix-profile tutorials).

Complexity

Time: O(n²) over the series length n, with an extra O(n · m) one-time cost on the first-column seed (where m = window). Practical wall-clock: ~3 ms at (n = 1 024, m = 32), ~49 ms at (n = 4 096, m = 128) on a modern core. Budget aggressively — doubling n quadruples the cost. Do not call on hot-path streams; reserve for forensic windows captured by the online crate::ShingledForest or triage batch jobs.

Memory: O(n) for the profile + O(n) for the scratch dot-product column.

Errors

Returns RcfError::InvalidConfig when window < MIN_WINDOW, window > MAX_WINDOW, when the series is too short (series.len() < 2 · window), when series contains a non-finite value, or when exclusion_zone would leave zero valid neighbours.

pub fn window(&self) -> usize

Window length used when computing this profile.

pub fn exclusion_zone(&self) -> usize

Exclusion-zone half-width used when computing this profile.

pub fn len(&self) -> usize

Number of subsequences (n − m + 1).

pub fn is_empty(&self) -> bool

true when the profile holds zero subsequences — never returned by Self::compute, provided as a total accessor.

pub fn profile(&self) -> &[f64]

Per-subsequence nearest-neighbour distance vector.

pub fn profile_index(&self) -> &[usize]

Per-subsequence nearest-neighbour index vector.

pub fn discord(&self) -> (usize, f64)

Discord — subsequence whose nearest neighbour is farthest. Returns (start_index, distance).

pub fn discord_topk(&self, k: usize) -> Vec<(usize, f64)>

Top-k discords ranked by descending distance. k is clamped to Self::len. Uses a greedy suppression pass that skips any candidate within the exclusion zone of an already-emitted discord — prevents the top-k from clustering inside a single anomalous region.

pub fn motif(&self) -> (usize, f64)

Motif — subsequence whose nearest neighbour is closest. Returns (start_index, distance).

Trait Implementations

impl Clone for MatrixProfile

fn clone(&self) -> MatrixProfile

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for MatrixProfile

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for MatrixProfile

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for MatrixProfile

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.