Struct ForestBuilder 

pub struct ForestBuilder<const D: usize> { /* private fields */ }

Fluent builder for RandomCutForest.

Defaults: num_trees = 100, sample_size = 256, time_decay = 0.1 / sample_size (matches AWS Java CompactSampler; call Self::time_decay with 0.0 to recover strict uniform sampling), RNG seeded from entropy.

D is the per-point dimensionality. Callers pin it at construction via turbofish: ForestBuilder::<4>::new().

Examples

use anomstream_core::ForestBuilder;

let mut forest = ForestBuilder::<4>::new()
    .num_trees(50)
    .sample_size(64)
    .seed(42)
    .build()
    .expect("AWS-conformant config");
forest.update([0.0, 0.0, 0.0, 0.0]).expect("dim matches");

Implementations

impl<const D: usize> ForestBuilder<D>

pub fn new() -> Self

Start a new builder for D-dimensional points.

pub fn num_trees(self, n: usize) -> Self

Override the number of trees.

pub fn sample_size(self, s: usize) -> Self

Override the per-tree reservoir size. When time_decay has not been explicitly set via Self::time_decay, the default 0.1 / sample_size is re-resolved against the new value so the effective recency bias stays consistent with AWS’s CompactSampler formula.

pub fn time_decay(self, d: f64) -> Self

Override the sampler time-decay factor. Pass 0.0 to disable recency bias and recover strict uniform reservoir sampling. Once called, the builder stops auto-resolving time_decay from subsequent Self::sample_size changes — the caller’s choice wins.

pub fn seed(self, seed: u64) -> Self

Pin the RNG seed for reproducible runs.

pub fn num_threads(self, n: usize) -> Self

Build a dedicated rayon thread pool of size n for this forest’s parallel score / attribution / update paths. Requires the parallel cargo feature. When unset (default) rayon’s global pool is used.

pub fn initial_accept_fraction(self, f: f64) -> Self

Override the warmup admission fraction forwarded to each per-tree reservoir. See crate::ReservoirSampler module-level docs for semantics. 1.0 (the default) disables the gate; AWS’s CompactSampler uses 0.125.

pub fn feature_scales(self, scales: [f64; D]) -> Self

Set per-dimension multiplicative weights applied to every point before it reaches the forest’s hot paths. See RcfConfig::feature_scales for semantics. Pass the exact [f64; D] array — length is checked against the builder’s compile-time D at Self::build time, contents are checked for finiteness and positivity by RcfConfig::validate.

pub fn clear_feature_scales(self) -> Self

Drop any previously-set feature_scales — returns the builder to the unweighted state. Useful for tests that want to clear a shared template’s scale vector before building a specialised forest.

pub fn config(&self) -> &RcfConfig

Read-only access to the config under construction.

pub const fn dimension(&self) -> usize

Per-point dimensionality (compile-time D).

pub fn build(self) -> RcfResult<RandomCutForest<D>>

Validate the config and instantiate the forest.

Errors

Forwards RcfConfig::validate errors and propagates any failure from the underlying RandomCutForest constructor.

Trait Implementations

impl<const D: usize> Clone for ForestBuilder<D>

fn clone(&self) -> ForestBuilder<D>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<const D: usize> Debug for ForestBuilder<D>

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

Formats the value using the given formatter.

impl<const D: usize> Default for ForestBuilder<D>

fn default() -> Self

Returns the “default value” for a type.