Struct EpsilonArchive 

pub struct EpsilonArchive<V> { /* private fields */ }

Grid-based epsilon-dominance archive with bounded size.

Unlike ParetoFrontier::with_eps (numerical tolerance for float comparison), this type deliberately coarsens objective space into grid cells of width eps per dimension. At most one point is kept per grid cell, guaranteeing an archive size bounded by prod_i(range_i / eps_i).

Use this for streaming / online optimization where memory must be bounded while maintaining an approximate Pareto front.

use pare::{Direction, EpsilonArchive};

let mut archive = EpsilonArchive::new_uniform(
    vec![Direction::Maximize, Direction::Maximize],
    0.1,
);

archive.push(vec![0.91, 0.12], "a");
archive.push(vec![0.92, 0.11], "b"); // same grid cell as "a", replaces if better
archive.push(vec![0.50, 0.50], "c"); // different cell, kept
assert!(archive.len() <= 2);

Implementations

impl<V> EpsilonArchive<V>

pub fn new(directions: Vec<Direction>, grid_eps: Vec<f64>) -> Self

Create an archive with per-dimension grid epsilon values.

Panics

Panics if grid_eps.len() != directions.len() or any epsilon is not positive.

pub fn new_uniform(directions: Vec<Direction>, eps: f64) -> Self

Create an archive with a uniform grid epsilon for all dimensions.

Panics

Panics if eps is not positive.

pub fn push(&mut self, values: Vec<f64>, data: V) -> bool

Insert a point into the archive if it is not epsilon-dominated.

Returns true if the point was added (or replaced an existing point in the same cell), false if it was rejected.

Panics

Panics if values.len() != directions.len() or any value is non-finite.

pub fn len(&self) -> usize

Number of points in the archive.

pub fn is_empty(&self) -> bool

Is the archive empty?

pub fn points(&self) -> &[Point<V>]

Get all points in the archive.

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

Get the grid epsilon values.

pub fn directions(&self) -> &[Direction]

Get the optimization directions.

pub fn into_frontier(self) -> ParetoFrontier<V>

Convert into a standard Pareto frontier for downstream analysis (crowding distance, hypervolume, scoring, etc.).

The resulting frontier uses the default numerical epsilon (1e-9), not the grid epsilon. All archived points are inserted; since they are mutually non-epsilon-dominated, most will survive standard dominance filtering (though some edge cases near cell boundaries may be filtered).

Trait Implementations

impl<V: Clone> Clone for EpsilonArchive<V>

fn clone(&self) -> EpsilonArchive<V>

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl<V: Debug> Debug for EpsilonArchive<V>

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

Formats the value using the given formatter.