Struct Percentile 

pub struct Percentile {
    pub p: f64,
    pub goal: Goal,
    pub warmup: usize,
    pub startup: usize,
}

Prune what is doing worse at this step than p percent of the others.

The median pruner is p = 50, which is why there is no Median scheme — median is a constructor. What is compared is each trial’s best so far and not its latest: one bad epoch is noise.

Fields

p: f64

Between 0 and 100, and the share that is kept — a smaller one prunes more, the way round optuna reads it. At 50 the better half survives; at 0 everything but the best does not.

goal: Goal

Which way is better.

warmup: usize

No verdict before this many reports, however bad it looks. What buys a slow starter the epochs it needs.

startup: usize

No verdict until this many other trials have reached this step. Without it the first trial to finish becomes the bar for everybody.

Implementations

impl Percentile

pub fn median(goal: Goal, warmup: usize, startup: usize) -> Self

The median: p = 50.

pub fn verdict(&self, mine: &[f64], others: &[Vec<f64>]) -> Verdict

Continue, or the reason not to.

Trait Implementations

impl Clone for Percentile

fn clone(&self) -> Percentile

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for Percentile

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

Formats the value using the given formatter.

impl Display for Percentile

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

Formats the value using the given formatter.

impl From<Percentile> for Pruner

fn from(rule: Percentile) -> Self

Converts to this type from the input type.

impl PartialEq for Percentile

fn eq(&self, other: &Percentile) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for Percentile