Struct LogLikelihoodTolerance 

pub struct LogLikelihoodTolerance { /* private fields */ }

A stopping rule that lets the EM algorithm run until the absolute difference in successive, normalised log-likelihood values falls below some tolerance.

The log-likelihood will be normalised by the number of sites, so that it becomes a per-site measure. This makes it easier to find a reasonable tolerance for a range of input sizes.

Implementations

impl LogLikelihoodTolerance

pub fn absolute_difference(&self) -> f64

Returns the absolute difference between the two most recent normalised log-likelihood values.

pub fn log_likelihood(&self) -> LogLikelihood

Returns the current, normalised log-likelihood value.

pub fn new(tolerance: f64) -> Self

Creates a new stopping rule that allows EM steps until the absolute difference in successive, normalised log-likelihood values falls below tolerance.

pub fn tolerance(&self) -> f64

Returns the tolerance defining convergence.

Trait Implementations

impl<T> Stop<T> for LogLikelihoodTolerance

where T: EmStep<Status = SumOf<LogLikelihood>>,

type Status = <T as EmStep>::Status

A status from the E-step that may be used for checking convergence.

fn stop<const N: usize>( &mut self, _em: &T, status: &Self::Status, _sfs: &Sfs<N>, ) -> bool

Returns true if the algorithm should stop, false otherwise.

impl StoppingRule for LogLikelihoodTolerance

fn and<S>(self, other: S) -> Both<Self, S>

where Self: Sized,

Returns a new stopping rule that requires that both this and another stopping rule must indicicate convergence before stopping.

fn inspect<F>(self, f: F) -> Inspect<Self, F>

where Self: Sized, F: FnMut(&Self),

Inspect the stopping rule after each E-step.

fn or<S>(self, other: S) -> Either<Self, S>

where Self: Sized,

Returns a new stopping rule that requires that either this or another stopping rule must indicicate convergence before stopping.