Struct changepoint::BocpdTruncated

pub struct BocpdTruncated<X, H, Fx, Pr> where
    H: Fn(usize) -> f64,
    Fx: Rv<X> + HasSuffStat<X>,
    Pr: ConjugatePrior<X, Fx>,
    Fx::Stat: Clone,  { /* fields omitted */ }

Online Bayesian Change Point Detection with a truncated tail

The truncation takes place after run length probabilites are computed. The truncation point is chosen based on the most recent point from which all successive mass is below the given threshold.

Implementations

impl<X, H, Fx, Pr> BocpdTruncated<X, H, Fx, Pr> where
    H: Fn(usize) -> f64,
    Fx: Rv<X> + HasSuffStat<X>,
    Pr: ConjugatePrior<X, Fx>,
    Fx::Stat: Clone, 

pub fn new(hazard: H, fx: Fx, predictive_prior: Pr) -> Self

Create a new Bocpd analyzer

Parameters

• hazard - The hazard function for P_{gap}.
• fx - Predictive distribution. Used for generating an empty SuffStat.
• predictive_prior - Prior for the predictive distribution.

Example

use changepoint::{BocpdTruncated, constant_hazard};
use rv::prelude::*;
use std::sync::Arc;

let cpd = BocpdTruncated::new(
    constant_hazard(250.0),
    Gaussian::standard(),
    NormalGamma::new_unchecked(0.0, 1.0, 1.0, 1.0),
);

pub fn with_cutoff(self, cutoff_threadhold: f64) -> Self

Change the cutoff for mass to be discarded on the tail end of run-lengths

Trait Implementations

impl<X, H, Fx, Pr> RunLengthDetector<X> for BocpdTruncated<X, H, Fx, Pr> where
    H: Fn(usize) -> f64,
    Fx: Rv<X> + HasSuffStat<X>,
    Pr: ConjugatePrior<X, Fx>,
    Fx::Stat: Clone, 

fn step(&mut self, data: &X) -> &[f64]

Update the model with a new datum and return the distribution of run lengths.

fn reset(&mut self)

Reset internal state, new run-lengths will refer to steps after this point.