Struct rv::ConjugateModel

pub struct ConjugateModel<X, Fx, Pr> where
    Fx: Rv<X> + HasSuffStat<X>,
    Pr: ConjugatePrior<X, Fx>,  { /* fields omitted */ }

A wrapper for a complete conjugate model

Paramters

X: The type of the data/observations to be modeled Fx: The type of the likelihood, f(x|θ) Pr: The type of the prior on the parameters of Fx, π(θ)

Methods

impl<X, Fx, Pr> ConjugateModel<X, Fx, Pr> where     Fx: Rv<X> + HasSuffStat<X>,     Pr: ConjugatePrior<X, Fx>,

pub fn new(fx: &Fx, pr: Arc<Pr>) -> Self

Create a new conjugate model

Arguments

fx:

Example

use std::sync::Arc;
use rv::prelude::*;
use rv::ConjugateModel;

let pr = Arc::new(Beta::jeffreys());
let fx = Bernoulli::uniform();
let model = ConjugateModel::<bool, Bernoulli, Beta>::new(&fx, pr);

pub fn ln_m(&self) -> f64

Log marginal likelihood, f(obs)

pub fn ln_pp(&self, y: &X) -> f64

Log posterior predictive, f(y|obs)

pub fn posterior(&self) -> Pr::Posterior

Return the posterior distribution

Example

use std::sync::Arc;
use rv::prelude::*;
use rv::ConjugateModel;

let flips: Vec<bool> = vec![true, false, true, false, false, false];

let pr = Arc::new(Beta::new(1.0, 1.0).unwrap());
let fx = Bernoulli::uniform();
let mut model = ConjugateModel::<bool, Bernoulli, Beta>::new(&fx, pr);

model.observe_many(&flips);

let post = model.posterior();

assert_eq!(post, Beta { alpha: 3.0, beta: 5.0 });

Trait Implementations

impl<X, Fx, Pr> SuffStat<X> for ConjugateModel<X, Fx, Pr> where     Fx: Rv<X> + HasSuffStat<X>,     Pr: ConjugatePrior<X, Fx>,

fn observe(&mut self, x: &X)

Assimilate the datum x into the statistic

fn forget(&mut self, x: &X)

Remove the datum x from the statistic

fn observe_many(&mut self, xs: &Vec<X>)

Assimilate several observations

fn forget_many(&mut self, xs: &Vec<X>)

Forget several observations

impl<X, Fx, Pr> Rv<X> for ConjugateModel<X, Fx, Pr> where     Fx: Rv<X> + HasSuffStat<X>,     Pr: ConjugatePrior<X, Fx>,

fn ln_f(&self, x: &X) -> f64

Un-normalized probability function

fn ln_normalizer(&self) -> f64

The log of the constant term in the PDF/PMF. Should not be a function of any of the parameters.

fn draw<R: Rng>(&self, rng: &mut R) -> X

Single draw from the Rv

fn sample<R: Rng>(&self, n: usize, rng: &mut R) -> Vec<X>

Multiple draws of the Rv

fn f(&self, x: &X) -> f64

Un-normalized probability function

fn normalizer(&self) -> f64

The constant term in the PDF/PMF. Should not be a function of any of the parameters.