pub fn sample<const N: usize>(
peak: &[f64; N],
covariance: &[[f64; N]; N],
draws: usize,
seed: u64,
log_target: impl FnMut(&[f64; N]) -> Option<f64>,
) -> Option<Posterior<N>>Expand description
The posterior whose log density, up to a constant, is log_target, by draws weighted draws
around peak with the Laplace covariance covariance.
log_target gives None where the posterior can’t be had; that draw gets no weight. Only the
lower triangle of covariance is read. None overall if peak isn’t finite, covariance
isn’t positive definite, draws is 0, or no draw has a finite weight.
The same seed and a log_target that gives the same answer for the same point give the same
answer, bit for bit, on one platform (the draws go through the platform’s ln, ln_1p,
sqrt and exp).