Edit on GitHub

Likelihoods

GaussianProcesses.BernLik — Type
BernLik <: Likelihood

Bernoulli likelihood

\[p(y = k | f) = θᵏ (1 - θ)^{1-k}\]

for $k ∈ \{0,1\}$, where $θ = Φ(f)$ and $f$ is the latent Gaussian process.

source
GaussianProcesses.BinLik — Type
BinLik <: Likelihood

Binomial likelihood

\[p(y = k | f) = k!/(n!(n-k)!) θᵏ(1 - θ)^{n-k}\]

for number of successes $k ∈ \{0, 1, …, n\}$ out of $n$ Bernoulli trials, where $θ = \exp(f)/(1 + \exp(f))$ and $f$ is the latent Gaussian process.

source
GaussianProcesses.ExpLik — Type
ExpLik <: Likelihood

Exponential likelihood

\[p(y | f) = θ\exp(-θy),\]

where $θ = \exp(-f)$ and $f$ is the latent Gaussian process.

source
GaussianProcesses.GaussLik — Type
GaussLik <: Likelihood

Gaussian, a.k.a. Normal, likelihood

\[p(y | f, σ) = 1 / √(2πσ²) \exp(-(y - f)²/(2σ²)),\]

where standard deviation $σ$ is a non-fixed hyperparameter and $f$ is the latent Gaussian process.

source
GaussianProcesses.PoisLik — Type
PoisLik <: Likelihood

Poisson likelihood

\[p(yᵢ = k | fᵢ) = θᵏ\exp(-θ)/k!\]

for $k ∈ N₀$, where $θ = \exp(f)$ and $f$ is the latent Gaussian process.

source
GaussianProcesses.StuTLik — Type
StuTLik <: Likelihood

Student-t likelihood (a.k.a. non-standardized Student's t-distribution)

\[p(y | f, σ) = Γ((ν + 1)/2)/[Γ(ν/2)√(πν)σ](1 + 1/ν((y - f)/σ)²)^{-(ν + 1)/2}\]

with degrees of freedom $ν ∈ N₀$, where scale $σ$ is a non-fixed hyperparameter and $f$ is the latent Gaussian process.

source