BayesianLinearRegressors.jl

BLRFunctionSample{Tw,Tϕ}

A function sampled from b::Union{BayesianLinearRegressor,BasisFunctionRegressor} by taking a fixed weight sample w ~ p(w). ϕ is the feature mapping if sampled from a BasisFunctionRegressor, or is the identity function if sampled from a BayesianLinearRegressor.

BasisFunctionRegressor{Tblr,Tϕ}

A Basis Function Regressor represents a Bayesian Linear Regressor where the input x is first mapped to a feature space through a basis function ϕ.

ϕ must be a function which accepts one of the allowed input types for BayesianLinearRegressors (ColVecs, RowVecs or Matrix{<:Real} - see the package readme for more details) and it must output one of these allowed types.

x = RowVecs(hcat(range(-1.0, 1.0, length=5)))
blr = BayesianLinearRegressor(zeros(2), Diagonal(ones(2)))

ϕ(x::RowVecs) = RowVecs(hcat(ones(length(x)), prod.(x)))
bfr = BasisFunctionRegressor(blr, ϕ)

var(bfr(x))

See [1], Section 3.1 for more details on basis function regression.

[1] - C. M. Bishop. "Pattern Recognition and Machine Learning". Springer, 2006.

BayesianLinearRegressor{Tmw, TΛw}

A Bayesian Linear Regressor is a distribution over linear functions given by

w ~ Normal(mw, Λw)
f(x) = dot(x, w)

where mw and Λw are the mean and precision of w, respectively.