Estimate Density Ratios via RuLSIF

Relative unconstrained Least-Squares Importance Fitting. Kernel-based closed-form density ratio estimation.

Usage

estimate_rulsif(x_num, x_den, kernel = kernel_spec(), lambda = 0.1, alpha = 0)

Arguments

x_num

Numeric matrix. Numerator samples.

x_den

Numeric matrix. Denominator samples.

kernel

Kernel specification. Default is RBF with median heuristic.

lambda

Regularisation parameter. Default is 0.1.

alpha

Relative parameter (0 = LSIF, 0.5 = symmetric). Default is 0.

Value

Named list with weights and ess.