init: Various parameter initialization methods for use with the...

In andreamrau/rpl: Randomized pairwise likelihood method for complex statistical inferences

Various parameter initialization methods for use with the rpl_optim function

Description

Various parameter initialization methods for use with the rpl_optim function

Usage

init(mydata, type = "unstructured", pi = NULL, block_indices = NULL)

Arguments

mydata	Data matrix
type	Type of intialization to perform, according to the desired structure of correlation matrix: unstructured ((d^2 - d)/2 correlation parameters estimated using the empirical Pearson correlation matrix), exchangeable (1 correlation parameter estimated by averaging off-diagonal of the empirical Pearson correlation matrix), one-factor (d theta parameters, where rho_ij = theta_i * theta_j, estimated using Nelson-Meader optimization of fit compared to the estimated Pearson correlation matrix), block_exchangeable (for b blocks, (b^2 - b)/2 correlation parameters estimated by averaging empirical Pearson correlation matrix within each block). In all cases, marginal parameters are initalized using the plug-in method with empirical marginal means. Several alternative initializations are also proposed for unstructured correlation matrices: PPearson correlation (unstructured_A, equivalent to unstructured), pairwise optimization with all data (unstructured_B), pairwise optimization with random subsample of data (unstructured_C), random initialization for the copula parameters. (unstructured_D)
pi	Bernoulli sampling parameter (0 < pi <= 1)
block_indices	For type="block_exchangeable", list containing the indices of variables belonging to each of the b blocks.

Value

Vector of parameter values to be used to initialize the rpl_optim function.

andreamrau/rpl documentation built on April 26, 2023, 3:57 p.m.