irwls.lr

vector of size p indicating types of columns in y (gaussian, binary, poisson)

var.type

positive number, value of the nuclear norm regularization parameter

lambda1

integer, maximum number of iterations

maxit

matrix of size (nb of ind.)x(number of variables), initial value of the individual effect, default 0

theta0

positive number, convergence criterion

thresh

boolean, if TRUE information about convergence will be displayes, default FALSE

trace.it

integer, maximum rank of interaction matrix theta

max.rank

positive number, backtracking line search parameter, default 0.01

Estimation of main effects and interactions in mixed data sets with missing values. Numeric, binary and count variables are supported. Main effects and interactions are modelled using an exponential family parametric model. Particular examples include the log-linear model for count data and the linear model for numeric data. Estimation is done through a convex program where main effects are assumed sparse and the interactions low-rank. Genevi<c3><a8>ve Robin, Olga Klopp, Julie Josse, <c3><89>ric Moulines, Robert Tibshirani (2018) <arXiv:1806.09734>.

irwls.lr: irwls.lr

Description

Usage

Arguments

Value