adam_gmic: The ADAM Algorithm for gMIC Optimization (experimental)

In liqun730/gMIC: Group Sparsity via Approximated Information Criteria

Description

The ADAM Algorithm for gMIC Optimization (experimental)

Usage

adam_gmic(X, y, a, lambda, gamma, group, family, stepsize, tol, maxit,
  b1 = 0.7, b2 = 0.9, e = 1e-08)

Arguments

X	Design matrix.
y	The response vector.
a	The approximation parameter for gMIC.
lambda	The penalization parameter for gMIC, e.g., 2 for AIC and long(n) for BIC.
gamma	The optimization parameter gamma.
group	The group structure of the model. For example, assume that X has 4 columns and group=c(1,1,2,2). It means the first 2 features form a group of variables and the last 2 features form another group of variables.
family	The type of glm model, should be one of "gaussian", "binomial" or "poisson".
stepsize	Stepsize for group coordinate descent.
tol	Convergence tolerance.
maxit	Maximum number of iterations.
b1	ADAM hyperparameter, default is 0.9.
b2	ADAM hyperparameter, default is 0.999.
e	ADAM hyperparameter, default is to 1e-8.

liqun730/gMIC documentation built on May 22, 2019, 11:52 p.m.