lme1en: lme1en

In oyvble/lme1en: Elastic Net With Random Intercept

Description

Fitting Elastic Net with random intercept

Usage

lme1en(y, X, batch, lambda = 1, alpha = 0.5, rho = 0, beta = NULL,
  glmnetPenalty = TRUE, glmnetWarmup = TRUE, maxit = 10000,
  toler = 0.001, verbose = FALSE)

Arguments

y	response vector. Should be standardized before input (use scale).
X	Design matrix belonging to fixed effects coefficients (beta). Should be standardized before input (use scale).
batch	factor with batch effect names (vector for each observations)
lambda	numeric, penalty levels for fixed effects betas
alpha	numeric, penalty levels for fixed effects betas (balancing LASSO/RIDGE erros)
rho	proportion of variation explained by batch effect
beta	numeric, initial values of the beta coefficients (using glmnet or marginal estimates if not provided)
glmnetPenalty	boolean, whether to use the original peanalty (FALSE) or the glmnet penalty (TRUE)
glmnetWarmup	boolean, whether to use glmnet beta-estimates as warmup (if not marginals are used)
maxit	maximum number of iterations (i.e. forloops) in the coordinate decent algorithm
toler	tolerance level of beta changes for each iterations (similar to 'thresh' in glmnet)
verbose	boolean, show progress Default: FALSE

Details

The model is fitted using the coordinate decent algorithm (exact solutions) The extended model includes the rho parameter (proportion of total variation) as argument. Calls a algorithm (iterate) in C which uses exact solutions of marginal beta's based on formula x = sgn(c/a)(|c/a| - b/a)_+, where ax + b*sgn(x) = c The function assume that the response y and design matrix X are centralized (no intercept returned from function)

Value

fitted beta values

Examples

## Not run:  
dat = genData(seed=1)
bhat = lme1en(y=dat$y,X=dat$X,batch=dat$batch, rho=0.3,lambda=0.1,alpha=0.5,verbose=TRUE)

## End(Not run)

oyvble/lme1en documentation built on April 30, 2020, 2:41 p.m.