LoglikPen: Compute the penalized log likelihood for GLM with MIC penalty...

In xgsu/glmMIC: Sparse Estimation of Generalized Linear Models (GLM) via Minimum approximated Inforamtion Criterion (MIC)

Description

Compute the penalized log likelihood for GLM with MIC penalty (Self-Written)

Usage

LoglikPen(beta, I.preselect = NULL, preselect.beta0 = FALSE, X, y, lambda,
  a, family = "gaussian")

Arguments

beta	A p-dimensional vector containing the regression ceofficients.
I.preselect	A 0-1 indicator vector of same length as beta showing preselected variables (value 0 for no penalty) that will not be penalized.
preselect.beta0	Indicator of whether or not the intercept is pre-selected. Default is FALSE.
X	An n by p design matrix.
y	The n by 1 response vector
lambda	The penalty parameter euqals either 2 in AIC or ln(n) in BIC (by default). It can be specified as any value of the user's own choice.
a	The scale parameter in the hyperbolic tangent function of the MIC penalty. By default, a = 50.
family	a description of the error distribution. To use this function, family needs to be one of "gaussian",

Details

This function is much faster than LoglikPenGLM, but it is only applicable for Gaussian linear regression, logistic regression, and loglinear or Poisson regression models. To take advantage, sepcify family as family="gaussian", family="binomial", or family="poisson" only.

Value

The value of the penalized log likelihood function evaluated at beta.

References

• Su, X. (2015). Variable selection via subtle uprooting. Journal of Computational and Graphical Statistics, 24(4): 1092–1113. URL http://www.tandfonline.com/doi/pdf/10.1080/10618600.2014.955176

• Su, X., Fan, J., Levine, R. A., Nunn, M. E., and Tsai, C.-L. (2016+). Sparse estimation of generalized linear models via approximated information criteria. Submitted, Statistica Sinica.

See Also

glm, family

xgsu/glmMIC documentation built on May 4, 2019, 1:06 p.m.