fmaglmnet: Frequentist model averaging of adaptive LASSO in GLM
In shaji948/FMAglm: Frequentist model averaging for generalised linear model

Description Usage Arguments Value Examples

Cross-validation frequentist model averaging for generalized linear models. The candidate models are selected by adaptive LASSO.

fma.glmnet( x, y, focus, family, nlambda = 100, nfolds = NULL, grouped,
            penalty.factor = NULL, force.nlambda,
            singleton.intercept.only = FALSE,
            type.measure = "mse", rule, solnptol, qp.scale,
            glmnetfit = NULL)

`x`	the matrix of regressors, excluding the intercept.
`y`	the vector of response variable.
`focus`	"mu" or "eta".
`family`	"binomial" or "poisson".
`nlambda`	the number of lambda. Default is 100.
`nfolds`	the number of folds in cross validation. Default equals the sample size.
`grouped`	same as grouped in the glmnet package.
`penalty.factor`	It will be passed to penalty.factor in the glmnet package. Default is a vector of 1, indicating same scaling. If penalty.factor is a numerical vector, it will be directly passed to cv.glment(). If penalty.factor="adaptiveLasso", the weight will be the reciporacal of the absolute value of the full model estimates.
`force.nlambda`	it is common that the user specify a value of nlambda, but the actual number of lambda is lower than nlambda. If force.nlambda = TRUE, the number of lambda values is forced to be nlambda by forming a better lambda sequence.
`singleton.intercept.only`	if TRUE, then the interpcet only model is also included in the singleton models. FALSE means that the intercept only model is left out.
`type.measure`	same as type.measure in the glmnet package.
`rule`	a vector that contains entries from "IMSE", "Singleton", and "Hybrid". "IMSE" averages the lambda sequence produced from the glmnet package, which aims to approximate the integrated MSE.
`solnptol`	tolerance used in solnp().
`qp.scale`	"max" or "1/n". "max" means that the raw symmetric matrix in QP is divided by the max number of the matrix. "1/n" means that it is divided by the sample size.
`glmnetfit`	The user can fit a glmnet object themselves and pass it to the function to save time. It is not provided, a glmnet object will be fitted in the function.

It returns a list of the following objects

fullcoef: estimated beta from the full model.
penalty.factor: the weights for each covariate.
IMSE: if "IMSE" belongs to rule, then it contains a list of the following objects.
- w: estimated weights.
- convergence: 0 or converged = converged.
- avecoef: averaged beta coefficients, if focus = "eta".
- elapsed: elapsed time in seconds, if focus = "eta".
- lambda: the lambda sequence.
- pos_def: whether QP is positive definite, if focus = "mu".
- coef: the matrix containing all estimated beta (excluding the full model), if focus = "mu".
Hybrid: if "Hybrid" belongs to rule, then it contains a list of the following objects.
- w: estimated weights.
- convergence: 0 or converged = converged.
- avecoef: averaged beta coefficients, if focus = "eta".
- elapsed: elapsed time in seconds, if focus = "eta".
- models: unique model structures.
- pos_def: whether QP is positive definite, if focus = "mu".
- coef: the matrix containing all estimated beta, if focus = "mu".
Singleton: if "Singleton" belongs to rule, then it contains a list of the following objects.
- w: estimated weights.
- convergence: 0 or converged = converged.
- avecoef: averaged beta coefficients, if focus = "eta".
- elapsed: elapsed time in seconds, if focus = "eta".
- pos_def: whether QP is positive definite, if focus = "mu".
- coef: the matrix containing all estimated beta, if focus = "mu".

data("ToyData_Binomial")
fma.glmnet(x = ToyBinomial[1 : 100, paste("x", 1 : 8, sep = "")],
           y = ToyBinomial[1 : 100, "y"],
           focus = "mu",
           family = "binomial",
           nlambda = 50,
           nfolds = 10,
           penalty.factor = "adaptiveLasso",
           grouped = FALSE,
           type.measure = "class",
           rule = c("IMSE", "Hybrid", "Singleton"),
           solnptol = 1e-08,
           force.nlambda = TRUE,
           qp.scale = "max",
           singleton.intercept.only = TRUE)

fma.glmnet(x = ToyBinomial[1 : 100, paste("x", 1 : 8, sep = "")],
           y = ToyBinomial[1 : 100, "y"],
           focus = "eta",
           family = "binomial",
           nlambda = 50,
           nfolds = 10,
           penalty.factor = "adaptiveLasso",
           grouped = FALSE,
           type.measure = "class",
           rule = c("IMSE", "Hybrid", "Singleton"),
           solnptol = 1e-08,
           force.nlambda = TRUE,
           qp.scale = "max",
           singleton.intercept.only = TRUE)

data("ToyData_Poisson")
fma.glmnet(x = ToyPoisson[1 : 100, paste("x", 1 : 8, sep = "")],
           y = ToyPoisson[1 : 100, "y"],
           focus = "mu",
           family = "poisson",
           nlambda = 50,
           nfolds = 10,
           penalty.factor = "adaptiveLasso",
           grouped = FALSE,
           type.measure = "mse",
           rule = c("IMSE", "Hybrid", "Singleton"),
           solnptol = 1e-08,
           force.nlambda = TRUE,
           qp.scale = "max",
           singleton.intercept.only = TRUE)

fma.glmnet(x = ToyPoisson[1 : 100, paste("x", 1 : 8, sep = "")],
           y = ToyPoisson[1 : 100, "y"],
           focus = "eta",
           family = "poisson",
           nlambda = 50,
           nfolds = 10,
           penalty.factor = "adaptiveLasso",
           grouped = FALSE,
           type.measure = "mse",
           rule = c("IMSE", "Hybrid", "Singleton"),
           solnptol = 1e-08,
           force.nlambda = TRUE,
           qp.scale = "max",
           singleton.intercept.only = TRUE)