Usage
correg(X = X, Y = Y, Z = NULL, B = NULL, compl = TRUE, expl = TRUE,
pred = TRUE, prednew = FALSE, select = "lar", criterion = c("MSE",
"BIC"), X_test = NULL, Y_test = NULL, intercept = TRUE, K = 10,
groupe = NULL, Amax = NULL, lambda = 1, returning = TRUE,
nbalter = 10, deltamin = 0.01, alpha = NULL, g = 5)Arguments
B
the (p+1)xp matrix associated to Z and that
contains the parameters of the sub-regressions
lambda
parameter for elasticnet (quadratic
penalty)
X
the data matrix (covariates) without the
intercept
Y
The response variable vector
Z
The structure (adjacency matrix) between the
covariates
compl
boolean to decide if the complete modele is
computed
expl
boolean to decide if the explicative model is
in the output
pred
boolean to decide if the predictive model is
computed
select
selection method in
("lar","lasso","forward.stagewise","stepwise",
"elasticnet",
"NULL","ridge","adalasso","clere","spikeslab")
criterion
the criterion used to compare the
models
K
the number of clusters for cross-validation
groupe
a vector to define the groups used for
cross-validation (to obtain a reproductible result)
Amax
the maximum number of covariates in the final
model
returning
boolean : second predictive step
(selection on I1 knowing I2 coefficients)
Y_test
response for the validation sample
intercept
boolean. If FALSE intercept will be set
to 0 in each model.
alpha
Coefficients of the explicative model to
coerce the predictive step. if not NULL explicative step
is not computed.
prednew
alternate optimisation for predictive
nbalter
number of alternance for prednew
deltamin
criterion to stop alternance
g
number of group of variables for clere