sureStep {qtlmt} | R Documentation |
SURE model selection
Description
Select a SURE model via model selection.
Usage
sureAdd1(object, y, x, range=NULL, iter=10000, tol=1e-12, ext=FALSE)
sureDrop1(object, y, x, range=NULL, iter=10000, tol=1e-12, ext=FALSE)
sureStep(object, y, x, cv, direction=c("both","backward","forward"),
range=NULL, iter=10000, steps=1000, tol=1e-12, ext=FALSE)
surStep(y, x, v, lower, upper, k, direction=c("both","backward",
"forward"), iter=10000, max.terms=200, steps=1000, tol=1e-12)
Arguments
object |
initial model in model search; can be an object of |
y |
a n by p matrix, whose columns are dependent variables. |
x |
a n by m matrix, whose columns are predictor variables to select from. |
range |
a list; range[[j]] indicates which x's (all if |
cv |
critical value used in the likelihood ratio test to determine adding/dropping a term. |
v |
a list; v[[j]] indicates which x's to start with for y[,j]. |
lower |
a list in the format of |
upper |
a list in the format of |
k |
penalty, 0 if missing or <0. |
direction |
forward selection, backward elimination or stepwise. |
iter |
maximum number of iterations in a numerical process to estimate model parameters. |
max.terms |
maximum number of terms in the final model. |
steps |
maximum number of search steps. |
tol |
convergence tolerance. |
ext |
whether to perform an extensive search for an "optimal" model with the same number of QTL per phenotype. |
Value
a list with the following components:
loglik |
log-likelihood of the model |
b |
estimates of model coefficients |
sigma |
estimates of residual variance-covariance |
v |
variables selected in the model |
Examples
data(etrait)
x<- as.matrix(mdat-1/2)
y<- as.matrix(traits)[,1:3]
v<- list()
upper<- list()
for(k in 1:ncol(y)){
v[[k]]<- numeric(0)
upper[[k]]<- 1:ncol(x)
}
## Not run:
o1<- surStep(y, x, v=v, upper=upper, k=19)
o2<- sureStep(o1, y, x, cv=50, ext=FALSE)
# search for optimal model of o1
o3<- sureStep(o1, y, x, cv=Inf, direction="forward", ext=TRUE)
## End(Not run)