Nothing
###############################################################
###
### The following function creates a connection list
### for the 2-dimensional Fused Lasso Signal Approximator
###
###############################################################
connListTwoDimensions = function(dimensions)
{
dimensions=as.integer(dimensions) # need integer
if(!is.vector(dimensions) || length(dimensions)>2 || (!is.numeric(dimensions) && !is.integer(dimensions)))
{
stop("dimensions has to be a numeric vector of length 2")
}
if(dimensions[1]<2 || dimensions[2]<2)
{
stop("Each dimension has to have at least length 2")
}
conn = .Call("conn2Dim", dimensions, PACKAGE="flsa")
### make the list
nodeNumbers = matrix(0:(dimensions[1]*dimensions[2]-1), nrow=dimensions[1])
names(conn)=nodeNumbers
class(conn) = "connListObj"
return(conn)
}
##################################################################
###
### checks an object if it conforms to the specifications of a connection List object
###
##################################################################
is.connListObj = function(obj)
{
if(! isa(obj, "connListObj"))
{
stop("Object does not have the right class")
}
### check that every element of the conn part is an integer vector or NULL
for(i in 1:length(obj))
{
if(!(is.null(obj[[i]]) || is.integer(obj[[i]])))
{
stop("All elements of the conn part have to be null or integer vectors")
}
}
### check that all node numbers that occur in the conn part are eligible
nodeNumbers = 0:(length(obj))
for(i in 1:length(obj))
{
if(sum(!is.element(obj[[i]], nodeNumbers))>0)
{
stop(paste("Node",i,"has a connection to a non-existing node"))
}
}
return(TRUE)
}
######################################################################
###
### Flsa function that works top down
###
###
######################################################################
flsaTopDown <- function(y, lambda1=0, groups=1:length(y), lambda2=NULL) {
## check the input variables
if(!is.vector(y) | !is.numeric(y)) {
stop("y has to be a numeric vector")
}
if(!is.double(y)) {
y <- as.double(y)
}
if(is.null(groups)) {
groups = numeric(0)
}
if(!is.vector(groups)) {
stop("Groups has to be a vector")
}
groups = as.integer(groups)
if(length(groups)>0 && min(groups)<1) {
warning("Number of groups should be >= 1")
}
if(length(groups)>0 && max(groups)> length(y)) {
warning("Number of groups should be <= length(y)")
}
if(is.null(lambda2)) {
lambda2 = numeric(0)
}
if(!is.vector(lambda2)) {
stop("lambda2 has to be a vector")
}
lambda2 = as.numeric(lambda2)
if(length(lambda2)>0 && min(lambda2) < 0) {
warning("Lambda2 should be greater than 0")
}
if(lambda1<0) {
stop("Lambda1 has to be >=0")
}
res = .Call("FLSATopDown", y, groups, lambda2, PACKAGE="flsa")
if(lambda1 > 0) {
res$Solution = softThresholding(res$Solution, lambda1)
}
## make a few small changes to the results
res$Solution = t(res$Solution)
rownames(res$Solution) = res$Lambdas
## remove duplicate values of lambda
lambda.not.duplicated <- !duplicated(round(res$Lambdas, 12))
res$Solution <- res$Solution[lambda.not.duplicated, , drop=FALSE]
res$Lambdas <- res$Lambdas[lambda.not.duplicated]
res$isBreakpoint <- res$isBreakpoint[lambda.not.duplicated]
return(res)
}
################################################################
###
### This is the main interface function for the FLSA
### it calls the right C++ function depending on whether it is a 2-dimensional
### or a 1-dimensional problem
###
################################################################
flsa = function(y, lambda1=0, lambda2=NULL, connListObj = NULL, splitCheckSize=1e9, verbose=FALSE, thr=10e-10, maxGrpNum=4*length(y)) {
splitCheckSize=as.integer(splitCheckSize)
if(is.null(connListObj)) { # call the appropriate method depending on dimension of y
if(is.vector(y)) { ## call the basic FLSA
if(!is.double(y)) { # y has to be REAL before passing it to C
y <- as.double(y)
}
solObj = .Call("FLSA",y, PACKAGE="flsa")
if(!is.null(lambda2)) {
resLambda1Is0 = FLSAOneDimExplicitSolution(solObj, lambda2)
if(lambda1!=0) {
res = softThresholding(resLambda1Is0, lambda1)
return(res)
}
else {
return(resLambda1Is0)
}
}
else { # no lambda2 value, so just return solution object
return(solObj)
}
}
else if(is.matrix(y)) { ## call the 2-dimensional FLSA
connListObj = connListTwoDimensions(dim(y))
## check that everything is ok with lambda2
if(!is.null(lambda2)) {
lambda2 = checkLambda2(lambda2)
res=.Call("FLSAGeneralMain", connListObj, as.double(y), lambda2, splitCheckSize, verbose,
thr, as.integer(maxGrpNum), PACKAGE="flsa")
## format the result in 2 dimensions
res= array(res, dim=c(length(lambda2), dim(y)))
## reset the names of the dimensions
myDimNames = list(lambda2, 1:(dim(y)[1]), 1:(dim(y)[2]))
dimnames(res) = myDimNames
## take lambda1 into account if necessary
if(lambda1!=0) {
res = softThresholding(res, lambda1)
}
}
else {
res=.Call("FLSAGeneralMain", connListObj, as.double(y), lambda2, splitCheckSize, verbose,thr, as.integer(maxGrpNum), PACKAGE="flsa")
}
return(res)
}
}
else {## call the general FLSA with the connection list
if(!is.null(lambda2)) {
lambda2=checkLambda2(lambda2)
}
if(length(connListObj)!=length(y)) {
stop("y has to have the same number of nodes as connListObj")
}
res=.Call("FLSAGeneralMain", connListObj, as.double(y), lambda2, splitCheckSize, verbose, thr, as.integer(maxGrpNum), PACKAGE="flsa")
## take lambda1 into account if necessary
if(!is.null(lambda2) && (lambda1!=0)) {
res = softThresholding(res, lambda1)
}
return(res)
}
}
##################################################################
###
### if an object with the solution tree was returned, this object can
### be used to generate explicit solutions
###
##################################################################
flsaGetSolution = function(solObj, lambda1=0, lambda2=NULL, dim=NULL)
{
### check that lambda2 is ok
if(is.null(lambda2)) {
stop("lambda2 has to be specified")
}
lambda2 = checkLambda2(lambda2)
if(isa(solObj, "FLSA"))
{
res = FLSAOneDimExplicitSolution(solObj, lambda2)
if(lambda1!=0)
{
res = softThresholding(res, lambda1)
}
return(res)
}
else if(isa(solObj, "FLSAGeneral"))
{
### calculate the explicit solution
nodes = as.integer(which(solObj$InitialNodeMap>=0)-1)
res = .Call("FLSAGeneralExplicitSolution",solObj,nodes, lambda2, PACKAGE="flsa")
### format in the right way if necessary
if(!is.null(dim))
{
### check that the dimensions are the same as the number of nodes
if(prod(dim)!=length(nodes))
{
stop("Dimensions are not compatible with solObj")
}
res = array(res, dim=c(length(lambda2),dim))
}
### take a look at lambda1
if(lambda1!=0)
{
res = softThresholding(res,lambda1)
}
return(res)
}
else
{
stop("solObj is not of class FLSA or FLSAGeneral")
}
}
##################################################################
###
### function that checks if lambda2 has the right format
###
##################################################################
checkLambda2 = function(lambda2)
{
### check that lambda2 is a numeric vector, increasing and only non-negative elements
if(!is.numeric(lambda2))
{
stop("lambda2 has to be a numeric vector")
}
### make sure that lambda2 is non-negative
if(sum(lambda2<0)>0)
{
stop("lambda2 has to be non-negative")
}
### sort lambda2, make sure all are unique
lambda2 = sort(unique(lambda2))
return(lambda2)
}
####################################################################
###
### get an explicit solution from a FLSA solution object
### this is an internal function and will be hidden in the package
###
####################################################################
FLSAOneDimExplicitSolution = function(solObj, lambda2)
{
lambda2 = checkLambda2(lambda2)
return(.Call("FLSAexplicitSolution",solObj, lambda2, PACKAGE="flsa"))
}
###################################################################
###
### Given a matrix with the solutions for lambda2, give back a three-dimensional
### array with soltions for varying lambda1
### This is an internal function
###
###################################################################
softThresholding = function(solMat, lambda1)
{
### check that lambda1 is a numeric vector, increasing and only non-negative elements
if(!is.numeric(lambda1))
{
stop("lambda1 has to be a numeric vector")
}
### make sure that lambda2 is non-negative
if(sum(lambda1<0)>0)
{
stop("lambda1 has to be non-negative")
}
### sort lambda1, make sure all are unique
lambda1 = sort(unique(lambda1))
### generate the new data array and set dimension and names right
oldDim = dim(solMat)
newDim = c(length(lambda1), oldDim)
oldDimNames = dimnames(solMat)
if(is.null(oldDimNames))
{
oldDimNames = vector("list",2)
}
newDimNames = c(list(lambda1), oldDimNames)
res = array(dim=newDim, dimnames = newDimNames)
### fill in the soft-thresholded data
if(length(oldDim)==2)
{
for(i in 1:length(lambda1))
{
foo = abs(solMat)-lambda1[i]
foo[foo<0]=0
res[i,,] = sign(solMat) * foo
}
}
else if(length(oldDim)==3)
{
for(i in 1:length(lambda1))
{
foo = abs(solMat)-lambda1[i]
foo[foo<0]=0
res[i,,,] = sign(solMat) * foo
}
}
else
{
stop("Wrong dimension of solMat; please inform the maintainer of this package")
}
return(res)
}
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.