Nothing
R/PMD.R
In PMA: Penalized Multivariate Analysis
Defines functions
print.PMDL1FL.cv
plot.PMDL1FL.cv
PMDL1FL.cv
plot.PMDL1L1.cv
print.PMDL1L1.cv
print.SPC.cv
plot.SPC.cv
SPC.cv
PMDL1L1.cv
MultiSMD
MultiSMDOrth
print.PMDL1FL
print.PMDL1L1
PMDL1L1
PMD.cv
PMD
print.SPC
SPC
CheckPMDV
SMDOrth
SMD
BinarySearch
safesvd
l2n
mean_na
soft
Documented in
plot.SPC.cv
PMD
PMD.cv
print.SPC
print.SPC.cv
SPC
SPC.cv
soft <- function(x,d){
return(sign(x)*pmax(0, abs(x)-d))
}
mean_na <- function(vec){
return(mean(vec[!is.na(vec)]))
}
l2n <- function(vec){
a <- sqrt(sum(vec^2))
if(a==0) a <- .05
return(a)
}
safesvd <- function(x){
i <- 1
out <- try(svd(x), silent=TRUE)
while(i<10 && inherits(out, "try-error")){
out <- try(svd(x), silent=TRUE)
i <- i+1
}
if(inherits(out, "try-error")) out <- svd(matrix(rnorm(nrow(x)*ncol(x)), ncol=ncol(x)))
return(out)
}
BinarySearch <- function(argu,sumabs){
if(l2n(argu)==0 || sum(abs(argu/l2n(argu)))<=sumabs) return(0)
lam1 <- 0
lam2 <- max(abs(argu))-1e-5
iter <- 1
while(iter < 150){
su <- soft(argu,(lam1+lam2)/2)
if(sum(abs(su/l2n(su)))<sumabs){
lam2 <- (lam1+lam2)/2
} else {
lam1 <- (lam1+lam2)/2
}
if((lam2-lam1)<1e-6) return((lam1+lam2)/2)
iter <- iter+1
}
warning("Didn't quite converge")
return((lam1+lam2)/2)
}
SMD <- function(x, sumabsu, sumabsv, niter=20,trace=TRUE, v, upos, uneg, vpos, vneg){
# This gets a single factor. Do MultiSMD to get multiple factors.
nas <- is.na(x)
v.init <- v
xoo <- x
if(sum(nas)>0) xoo[nas] <- mean(x[!nas])
oldv <- rnorm(ncol(x))
for(iter in 1:niter){
if(sum(abs(oldv-v))>1e-7){
oldv <- v
if(trace) cat(iter,fill=F)
# update u #
argu <- xoo%*%v
if(upos) argu <- pmax(argu,0)
if(uneg) argu <- pmin(argu,0)
lamu <- BinarySearch(argu,sumabsu)
su <- soft(argu,lamu)
u <- matrix(su/l2n(su),ncol=1)
# done updating u #
# update v #
argv <- t(u)%*%xoo
if(vpos) argv <- pmax(argv,0)
if(vneg) argv <- pmin(argv,0)
lamv <- BinarySearch(argv, sumabsv)
sv <- soft(argv,lamv)
v <- matrix(sv/l2n(sv),ncol=1)
# done updating v #
}
}
d <- as.numeric(t(u)%*%(xoo%*%v))
if(trace) cat(fill=TRUE)
return(list(d=d, u=u, v=v, v.init=v.init))
}
SMDOrth <- function(x, us, sumabsv=NULL, niter=20, trace=TRUE,v, vpos, vneg){
# Gets the next u for sparse PCA, using Trevor's method of requiring this new u to be orthog
# to all previous us (contained in us)
nas <- is.na(x)
v.init <- v
xoo <- x
if(sum(nas)>0) xoo[nas] <- mean(x[!nas])
u <- rnorm(nrow(x))
oldu <- rnorm(nrow(x))
oldv <- rnorm(ncol(x))
for(iter in 1:niter){
if(sum(abs(oldu-u))>1e-6 || sum(abs(oldv-v))>1e-6){
oldu <- u
oldv <- v
if(trace) cat(iter,fill=F)
# update u #
argu <- xoo%*%v
numer <- lsfit(y=argu, x=us, intercept=FALSE)$res
u <- numer/l2n(numer)
# done updating u #
# update v #
argv <- t(u)%*%xoo
if(vpos) argv <- pmax(argv,0)
if(vneg) argv <- pmin(argv,0)
lamv <- BinarySearch(argv, sumabsv)
sv <- soft(argv,lamv)
v <- matrix(sv/l2n(sv),ncol=1)
# done updating v #
}
}
d <- as.numeric(t(u)%*%(xoo%*%v))
if(trace) cat(fill=TRUE)
return(list(d=d,u=u,v=v, v.init=v.init))
}
CheckPMDV <- function(v,x,K){
if(!is.null(v) && is.matrix(v) && ncol(v)>=K){
v <- matrix(v[,1:K], ncol=K)
} else if(ncol(x)>nrow(x)){
x[is.na(x)] <- mean_na(x)
v <- matrix(t(x)%*%(safesvd(x%*%t(x))$v[,1:K]),ncol=K)
if(sum(is.na(v))>0) v <- matrix(safesvd(x)$v[,1:K], ncol=K)
v <- sweep(v,2,apply(v, 2, l2n), "/")
if(sum(is.na(v))>0) stop("some are NA")
} else if (ncol(x)<=nrow(x)){
x[is.na(x)] <- mean_na(x)
v <- matrix(safesvd(t(x)%*%x)$v[,1:K],ncol=K)
}
return(v)
}
#' Perform sparse principal component analysis
#'
#' Performs sparse principal components analysis by applying PMD to a data
#' matrix with lasso ($L_1$) penalty on the columns and no penalty on the rows.
#'
#' PMD(x,sumabsu=sqrt(nrow(x)), sumabsv=3, K=1) and SPC(x,sumabsv=3, K=1) give
#' the same result, since the SPC method is simply PMD with an L1 penalty on
#' the columns and no penalty on the rows.
#'
#' In Witten, Tibshirani, and Hastie (2008), two methods are presented for
#' obtaining multiple factors for SPC. The methods are as follows:
#'
#' (1) If one has already obtained factors $k-1$ factors then oen can compute
#' residuals by subtracting out these factors. Then $u_k$ and $v_k$ can be
#' obtained by applying the SPC/PMD algorithm to the residuals.
#'
#' (2) One can require that $u_k$ be orthogonal to $u_i$'s with $i<k$; the
#' method is slightly more complicated, and is explained in WT&H(2008).
#'
#' Method 1 is performed by running SPC with option orth=FALSE (the default)
#' and Method 2 is performed using option orth=TRUE. Note that Methods 1 and 2
#' always give identical results for the first component, and often given quite
#' similar results for later components.
#'
#' @aliases SPC print.SPC
#' @param x Data matrix of dimension $n x p$, which can contain NA for missing
#' values. We are interested in finding sparse principal components of
#' dimension $p$.
#' @param sumabsv How sparse do you want v to be? This is the sum of absolute
#' values of elements of v. It must be between 1 and square root of number of
#' columns of data. The smaller it is, the sparser v will be.
#' @param niter How many iterations should be performed. It is best to run at
#' least 20 of so. Default is 20.
#' @param K The number of factors in the PMD to be returned; default is 1.
#' @param v The first right singular vector(s) of the data. (If missing data is
#' present, then the missing values are imputed before the singular vectors are
#' calculated.) v is used as the initial value for the iterative PMD($L_1$,
#' $L_1$) algorithm. If x is large, then this step can be time-consuming;
#' therefore, if PMD is to be run multiple times, then v should be computed
#' once and saved.
#' @param trace Print out progress as iterations are performed? Default is
#' TRUE.
#' @param orth If TRUE, then use method of Section 3.2 of Witten, Tibshirani
#' and Hastie (2008) to obtain multiple sparse principal components. Default is
#' FALSE.
#' @param center Subtract out mean of x? Default is TRUE
#' @param cnames An optional vector containing a name for each column.
#' @param vpos Constrain the elements of v to be positive? TRUE or FALSE.
#' @param vneg Constrain the elements of v to be negative? TRUE or FALSE.
#' @param compute.pve Compute percent variance explained? Default TRUE. If not
#' needed, then choose FALSE to save time.
#' @return \item{u}{u is output. If you asked for multiple factors then each
#' column of u is a factor. u has dimension nxK if you asked for K factors.}
#' \item{v}{v is output. These are the sparse principal components. If you
#' asked for multiple factors then each column of v is a factor. v has
#' dimension pxK if you asked for K factors.} \item{d}{d is output; it is the
#' diagonal of the matrix $D$ in the penalized matrix decomposition. In the
#' case of the rank-1 decomposition, it is given in the formulation
#' $||X-duv'||_F^2$ subject to $||u||_1 <= sumabsu$, $||v||_1 <= sumabsv$.
#' Computationally, $d=u'Xv$ where $u$ and $v$ are the sparse factors output by
#' the PMD function and $X$ is the data matrix input to the PMD function.}
#' \item{prop.var.explained}{A vector containing the proportion of variance
#' explained by the first 1, 2, ..., K sparse principal components obtaineds.
#' Formula for proportion of variance explained is on page 20 of Shen & Huang
#' (2008), Journal of Multivariate Analysis 99: 1015-1034.} \item{v.init}{The
#' first right singular vector(s) of the data; these are returned to save on
#' computation time if PMD will be run again.} \item{meanx}{Mean of x that was
#' subtracted out before SPC was performed.}
#' @seealso \link{SPC.cv}, \link{PMD}, \link{PMD.cv}
#' @references
#' Witten D. M., Tibshirani R., and Hastie, T. (2009)
#' \emph{A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis}, \emph{Biostatistics, Gol 10 (3), 515-534, Jul 2009}\cr
#' @examples
#'
#'
#' # A simple simulated example
#' #NOT RUN
#' #set.seed(1)
#' #u <- matrix(c(rnorm(50), rep(0,150)),ncol=1)
#' #v <- matrix(c(rnorm(75),rep(0,225)), ncol=1)
#' #x <- u%*%t(v)+matrix(rnorm(200*300),ncol=300)
#' ## Perform Sparse PCA - that is, decompose a matrix w/o penalty on rows
#' ## and w/ L1 penalty on columns
#' ## First, we perform sparse PCA and get 4 components, but we do not
#' ## require subsequent components to be orthogonal to previous components
#' #out <- SPC(x,sumabsv=3, K=4)
#' #print(out,verbose=TRUE)
#' ## We could have selected sumabsv by cross-validation, using function SPC.cv
#' ## Now, we do sparse PCA using method in Section 3.2 of WT&H(2008) for getting
#' ## multiple components - that is, we require components to be orthogonal
#' #out.orth <- SPC(x,sumabsv=3, K=4, orth=TRUE)
#' #print(out.orth,verbose=TRUE)
#' #par(mfrow=c(1,1))
#' #plot(out$u[,1], out.orth$u[,1], xlab="", ylab="")
#' ## Note that the first components w/ and w/o orth option are identical,
#' ## since the orth option only affects the way that subsequent components
#' ## are found
#' #print(round(t(out$u)%*%out$u,4)) # not orthogonal
#' #print(round(t(out.orth$u)%*%out.orth$u,4)) # orthogonal
#' #
#' ## Use SPC.cv to choose tuning parameters:
#' #cv.out <- SPC.cv(x)
#' #print(cv.out)
#' #plot(cv.out)
#' #out <- SPC(x, sumabsv=cv.out$bestsumabsv)
#' #print(out)
#' ## or we could do
#' #out <- SPC(x, sumabsv=cv.out$bestsumabsv1se)
#' #print(out)
#' #
#' #
#' @export SPC
SPC <- function(x, sumabsv=4, niter=20, K=1, orth=FALSE, trace=TRUE, v=NULL, center=TRUE, cnames=NULL, vpos=FALSE, vneg=FALSE, compute.pve=TRUE){
if(vpos&&vneg) stop("Cannot constrain elements to be positive AND negative.")
out <- PMDL1L1(x,sumabsu=sqrt(nrow(x)), sumabsv=sumabsv, niter=niter,K=K,orth=orth,trace=trace,v=v,center=center,cnames=cnames, upos=FALSE, uneg=FALSE, vpos=vpos, vneg=vneg)
if(compute.pve){
# Calculate percent variance explained, using definition on page 7 of Shen and Huang (2008) Journal of Multivariate Analysis vol 99: 1015-1034
v <- matrix(out$v, ncol=K)
ve <- NULL # variance explained
xfill <- x
if(center) xfill <- x-mean_na(x)
xfill[is.na(x)] <- mean_na(xfill)
for(k in 1:K){
vk <- matrix(v[,1:k], ncol=k)
xk <- xfill%*%vk%*%solve(t(vk)%*%vk)%*%t(vk)
svdxk <- svd(xk)
ve <- c(ve, sum(svdxk$d^2))
}
pve <- ve/sum(svd(xfill)$d^2) # proportion of variance explained
out$prop.var.explained <- pve
}
out$vpos <- vpos
out$vneg <- vneg
class(out) <- "SPC"
return(out)
}
#' @method print SPC
#' @export
print.SPC <- function(x,verbose=FALSE,...){
cat("Call: ")
dput(x$call)
cat("\n\n")
cat("Num non-zero v's: ", apply(x$v!=0, 2, sum), "\n")
cat("Sumabsv used: ", x$sumabsv, "\n")
cat("Proportion of variance explained by first k components: \n")
tab <- cbind(paste(paste("First ", sep="", 1:length(x$prop.var.explained)), sep="", " Components"), round(x$prop.var.explained, 3))
dimnames(tab) <- list(1:length(x$prop.var.explained), c("", "Prop. Var. Explained"))
print(tab, quote=FALSE, sep="\t")
if(!is.null(x$meanx)) cat("Subtracted out mean of x: ", x$meanx, "\n")
if(verbose){
for(k in 1:x$K){
cat("\n Component ", k, ":\n")
v <- x$v[,k]
if(is.null(x$cnames)) x$cnames <- 1:length(v)
cat(fill=T)
vs <- cbind(x$cnames[v!=0], round(v[v!=0],3))
dimnames(vs) <- list(1:sum(v!=0), c("Feature Name", "Feature Weight"))
print(vs, quote=FALSE, sep="\t")
}
}
if(x$vpos) cat("Elements of v constrained to be positive.", fill=TRUE)
if(x$vneg) cat("Elements of v constrained to be negative.", fill=TRUE)
}
#' Get a penalized matrix decomposition for a data matrix.
#'
#' Performs a penalized matrix decomposition for a data matrix. Finds factors u
#' and v that summarize the data matrix well. u and v will both be sparse, and
#' v can optionally also be smooth.
#'
#' The criterion for the PMD is as follows: we seek vectors $u$ and $v$ such
#' that $u'Xv$ is large, subject to $||u||_2=1, ||v||_2=1$ and additional
#' penalties on $u$ and $v$. These additional penalties are as follows: If type
#' is "standard", then lasso ($L_1$) penalties (promoting sparsity) are placed
#' on u and v. If type is "ordered", then lasso penalty is placed on u and a
#' fused lasso penalty (promoting sparsity and smoothness) is placed on v.
#'
#' If type is "standard", then arguments sumabs OR sumabsu&sumabsv are used. If
#' type is "ordered", then sumabsu AND lambda are used. Sumabsu is the bound of
#' absolute value of elements of u. Sumabsv is bound of absolute value of
#' elements of v. If sumabs is given, then sumabsu is set to
#' sqrt(nrow(x))*sumabs and sumabsv is set to sqrt(ncol(x))*sumabs. $lambda$ is
#' the parameter for the fused lasso penalty on v when type is "ordered":
#' $lambda(||v||_1 + sum_j |v_j - v_(j-1))$.
#'
#' @param x Data matrix of dimension $n x p$, which can contain NA for missing
#' values.
#' @param type "standard" or "ordered": Do we want v to simply be sparse, or
#' should it also be smooth? If the columns of x are ordered (e.g. CGH spots
#' along a chromosome) then choose "ordered". Default is "standard". If
#' "standard", then the PMD function will make use of sumabs OR
#' sumabsu&sumabsv. If "ordered", then the function will make use of sumabsu
#' and lambda.
#' @param sumabs Used only if type is "standard". A measure of sparsity for u
#' and v vectors, between 0 and 1. When sumabs is specified, and sumabsu and
#' sumabsv are NULL, then sumabsu is set to $sqrt(n)*sumabs$ and sumabsv is set
#' to $sqrt(p)*sumabs$. If sumabs is specified, then sumabsu and sumabsv should
#' be NULL. Or if sumabsu and sumabsv are specified, then sumabs should be
#' NULL.
#' @param sumabsu Used for types "ordered" AND "standard". How sparse do you
#' want u to be? This is the sum of absolute values of elements of u. It must
#' be between 1 and the square root of the number of rows in data matrix. The
#' smaller it is, the sparser u will be.
#' @param sumabsv Used only if type is "standard". How sparse do you want v to
#' be? This is the sum of absolute values of elements of v. It must be between
#' 1 and square root of number of columns of data. The smaller it is, the
#' sparser v will be.
#' @param lambda Used only if type is "ordered". This is the tuning parameter
#' for the fused lasso penalty on v, which takes the form $lambda ||v||_1 +
#' lambda |v_j - v_(j-1)|$. $lambda$ must be non-negative. If NULL, then it is
#' chosen adaptively from the data.
#' @param niter How many iterations should be performed. It is best to run at
#' least 20 of so. Default is 20.
#' @param K The number of factors in the PMD to be returned; default is 1.
#' @param v The first right singular vector(s) of the data. (If missing data is
#' present, then the missing values are imputed before the singular vectors are
#' calculated.) v is used as the initial value for the iterative PMD algorithm.
#' If x is large, then this step can be time-consuming; therefore, if PMD is to
#' be run multiple times, then v should be computed once and saved.
#' @param trace Print out progress as iterations are performed? Default is
#' TRUE.
#' @param center Subtract out mean of x? Default is TRUE.
#' @param chrom If type is "ordered", then this gives the option to specify
#' that some columns of x (corresponding to CGH spots) are on different
#' chromosomes. Then v will be sparse, and smooth *within* each chromosome but
#' not *between* chromosomes. Length of chrom should equal number of columns of
#' x, and each entry in chrom should be a number corresponding to which
#' chromosome the CGH spot is on.
#' @param rnames An optional vector containing a name for each row of x.
#' @param cnames An optional vector containing a name for each column of x.
#' @param upos Constrain the elements of u to be positive? TRUE or FALSE.
#' @param uneg Constrain the elements of u to be negative? TRUE or FALSE.
#' @param vpos Constrain the elements of v to be positive? TRUE or FALSE.
#' Cannot be used if type is "ordered".
#' @param vneg Constrain the elements of v to be negative? TRUE or FALSE.
#' Cannot be used if type is "ordered."
#' @return \item{u}{u is output. If you asked for multiple factors then each
#' column of u is a factor. u has dimension nxK if you asked for K factors.}
#' \item{v}{v is output. If you asked for multiple factors then each column of
#' v is a factor. v has dimension pxK if you asked for K factors.} \item{d}{d
#' is output. Computationally, $d=u'Xv$ where $u$ and $v$ are the sparse
#' factors output by the PMD function and $X$ is the data matrix input to the
#' PMD function. When K=1, the residuals of the rank-1 PMD are given by $X -
#' duv'$.} \item{v.init}{The first right singular vector(s) of the data; these
#' are returned to save on computation time if PMD will be run again.}
#' \item{meanx}{Mean of x that was subtracted out before PMD was performed.}
#' @seealso \link{PMD.cv}, \link{SPC}
#' @references
#' Witten D. M., Tibshirani R., and Hastie, T. (2009)
#' \emph{A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis}, \emph{Biostatistics, Gol 10 (3), 515-534, Jul 2009}\cr
#' @examples
#'
#'
#' # Try PMD with L1 penalty on rows and columns: type="standard"
#' # A simple simulated example
#' set.seed(1)
#' # Our data is a rank-one matrix, plus noise. The underlying components
#' # contain 50 and 75 non-zero elements, respectively.
#' u <- matrix(c(rnorm(50), rep(0,150)),
#' ncol=1)
#' v <- matrix(c(rnorm(75),rep(0,225)), ncol=1)
#' x <- u%*%t(v)+
#' matrix(rnorm(200*300),ncol=300)
#' # We can use cross-validation to try to find optimal value of sumabs
#' cv.out <- PMD.cv(x, type="standard", sumabss=seq(0.1, 0.6, len=20))
#' print(cv.out)
#' plot(cv.out)
#' # The optimal value of sumabs is 0.4157, but we can get within one
#' # standard error of that CV error using sumabs=0.337, which corresponds to
#' # an average of 45.8 and 71.8 non-zero elements in each component - pretty
#' # close to the true model.
#' # We can fit the model corresponding to the lowest cross-validation error:
#' out <- PMD(x, type="standard", sumabs=cv.out$bestsumabs, K=1, v=cv.out$v.init)
#' print(out)
#' par(mfrow=c(2,2))
#' par(mar=c(2,2,2,2))
#' plot(out$u[,1], main="Est. u")
#' plot(out$v[,1], main="Est. v")
#' plot(u, main="True u")
#' plot(v, main="True v")
#' # And if we want to control sumabsu and sumabsv separately, we can do
#' # that too. Let's get 2 components while we're at it:
#' out2 <- PMD(x, type="standard", K=2, sumabsu=6, sumabsv=8, v=out$v.init,
#' cnames=paste("v", sep=" ", 1:ncol(x)), rnames=paste("u", sep=" ", 1:nrow(x)))
#' print(out2)
#'
#' # Now check out PMD with L1 penalty on rows and fused lasso penalty on
#' # columns: type="ordered". We'll use the Chin et al (2006) Cancer Cell
#' # data set; try "?breastdata" for more info.
#' \dontrun{
#' breastdata <- download_breast_data()
#' with(breastdata, {
#' # dna contains CGH data and chrom contains chromosome of each CGH spot;
#' # nuc contains position of each CGH spot.
#' dna <- t(dna) # Need samples on rows and CGH spots on columns
#' # First, look for shared regions of gain/loss on chromosome 1.
#' # Use cross-validation to choose tuning parameter value
#' par(mar=c(2,2,2,2))
#' ch1 = which(chrom == 1)
#' cv.out <- PMD.cv(dna[, ch1],type="ordered",chrom=chrom[ch1],
#' nuc=nuc[ch1],
#' sumabsus=seq(1, sqrt(nrow(dna)), len=15))
#' print(cv.out)
#' plot(cv.out)
#' out <- PMD(dna[,chrom==1],type="ordered",
#' sumabsu=cv.out$bestsumabsu,chrom=chrom[chrom==1],K=1,v=cv.out$v.init,
#' cnames=paste("Pos",sep="",
#' nuc[chrom==1]), rnames=paste("Sample", sep=" ", 1:nrow(dna)))
#' print(out, verbose=TRUE)
#' # Which samples actually have that region of gain/loss?
#' par(mfrow=c(3,1))
#' par(mar=c(2,2,2,2))
#' PlotCGH(dna[which.min(out$u[,1]),chrom==1],chrom=chrom[chrom==1],
#' main=paste(paste(paste("Sample ", sep="", which.min(out$u[,1])),
#' sep="; u=", round(min(out$u[,1]),3))),nuc=nuc[chrom==1])
#' PlotCGH(dna[88,chrom==1], chrom=chrom[chrom==1],
#' main=paste("Sample 88; u=", sep="", round(out$u[88,1],3)),
#' nuc=nuc[chrom==1])
#' PlotCGH(out$v[,1],chrom=chrom[chrom==1], main="V",nuc=nuc[chrom==1])
#' } )
#' }
#' @export PMD
PMD <- function(x, type=c("standard", "ordered"), sumabs=.4, sumabsu=5, sumabsv=NULL, lambda=NULL, niter=20, K=1, v=NULL, trace=TRUE, center=TRUE, chrom=NULL, rnames=NULL, cnames=NULL, upos=FALSE, uneg=FALSE, vpos=FALSE, vneg=FALSE){
if(upos&&uneg || vpos&&vneg) stop("Cannot contrain elements to be both positive and negative!")
if(type=="ordered" && (vpos||vneg)) stop("Cannot constrain signs of v if type is ordered.")
call <- match.call()
type <- match.arg(type)
if(type=="standard"){
out <- PMDL1L1(x, sumabs=sumabs, sumabsu=sumabsu, sumabsv=sumabsv, niter=niter, K=K, v=v, trace=trace, center=center, rnames=rnames, cnames=cnames, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg)
class(out) <- "PMDL1L1"
} else if(type=="ordered"){
out <- PMDL1FL(x,K=K,sumabsu=sumabsu,lambda=lambda,chrom=chrom,niter=niter, v=v, trace=trace, center=center, rnames=rnames, cnames=cnames, upos=upos, uneg=uneg)
class(out) <- "PMDL1FL"
}
out$upos <- upos
out$uneg <- uneg
out$vpos <- vpos
out$vneg <- vneg
out$call <- call
return(out)
}
#' Do tuning parameter selection for PMD via cross-validation
#'
#' Performs cross-validation to select tuning parameters for rank-1 PMD, the
#' penalized matrix decomposition for a data matrix.
#'
#' If type is "standard", then lasso ($L_1$) penalties (promoting sparsity) are
#' placed on u and v. If type is "ordered", then lasso penalty is placed on u
#' and a fused lasso penalty (promoting sparsity and smoothness) is placed on
#' v.
#'
#' Cross-validation of the rank-1 PMD is performed over sumabss (if type is
#' "standard") or over sumabsus (if type is "ordered"). If type is "ordered",
#' then lambda is chosen from the data without cross-validation.
#'
#' The cross-validation works as follows: Some percent of the elements of $x$
#' is removed at random from the data matrix. The PMD is performed for a range
#' of tuning parameter values on this partially-missing data matrix; then,
#' missing values are imputed using the decomposition obtained. The value of
#' the tuning parameter that results in the lowest sum of squared errors of the
#' missing values if "best".
#'
#' To do cross-validation on the rank-2 PMD, first the rank-1 PMD should be
#' computed, and then this function should be performed on the residuals, given
#' by $x-udv'$.
#'
#' @param x Data matrix of dimension $n x p$, which can contain NA for missing
#' values.
#' @param type "standard" or "ordered": Do we want v to simply be sparse, or
#' should it also be smooth? If the columns of x are ordered (e.g. CGH spots
#' along a chromosome) then choose "ordered". Default is "standard". If
#' "standard", then the PMD function will make use of sumabs OR
#' sumabsu&sumabsv. If "ordered", then the function will make use of sumabsu
#' and lambda.
#' @param sumabss Used only if type is "standard". A vector of sumabs values to
#' be used. Sumabs is a measure of sparsity for u and v vectors, between 0 and
#' 1. When sumabss is specified, and sumabsus and sumabsvs are NULL, then
#' sumabsus is set to $sqrt(n)*sumabss$ and sumabsvs is set at
#' $sqrt(p)*sumabss$. If sumabss is specified, then sumabsus and sumabsvs
#' should be NULL. Or if sumabsus and sumabsvs are specified, then sumabss
#' should be NULL.
#' @param sumabsus Used only for type "ordered". A vector of sumabsu values to
#' be used. Sumabsu measures sparseness of u - it is the sum of absolute values
#' of elements of u. Must be between 1 and sqrt(n).
#' @param lambda Used only if type is "ordered". This is the tuning parameter
#' for the fused lasso penalty on v, which takes the form $lambda ||v||_1 +
#' lambda |v_j - v_(j-1)|$. $lambda$ must be non-negative. If NULL, then it is
#' chosen adaptively from the data.
#' @param nfolds How many cross-validation folds should be performed? Default
#' is 5.
#' @param niter How many iterations should be performed. For speed, only 5 are
#' performed by default.
#' @param v The first right singular vector(s) of the data. (If missing data is
#' present, then the missing values are imputed before the singular vectors are
#' calculated.) v is used as the initial value for the iterative PMD algorithm.
#' If x is large, then this step can be time-consuming; therefore, if PMD is to
#' be run multiple times, then v should be computed once and saved.
#' @param chrom If type is "ordered", then this gives the option to specify
#' that some columns of x (corresponding to CGH spots) are on different
#' chromosomes. Then v will be sparse, and smooth *within* each chromosome but
#' not *between* chromosomes. Length of chrom should equal number of columns of
#' x, and each entry in chrom should be a number corresponding to which
#' chromosome the CGH spot is on.
#' @param nuc If type is "ordered", can specify the nucleotide position of each
#' CGH spot (column of x), to be used in plotting. If NULL, then it is assumed
#' that CGH spots are equally spaced.
#' @param trace Print out progress as iterations are performed? Default is
#' TRUE.
#' @param center Subtract out mean of x? Default is TRUE
#' @param upos Constrain the elements of u to be positive? TRUE or FALSE.
#' @param uneg Constrain the elements of u to be negative? TRUE or FALSE.
#' @param vpos Constrain the elements of v to be positive? TRUE or FALSE.
#' Cannot be used if type is "ordered".
#' @param vneg Constrain the elements of v to be negative? TRUE or FALSE.
#' Cannot be used if type is "ordered."
#' @return \item{cv}{Average sum of squared errors obtained over
#' cross-validation folds.} \item{cv.error}{Standard error of average sum of
#' squared errors obtained over cross-validation folds.} \item{bestsumabs}{If
#' type="standard", then value of sumabss resulting in smallest CV error is
#' returned.} \item{bestsumabsu}{If type="ordered", then value of sumabsus
#' resulting in smallest CV error is returned.} \item{v.init}{The first right
#' singular vector(s) of the data; these are returned to save on computation
#' time if PMD will be run again.}
#' @seealso \link{PMD}, \link{SPC}
#' @references
#' Witten D. M., Tibshirani R., and Hastie, T. (2009)
#' \emph{A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis}, \emph{Biostatistics, Gol 10 (3), 515-534, Jul 2009}\cr
#' @examples
#' # See examples in PMD help file
#' @export PMD.cv
PMD.cv <- function(x, type=c("standard", "ordered"), sumabss=seq(0.1,0.7,len=10), sumabsus=NULL, lambda=NULL, nfolds=5, niter=5, v=NULL, chrom=NULL, nuc=NULL, trace=TRUE, center=TRUE, upos=FALSE, uneg=FALSE, vpos=FALSE, vneg=FALSE){
if(upos&&uneg || vpos&&vneg) stop("Cannot contrain elements to be both positive and negative!")
if(type=="ordered" && (vpos||vneg)) stop("Cannot constrain signs of v if type is ordered.")
call <- match.call()
type <- match.arg(type)
if(type=="standard"){
out <- PMDL1L1.cv(x, sumabss=sumabss, nfolds=nfolds, niter=niter, v=v, trace=trace, center=center, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg)
class(out) <- "PMDL1L1.cv"
} else if(type=="ordered"){
out <- PMDL1FL.cv(x, nfolds=nfolds, niter=niter, lambda=lambda, sumabsus=sumabsus, chrom=chrom, v=v, trace=trace, nuc=nuc, center=center, upos=upos, uneg=uneg)
class(out) <- "PMDL1FL.cv"
}
out$upos <- upos
out$uneg <- uneg
out$vpos <- vpos
out$vneg <- vneg
out$call <- call
return(out)
}
PMDL1L1 <- function(x,sumabs=.4,sumabsu=NULL,sumabsv=NULL,niter=20,K=1,v=NULL, trace=TRUE, orth=FALSE, center=TRUE, rnames=NULL, cnames=NULL, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg){
if(center){
meanx <- mean_na(x)
x <- x-meanx
} else {
meanx <- NULL
}
if(orth){
if(is.null(sumabsu) || sumabsu < sqrt(nrow(x))){
orth <- FALSE
warning("Orth option ignored because sparse PCA results only when sumabsu equals sqrt(nrow(x))")
}
}
# if((is.null(sumabsu) && !is.null(sumabsv)) || (is.null(sumabsv) && !is.null(sumabsu))) warning("Sumabsu and sumabsv BOTH must be input when type=standard. Since only one was given, it was ignored and sumabs was used instead.")
if(is.null(sumabsu) || is.null(sumabsv)){
sumabsu <- sqrt(nrow(x))*sumabs
sumabsv <- sqrt(ncol(x))*sumabs
}
call <- match.call()
if(trace && abs(mean_na(x)) > 1e-15) warning("PMDL1L1 was run without first subtracting out the mean of x.")
if(!is.null(sumabsu) && (sumabsu<1 || sumabsu>sqrt(nrow(x)))) stop("sumabsu must be between 1 and sqrt(n)")
if(!is.null(sumabsv) && (sumabsv<1 || sumabsv>sqrt(ncol(x)))) stop("sumabsv must be between 1 and sqrt(p)")
v <- CheckPMDV(v,x,K)
if(K>1 && !orth) out <- (MultiSMD(x,sumabsu=sumabsu,sumabsv=sumabsv,niter=niter,K=K, trace=trace, v=v, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg))
if(K>1 && orth) out <- MultiSMDOrth(x,sumabsu=sumabsu,sumabsv=sumabsv,niter=niter,K=K, trace=trace, v=v, vpos=vpos, vneg=vneg)
if(K==1) out <- SMD(x,sumabsu=sumabsu,sumabsv=sumabsv,niter=niter, trace=trace, v=v, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg)
obj <- (list(u=out$u,v=out$v, d=out$d, v.init=out$v.init, call=call, meanx=meanx,sumabsu=sumabsu, sumabsv=sumabsv, rnames=rnames, cnames=cnames, K=K, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg))
class(obj) <- "PMDL1L1"
return(obj)
}
#' @method print PMDL1L1
#' @export
print.PMDL1L1 <- function(x,verbose=FALSE,...){
cat("Call: ")
dput(x$call)
cat("\n\n")
cat("Num non-zero u's: ", apply(x$u!=0, 2, sum), "\n")
cat("Num non-zero v's: ", apply(x$v!=0, 2, sum), "\n")
cat("Sumabsu used: ", x$sumabsu, "\n")
cat("Sumabsv used: ", x$sumabsv, "\n")
if(!is.null(x$meanx)) cat("Subtracted out mean of x: ", x$meanx, "\n")
if(verbose){
for(k in 1:x$K){
cat("\n Component ", k, ":\n")
u <- x$u[,k]
v <- x$v[,k]
if(is.null(x$rnames)) x$rnames <- 1:length(u)
if(is.null(x$cnames)) x$cnames <- 1:length(v)
cat(fill=T)
us <- cbind(x$rnames[u!=0], round(u[u!=0],3))
dimnames(us) <- list(1:sum(u!=0), c("Row Feature Name", "Row Feature Weight"))
vs <- cbind(x$cnames[v!=0], round(v[v!=0],3))
dimnames(vs) <- list(1:sum(v!=0), c("Column Feature Name", "Column Feature Weight"))
print(us, quote=FALSE, sep="\t")
cat(fill=T)
print(vs, quote=FALSE, sep="\t")
}
}
if(x$upos) cat("Elements of u constrained to be positive.", fill=TRUE)
if(x$vpos) cat("Elements of v constrained to be positive.", fill=TRUE)
if(x$uneg) cat("Elements of u constrained to be negative.", fill=TRUE)
if(x$vneg) cat("Elements of v constrained to be negative.", fill=TRUE)
}
#' @method print PMDL1FL
#' @export
print.PMDL1FL <- function(x,verbose=FALSE,...){
cat("Call: ")
dput(x$call)
cat("\n\n")
cat("Num non-zero u's: ", apply(x$u!=0, 2, sum), "\n")
cat("Num non-zero v's: ", apply(x$v!=0, 2, sum), "\n")
cat("Sumabsu used: ", x$sumabsu, "\n")
cat("Lambda used: ", x$lambda, "\n")
if(!is.null(x$meanx)) cat("Subtracted out mean of x: ", x$meanx, "\n")
if(verbose){
for(k in 1:x$K){
cat("\n Component ", k, ":\n")
u <- x$u[,k]
v <- x$v[,k]
if(is.null(x$rnames)) x$rnames <- 1:length(u)
if(is.null(x$cnames)) x$cnames <- 1:length(v)
cat(fill=T)
us <- cbind(x$rnames[u!=0], round(u[u!=0],3))
dimnames(us) <- list(1:sum(u!=0), c("Row Feature Name", "Row Feature Weight"))
vs <- cbind(x$cnames[v!=0], round(v[v!=0],3))
dimnames(vs) <- list(1:sum(v!=0), c("Column Feature Name", "Column Feature Weight"))
print(us, quote=FALSE, sep="\t")
cat(fill=T)
print(vs, quote=FALSE, sep="\t")
}
}
if(x$upos) cat("Elements of u constrained to be positive.", fill=TRUE)
if(x$uneg) cat("Elements of u constrained to be negative.", fill=TRUE)
}
MultiSMDOrth <- function(x,sumabsu,sumabsv,niter,K, trace, v, vpos, vneg){
if(sumabsu < sqrt(nrow(x))){
warning("sumabsu was less than sqrt(nrow(x)), so the orthogonal option was not implemented.")
return(MultiSMD(x,sumabsu=sumabsu,sumabsv=sumabsv,niter=niter,K=K, trace=trace, v=v, vpos=vpos, vneg=vneg))
}
nas <- is.na(x)
v.init <- v
xuse <- x
ds <- numeric(K)
us <- matrix(0,nrow=nrow(x),ncol=K)
vs <- matrix(0,nrow=ncol(x),ncol=K)
for(k in 1:K){
if(k==1) out <- SMD(xuse,sumabsu=sumabsu,sumabsv=sumabsv,niter=niter,v=matrix(v[,k],ncol=1), trace=trace, upos=FALSE, uneg=FALSE, vpos=vpos, vneg=vneg)
if(k>1) out <- SMDOrth(xuse,us[,1:(k-1)],sumabsv=sumabsv,niter=niter,v=matrix(v[,k],ncol=1), trace=trace, vpos=vpos, vneg=vneg)
us[,k] <- out$u
vs[,k] <- out$v
ds[k] <- out$d
}
return(list(u=us,v=vs,d=ds, v.init=v.init))
}
MultiSMD <- function(x, sumabsu, sumabsv, K=3, niter=20,v, trace=TRUE, upos, uneg, vpos, vneg){
nas <- is.na(x)
v.init <- v
xuse <- x
ds <- numeric(K)
us <- matrix(0,nrow=nrow(x),ncol=K)
vs <- matrix(0,nrow=ncol(x),ncol=K)
for(k in 1:K){
out <- SMD(xuse, sumabsu=sumabsu,sumabsv=sumabsv,niter=niter,v=matrix(v[,k],ncol=1), trace=trace, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg)
us[,k] <- out$u
vs[,k] <- out$v
ds[k] <- out$d
res <- xuse - out$d*out$u%*%t(out$v)
xuse[!nas] <- res[!nas] # [!nas] is new on July 24 2009
}
return(list(u=us,v=vs,d=ds, v.init=v.init))
}
PMDL1L1.cv <- function(x, sumabss=seq(0.1,0.7,len=10), nfolds=5, niter=5, v=NULL, trace=TRUE, center=TRUE, upos, uneg, vpos, vneg){
if(nfolds<2) stop("Must run at least 2 cross-validation folds.")
percentRemove <- min(.25, 1/nfolds)
call <- match.call()
if(max(sumabss)>1 || min(sumabss)<0) stop("sumabs must be between 0 and 1")
# PMD only cross-validates the single-factor model.
# and it requires just one sparsity measure.
xfill <- x
missing <- is.na(x)
v <- CheckPMDV(v,x,K=1)
errs <- matrix(NA, nrow=nfolds, ncol=length(sumabss))
nonzerous <- matrix(NA, nrow=nfolds, ncol=length(sumabss))
nonzerovs <- matrix(NA, nrow=nfolds, ncol=length(sumabss))
rands <- matrix(runif(nrow(x)*ncol(x)),ncol=ncol(x))
for(i in 1:nfolds){
if(trace) cat(" Fold ", i, " out of ", nfolds, "\n")
rm <- ((i-1)*percentRemove < rands)&(rands < i*percentRemove)
xrm <- x
xrm[rm] <- NA
for(j in 1:length(sumabss)){
out <- PMDL1L1(xrm, sumabs=sumabss[j], niter=niter, v=v, trace=FALSE, center=center,K=1, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg)
xhat <- as.numeric(out$d)*out$u%*%t(out$v)
errs[i,j] <- sum(((xhat-x)[rm & !missing])^2)
nonzerous[i,j] <- sum(out$u!=0)
nonzerovs[i,j] <- sum(out$v!=0)
}
}
if(trace) cat(fill=TRUE)
err.means <- apply(errs, 2, mean)
err.sds <- apply(errs, 2, sd)/sqrt(nfolds)
nonzerous.mean <- apply(nonzerous,2,mean)
nonzerovs.mean <- apply(nonzerovs,2,mean)
bestsumabs <- sumabss[which.min(err.means)]
object <- (list(cv=err.means, cv.error=err.sds,bestsumabs=bestsumabs, nonzerous=nonzerous.mean, nonzerovs=nonzerovs.mean, v.init=v, call=call, sumabss=sumabss, nfolds=nfolds))
class(object) <- "PMDL1L1.cv"
return(object)
}
#' Perform cross-validation on sparse principal component analysis
#'
#' Selects tuning parameter for the sparse principal component analysis method
#' of Witten, Tibshirani, and Hastie (2008), which involves applying PMD to a
#' data matrix with lasso ($L_1$) penalty on the columns and no penalty on the
#' rows. The tuning parameter controls the sum of absolute values - or $L_1$
#' norm - of the elements of the sparse principal component.
#'
#' This method only performs cross-validation for the first sparse principal
#' component. It does so by performing the following steps nfolds times: (1)
#' replace a fraction of the data with missing values, (2) perform SPC on this
#' new data matrix using a range of tuning parameter values, each time getting
#' a rank-1 approximationg $udv'$ where $v$ is sparse, (3) measure the mean
#' squared error of the rank-1 estimate of the missing values created in step
#' 1.
#'
#' Then, the selected tuning parameter value is that which resulted in the
#' lowest average mean squared error in step 3.
#'
#' In order to perform cross-validation for the second sparse principal
#' component, apply this function to $X-udv'$ where $udv'$ are the output of
#' running SPC on the raw data $X$.
#'
#' @aliases SPC.cv plot.SPC.cv print.SPC.cv
#' @param x Data matrix of dimension $n x p$, which can contain NA for missing
#' values. We are interested in finding sparse principal components of
#' dimension $p$.
#' @param sumabsvs Range of sumabsv values to be considered in
#' cross-validation. Sumabsv is the sum of absolute values of elements of v. It
#' must be between 1 and square root of number of columns of data. The smaller
#' it is, the sparser v will be.
#' @param nfolds Number of cross-validation folds performed.
#' @param niter How many iterations should be performed. By default, perform
#' only 5 for speed reasons.
#' @param v The first right singular vector(s) of the data. (If missing data is
#' present, then the missing values are imputed before the singular vectors are
#' calculated.) v is used as the initial value for the iterative PMD($L_1$,
#' $L_1$) algorithm. If x is large, then this step can be time-consuming;
#' therefore, if PMD is to be run multiple times, then v should be computed
#' once and saved.
#' @param trace Print out progress as iterations are performed? Default is
#' TRUE.
#' @param orth If TRUE, then use method of Section 3.2 of Witten, Tibshirani
#' and Hastie (2008) to obtain multiple sparse principal components. Default is
#' FALSE.
#' @param center Subtract out mean of x? Default is TRUE
#' @param vpos Constrain elements of v to be positive? Default is FALSE.
#' @param vneg Constrain elements of v to be negative? Default is FALSE.
#' @return \item{cv}{Average sum of squared errors that results for each tuning
#' parameter value.} \item{cv.error}{Standard error of the average sum of
#' squared error that results for each tuning parameter value.}
#' \item{bestsumabsv}{Value of sumabsv that resulted in lowest CV error.}
#' \item{nonzerovs}{Average number of non-zero elements of v for each candidate
#' value of sumabsvs.} \item{v.init}{Initial value of v that was passed in. Or,
#' if that was NULL, then first right singular vector of X.}
#' \item{bestsumabsv1se}{The smallest value of sumabsv that is within 1
#' standard error of smallest CV error.}
#' @seealso \link{SPC}, \link{PMD}, \link{PMD.cv}
#' @references
#' Witten D. M., Tibshirani R., and Hastie, T. (2009)
#' \emph{A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis}, \emph{Biostatistics, Gol 10 (3), 515-534, Jul 2009}\cr
#' @examples
#'
#' #NOT RUN
#' ## A simple simulated example
#' #set.seed(1)
#' #u <- matrix(c(rnorm(50), rep(0,150)),ncol=1)
#' #v <- matrix(c(rnorm(75),rep(0,225)), ncol=1)
#' #x <- u%*%t(v)+matrix(rnorm(200*300),ncol=300)
#' ## Perform Sparse PCA - that is, decompose a matrix w/o penalty on rows
#' ## and w/ L1 penalty on columns
#' ## First, we perform sparse PCA and get 4 components, but we do not
#' ## require subsequent components to be orthogonal to previous components
#' #cv.out <- SPC.cv(x, sumabsvs=seq(1.2, sqrt(ncol(x)), len=6))
#' #print(cv.out)
#' #plot(cv.out)
#' #out <- SPC(x,sumabsv=cv.out$bestsumabs, K=4) # could use
#' ## cv.out$bestsumabvsv1se instead
#' #print(out,verbose=TRUE)
#' ## Now, we do sparse PCA using method in Section 3.2 of WT&H(2008) for getting
#' ## multiple components - that is, we require components to be orthogonal
#' #cv.out <- SPC.cv(x, sumabsvs=seq(1.2, sqrt(ncol(x)), len=6), orth=TRUE)
#' #print(cv.out)
#' #plot(cv.out)
#' #out.orth <- SPC(x,sumabsv=cv.out$bestsumabsv, K=4, orth=TRUE)
#' #print(out.orth,verbose=TRUE)
#' #par(mfrow=c(1,1))
#' #plot(out$u[,1], out.orth$u[,1], xlab="", ylab="")
#' #
#' #
#' @export SPC.cv
SPC.cv <- function(x, sumabsvs=seq(1.2,5,len=10), nfolds=5, niter=5, v=NULL, trace=TRUE, orth=FALSE, center=TRUE, vpos=FALSE, vneg=FALSE){
if(vpos&&vneg) stop("Cannot constrain elements of v to be both positive and negative.")
if(nfolds<2) stop("Must run at least 2 cross-validation folds.")
percentRemove <- min(.25, 1/nfolds)
call <- match.call()
if(max(sumabsvs)>sqrt(ncol(x)) || min(sumabsvs)<1) stop("sumabs must be between 1 and sqrt(ncol(x))")
xfill <- x
missing <- is.na(x)
v <- CheckPMDV(v,x,K=1)
errs <- matrix(NA, nrow=nfolds, ncol=length(sumabsvs))
nonzerovs <- matrix(NA, nrow=nfolds, ncol=length(sumabsvs))
rands <- matrix(runif(nrow(x)*ncol(x)),ncol=ncol(x))
for(i in 1:nfolds){
if(trace) cat(" Fold ", i, " out of ", nfolds, "\n")
rm <- ((i-1)*percentRemove < rands)&(rands < i*percentRemove)
xrm <- x
xrm[rm] <- NA
for(j in 1:length(sumabsvs)){
out <- SPC(xrm, sumabsv=sumabsvs[j], orth=orth, niter=niter, v=v, trace=FALSE, center=center,K=1, vpos=vpos, vneg=vneg)
xhat <- as.numeric(out$d)*out$u%*%t(out$v)
errs[i,j] <- sum(((xhat-x)[rm & !missing])^2)
nonzerovs[i,j] <- sum(out$v!=0)
}
}
if(trace) cat(fill=TRUE)
err.means <- apply(errs, 2, mean)
err.sds <- apply(errs, 2, sd)/sqrt(nfolds)
nonzerovs.mean <- apply(nonzerovs,2,mean)
bestsumabsv <- sumabsvs[which.min(err.means)]
bestsumabsv1se <- sumabsvs[min(which(err.means < min(err.means) + err.sds[which.min(err.means)]))]
object <- (list(cv=err.means, cv.error=err.sds,bestsumabsv=bestsumabsv,nonzerovs=nonzerovs.mean, v.init=v, call=call, sumabsvs=sumabsvs, nfolds=nfolds, bestsumabsv1se=bestsumabsv1se, vpos=vpos, vneg=vneg))
class(object) <- "SPC.cv"
return(object)
}
#' @method plot SPC.cv
#' @export
plot.SPC.cv <- function(x,...){
sumabsvs <- x$sumabsvs
err.means <- x$cv
err.sds <- x$cv.error
plot(sumabsvs, err.means, ylim=range(c(err.means+err.sds, err.means-err.sds)), main="CV Error for SPC", xlab="Value of sumabsv", ylab="Average SSE", type="l")
points(sumabsvs, err.means)
lines(sumabsvs, err.means+err.sds, lty="dashed")
lines(sumabsvs, err.means-err.sds, lty="dashed")
}
#' @method print SPC.cv
#' @export
print.SPC.cv <- function(x,...){
cat("Call:\n")
dput(x$call)
cat("\n \n")
cat('Cross-validation errors: \n')
mat <- round(cbind(round(x$sumabsvs, 3), x$cv, x$cv.error, x$nonzerovs),3)
dimnames(mat) <- list(1:length(x$sumabsvs), c("Sumabsvs", "CV Error", "CV S.E.", "# non-zero v's"))
print(mat, quote=F)
cat('\n Best sumabsv value (lowest CV error): ', x$bestsumabsv, "\n")
cat('\n Smallest sumabsv value that has CV error within 1 SE of best CV error: ', x$bestsumabsv1se, "\n")
if(x$vpos) cat("Elements of v constrained to be positive.", fill=TRUE)
if(x$vneg) cat("Elements of v constrained to be negative.", fill=TRUE)
}
#' @method print PMDL1L1.cv
#' @export
print.PMDL1L1.cv <- function(x,...){
cat("Call:\n")
dput(x$call)
cat("\n \n")
cat('Cross-validation errors: \n')
mat <- round(cbind(round(x$sumabss, 3), x$cv, x$cv.error, x$nonzerous, x$nonzerovs),3)
dimnames(mat) <- list(1:length(x$sumabss), c("Sumabss", "CV Error", "CV S.E.", "# non-zero u's", "# non-zero v's"))
print(mat, quote=F)
cat('\n Best sumabs value (lowest CV error): ', x$bestsumabs, "\n")
if(x$upos) cat("Elements of u constrained to be positive.", fill=TRUE)
if(x$uneg) cat("Elements of u constrained to be negative.", fill=TRUE)
if(x$vpos) cat("Elements of v constrained to be positive.", fill=TRUE)
if(x$vneg) cat("Elements of v constrained to be negative.", fill=TRUE)
}
#' @method plot PMDL1L1.cv
#' @export
plot.PMDL1L1.cv <- function(x,...){
sumabss <- x$sumabss
err.means <- x$cv
err.sds <- x$cv.error
plot(sumabss, err.means, ylim=range(c(err.means+err.sds, err.means-err.sds)), main="CV Error for PMD(L1,L1)", xlab="Value of sumabs", ylab="Average SSE", type="l")
points(sumabss, err.means)
lines(sumabss, err.means+err.sds, lty="dashed")
lines(sumabss, err.means-err.sds, lty="dashed")
}
PMDL1FL.cv <- function(x, nfolds=5, niter=5, lambda=NULL, sumabsus=NULL, chrom=NULL, v=NULL, trace=TRUE, nuc=NULL, center=TRUE, upos, uneg){
if(is.null(sumabsus)) sumabsus <- seq(1, .7*sqrt(nrow(x)),len=10)
percentRemove <- min(.25, 1/nfolds)
call <- match.call()
if(is.null(chrom)) chrom <- rep(1, ncol(x))
v <- CheckPMDV(v,x,K=1)
if(is.null(lambda)) lambda <- ChooseLambda1Lambda2(as.numeric(v[,1]))
errs <- matrix(NA, nrow=nfolds, ncol=length(sumabsus))
nonzerous <- matrix(NA, nrow=nfolds, ncol=length(sumabsus))
nonzerovs <- matrix(NA, nrow=nfolds, ncol=length(sumabsus))
rands <- matrix(runif(nrow(x)*ncol(x)), nrow=nrow(x))
missing <- is.na(x)
for(i in 1:nfolds){
if(trace) cat(" Fold ", i, " out of ", nfolds, "\n")
rm <- ((i-1)*percentRemove < rands)&(rands < i*percentRemove)
xrm <- x
xrm[rm] <- NA
for(j in 1:length(sumabsus)){
if(trace) cat(j,fill=F)
out <- PMDL1FL(xrm, K=1, lambda=lambda,sumabsu=sumabsus[j], niter=niter, chrom=chrom, v=v, trace=FALSE, center=center, upos=upos, uneg=uneg)
xhat <- as.numeric(out$d)*out$u%*%t(out$v)
errs[i,j] <- sum(((xhat-x)[rm & !missing])^2)
nonzerous[i,j] <- sum(out$u!=0)
nonzerovs[i,j] <- sum(out$v!=0)
}
}
if(trace) cat(fill=TRUE)
err.means <- apply(errs, 2, mean)
err.sds <- apply(errs, 2, sd)/sqrt(nfolds)
nonzerous.mean <- apply(nonzerous,2,mean)
nonzerovs.mean <- apply(nonzerovs,2,mean)
bestind <- which.min(err.means)
bestsumabsu <- sumabsus[bestind]
object <- (list(cv=err.means, lambda=lambda,cv.error=err.sds,bestsumabsu=bestsumabsu, nonzerous=nonzerous.mean, nonzerovs=nonzerovs.mean, v.init=v, call=call, sumabsus=sumabsus, nfolds=nfolds,x=x,chrom=chrom,nuc=nuc, niter=niter))
class(object) <- "PMDL1FL.cv"
return(object)
}
#' @method plot PMDL1FL.cv
#' @export
plot.PMDL1FL.cv <- function(x,...){
mat <- x$x
sumabsus <- x$sumabsus
err.means <- x$cv
err.sds <- x$cv.error
chrom <- x$chrom
niter <- x$niter
nuc <- x$nuc
lambda <- x$lambda
v <- x$v.init
bestsumabsu <- x$bestsumabsu
par(mfrow=c(2,1))
plot(sumabsus, err.means, ylim=range(c(err.means+err.sds, err.means-err.sds)), main="CV Error for PMD(L1, FL)", xlab="Value of sumabsu", ylab="Average SSE", type="l")
points(sumabsus, err.means)
lines(sumabsus, err.means+err.sds, lty="dashed")
lines(sumabsus, err.means-err.sds, lty="dashed")
out <- PMDL1FL(mat, sumabsu=bestsumabsu, K=1, chrom=chrom, niter=niter, v=v, trace=FALSE, lambda=lambda, upos=x$upos, uneg=x$uneg)
PlotCGH(out$v, main="Best V", chrom=chrom, nuc=nuc)
}
#' @method print PMDL1FL.cv
#' @export
print.PMDL1FL.cv <- function(x,...){
cat("Call:\n")
dput(x$call)
cat("\n \n")
cat('Cross-validation errors: \n')
mat <- round(cbind(round(x$sumabsus, 3), x$cv, x$cv.error, x$nonzerous, x$nonzerovs),3)
dimnames(mat) <- list(1:length(x$sumabsus), c("Sumabsus", "CV Error", "CV S.E.", "# non-zero u's", "# non-zero v's"))
print(mat, quote=F)
cat('\n Best sumabs value : ', x$bestsumabsu, "\n")
cat('\n Lambda Used : ', x$lambda, "\n")
if(x$upos) cat("Elements of u constrained to be positive.", fill=TRUE)
if(x$uneg) cat("Elements of u constrained to be negative.", fill=TRUE)
}
Try the PMA package in your browser
Any scripts or data that you put into this service are public.
PMA documentation built on Sept. 11, 2024, 7:02 p.m.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Defines functions print.PMDL1FL.cv plot.PMDL1FL.cv PMDL1FL.cv plot.PMDL1L1.cv print.PMDL1L1.cv print.SPC.cv plot.SPC.cv SPC.cv PMDL1L1.cv MultiSMD MultiSMDOrth print.PMDL1FL print.PMDL1L1 PMDL1L1 PMD.cv PMD print.SPC SPC CheckPMDV SMDOrth SMD BinarySearch safesvd l2n mean_na soft
Documented in plot.SPC.cv PMD PMD.cv print.SPC print.SPC.cv SPC SPC.cv
soft <- function(x,d){
return(sign(x)*pmax(0, abs(x)-d))
}
mean_na <- function(vec){
return(mean(vec[!is.na(vec)]))
}
l2n <- function(vec){
a <- sqrt(sum(vec^2))
if(a==0) a <- .05
return(a)
}
safesvd <- function(x){
i <- 1
out <- try(svd(x), silent=TRUE)
while(i<10 && inherits(out, "try-error")){
out <- try(svd(x), silent=TRUE)
i <- i+1
}
if(inherits(out, "try-error")) out <- svd(matrix(rnorm(nrow(x)*ncol(x)), ncol=ncol(x)))
return(out)
}
BinarySearch <- function(argu,sumabs){
if(l2n(argu)==0 || sum(abs(argu/l2n(argu)))<=sumabs) return(0)
lam1 <- 0
lam2 <- max(abs(argu))-1e-5
iter <- 1
while(iter < 150){
su <- soft(argu,(lam1+lam2)/2)
if(sum(abs(su/l2n(su)))<sumabs){
lam2 <- (lam1+lam2)/2
} else {
lam1 <- (lam1+lam2)/2
}
if((lam2-lam1)<1e-6) return((lam1+lam2)/2)
iter <- iter+1
}
warning("Didn't quite converge")
return((lam1+lam2)/2)
}
SMD <- function(x, sumabsu, sumabsv, niter=20,trace=TRUE, v, upos, uneg, vpos, vneg){
# This gets a single factor. Do MultiSMD to get multiple factors.
nas <- is.na(x)
v.init <- v
xoo <- x
if(sum(nas)>0) xoo[nas] <- mean(x[!nas])
oldv <- rnorm(ncol(x))
for(iter in 1:niter){
if(sum(abs(oldv-v))>1e-7){
oldv <- v
if(trace) cat(iter,fill=F)
# update u #
argu <- xoo%*%v
if(upos) argu <- pmax(argu,0)
if(uneg) argu <- pmin(argu,0)
lamu <- BinarySearch(argu,sumabsu)
su <- soft(argu,lamu)
u <- matrix(su/l2n(su),ncol=1)
# done updating u #
# update v #
argv <- t(u)%*%xoo
if(vpos) argv <- pmax(argv,0)
if(vneg) argv <- pmin(argv,0)
lamv <- BinarySearch(argv, sumabsv)
sv <- soft(argv,lamv)
v <- matrix(sv/l2n(sv),ncol=1)
# done updating v #
}
}
d <- as.numeric(t(u)%*%(xoo%*%v))
if(trace) cat(fill=TRUE)
return(list(d=d, u=u, v=v, v.init=v.init))
}
SMDOrth <- function(x, us, sumabsv=NULL, niter=20, trace=TRUE,v, vpos, vneg){
# Gets the next u for sparse PCA, using Trevor's method of requiring this new u to be orthog
# to all previous us (contained in us)
nas <- is.na(x)
v.init <- v
xoo <- x
if(sum(nas)>0) xoo[nas] <- mean(x[!nas])
u <- rnorm(nrow(x))
oldu <- rnorm(nrow(x))
oldv <- rnorm(ncol(x))
for(iter in 1:niter){
if(sum(abs(oldu-u))>1e-6 || sum(abs(oldv-v))>1e-6){
oldu <- u
oldv <- v
if(trace) cat(iter,fill=F)
# update u #
argu <- xoo%*%v
numer <- lsfit(y=argu, x=us, intercept=FALSE)$res
u <- numer/l2n(numer)
# done updating u #
# update v #
argv <- t(u)%*%xoo
if(vpos) argv <- pmax(argv,0)
if(vneg) argv <- pmin(argv,0)
lamv <- BinarySearch(argv, sumabsv)
sv <- soft(argv,lamv)
v <- matrix(sv/l2n(sv),ncol=1)
# done updating v #
}
}
d <- as.numeric(t(u)%*%(xoo%*%v))
if(trace) cat(fill=TRUE)
return(list(d=d,u=u,v=v, v.init=v.init))
}
CheckPMDV <- function(v,x,K){
if(!is.null(v) && is.matrix(v) && ncol(v)>=K){
v <- matrix(v[,1:K], ncol=K)
} else if(ncol(x)>nrow(x)){
x[is.na(x)] <- mean_na(x)
v <- matrix(t(x)%*%(safesvd(x%*%t(x))$v[,1:K]),ncol=K)
if(sum(is.na(v))>0) v <- matrix(safesvd(x)$v[,1:K], ncol=K)
v <- sweep(v,2,apply(v, 2, l2n), "/")
if(sum(is.na(v))>0) stop("some are NA")
} else if (ncol(x)<=nrow(x)){
x[is.na(x)] <- mean_na(x)
v <- matrix(safesvd(t(x)%*%x)$v[,1:K],ncol=K)
}
return(v)
}
#' Perform sparse principal component analysis
#'
#' Performs sparse principal components analysis by applying PMD to a data
#' matrix with lasso ($L_1$) penalty on the columns and no penalty on the rows.
#'
#' PMD(x,sumabsu=sqrt(nrow(x)), sumabsv=3, K=1) and SPC(x,sumabsv=3, K=1) give
#' the same result, since the SPC method is simply PMD with an L1 penalty on
#' the columns and no penalty on the rows.
#'
#' In Witten, Tibshirani, and Hastie (2008), two methods are presented for
#' obtaining multiple factors for SPC. The methods are as follows:
#'
#' (1) If one has already obtained factors $k-1$ factors then oen can compute
#' residuals by subtracting out these factors. Then $u_k$ and $v_k$ can be
#' obtained by applying the SPC/PMD algorithm to the residuals.
#'
#' (2) One can require that $u_k$ be orthogonal to $u_i$'s with $i<k$; the
#' method is slightly more complicated, and is explained in WT&H(2008).
#'
#' Method 1 is performed by running SPC with option orth=FALSE (the default)
#' and Method 2 is performed using option orth=TRUE. Note that Methods 1 and 2
#' always give identical results for the first component, and often given quite
#' similar results for later components.
#'
#' @aliases SPC print.SPC
#' @param x Data matrix of dimension $n x p$, which can contain NA for missing
#' values. We are interested in finding sparse principal components of
#' dimension $p$.
#' @param sumabsv How sparse do you want v to be? This is the sum of absolute
#' values of elements of v. It must be between 1 and square root of number of
#' columns of data. The smaller it is, the sparser v will be.
#' @param niter How many iterations should be performed. It is best to run at
#' least 20 of so. Default is 20.
#' @param K The number of factors in the PMD to be returned; default is 1.
#' @param v The first right singular vector(s) of the data. (If missing data is
#' present, then the missing values are imputed before the singular vectors are
#' calculated.) v is used as the initial value for the iterative PMD($L_1$,
#' $L_1$) algorithm. If x is large, then this step can be time-consuming;
#' therefore, if PMD is to be run multiple times, then v should be computed
#' once and saved.
#' @param trace Print out progress as iterations are performed? Default is
#' TRUE.
#' @param orth If TRUE, then use method of Section 3.2 of Witten, Tibshirani
#' and Hastie (2008) to obtain multiple sparse principal components. Default is
#' FALSE.
#' @param center Subtract out mean of x? Default is TRUE
#' @param cnames An optional vector containing a name for each column.
#' @param vpos Constrain the elements of v to be positive? TRUE or FALSE.
#' @param vneg Constrain the elements of v to be negative? TRUE or FALSE.
#' @param compute.pve Compute percent variance explained? Default TRUE. If not
#' needed, then choose FALSE to save time.
#' @return \item{u}{u is output. If you asked for multiple factors then each
#' column of u is a factor. u has dimension nxK if you asked for K factors.}
#' \item{v}{v is output. These are the sparse principal components. If you
#' asked for multiple factors then each column of v is a factor. v has
#' dimension pxK if you asked for K factors.} \item{d}{d is output; it is the
#' diagonal of the matrix $D$ in the penalized matrix decomposition. In the
#' case of the rank-1 decomposition, it is given in the formulation
#' $||X-duv'||_F^2$ subject to $||u||_1 <= sumabsu$, $||v||_1 <= sumabsv$.
#' Computationally, $d=u'Xv$ where $u$ and $v$ are the sparse factors output by
#' the PMD function and $X$ is the data matrix input to the PMD function.}
#' \item{prop.var.explained}{A vector containing the proportion of variance
#' explained by the first 1, 2, ..., K sparse principal components obtaineds.
#' Formula for proportion of variance explained is on page 20 of Shen & Huang
#' (2008), Journal of Multivariate Analysis 99: 1015-1034.} \item{v.init}{The
#' first right singular vector(s) of the data; these are returned to save on
#' computation time if PMD will be run again.} \item{meanx}{Mean of x that was
#' subtracted out before SPC was performed.}
#' @seealso \link{SPC.cv}, \link{PMD}, \link{PMD.cv}
#' @references
#' Witten D. M., Tibshirani R., and Hastie, T. (2009)
#' \emph{A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis}, \emph{Biostatistics, Gol 10 (3), 515-534, Jul 2009}\cr
#' @examples
#'
#'
#' # A simple simulated example
#' #NOT RUN
#' #set.seed(1)
#' #u <- matrix(c(rnorm(50), rep(0,150)),ncol=1)
#' #v <- matrix(c(rnorm(75),rep(0,225)), ncol=1)
#' #x <- u%*%t(v)+matrix(rnorm(200*300),ncol=300)
#' ## Perform Sparse PCA - that is, decompose a matrix w/o penalty on rows
#' ## and w/ L1 penalty on columns
#' ## First, we perform sparse PCA and get 4 components, but we do not
#' ## require subsequent components to be orthogonal to previous components
#' #out <- SPC(x,sumabsv=3, K=4)
#' #print(out,verbose=TRUE)
#' ## We could have selected sumabsv by cross-validation, using function SPC.cv
#' ## Now, we do sparse PCA using method in Section 3.2 of WT&H(2008) for getting
#' ## multiple components - that is, we require components to be orthogonal
#' #out.orth <- SPC(x,sumabsv=3, K=4, orth=TRUE)
#' #print(out.orth,verbose=TRUE)
#' #par(mfrow=c(1,1))
#' #plot(out$u[,1], out.orth$u[,1], xlab="", ylab="")
#' ## Note that the first components w/ and w/o orth option are identical,
#' ## since the orth option only affects the way that subsequent components
#' ## are found
#' #print(round(t(out$u)%*%out$u,4)) # not orthogonal
#' #print(round(t(out.orth$u)%*%out.orth$u,4)) # orthogonal
#' #
#' ## Use SPC.cv to choose tuning parameters:
#' #cv.out <- SPC.cv(x)
#' #print(cv.out)
#' #plot(cv.out)
#' #out <- SPC(x, sumabsv=cv.out$bestsumabsv)
#' #print(out)
#' ## or we could do
#' #out <- SPC(x, sumabsv=cv.out$bestsumabsv1se)
#' #print(out)
#' #
#' #
#' @export SPC
SPC <- function(x, sumabsv=4, niter=20, K=1, orth=FALSE, trace=TRUE, v=NULL, center=TRUE, cnames=NULL, vpos=FALSE, vneg=FALSE, compute.pve=TRUE){
if(vpos&&vneg) stop("Cannot constrain elements to be positive AND negative.")
out <- PMDL1L1(x,sumabsu=sqrt(nrow(x)), sumabsv=sumabsv, niter=niter,K=K,orth=orth,trace=trace,v=v,center=center,cnames=cnames, upos=FALSE, uneg=FALSE, vpos=vpos, vneg=vneg)
if(compute.pve){
# Calculate percent variance explained, using definition on page 7 of Shen and Huang (2008) Journal of Multivariate Analysis vol 99: 1015-1034
v <- matrix(out$v, ncol=K)
ve <- NULL # variance explained
xfill <- x
if(center) xfill <- x-mean_na(x)
xfill[is.na(x)] <- mean_na(xfill)
for(k in 1:K){
vk <- matrix(v[,1:k], ncol=k)
xk <- xfill%*%vk%*%solve(t(vk)%*%vk)%*%t(vk)
svdxk <- svd(xk)
ve <- c(ve, sum(svdxk$d^2))
}
pve <- ve/sum(svd(xfill)$d^2) # proportion of variance explained
out$prop.var.explained <- pve
}
out$vpos <- vpos
out$vneg <- vneg
class(out) <- "SPC"
return(out)
}
#' @method print SPC
#' @export
print.SPC <- function(x,verbose=FALSE,...){
cat("Call: ")
dput(x$call)
cat("\n\n")
cat("Num non-zero v's: ", apply(x$v!=0, 2, sum), "\n")
cat("Sumabsv used: ", x$sumabsv, "\n")
cat("Proportion of variance explained by first k components: \n")
tab <- cbind(paste(paste("First ", sep="", 1:length(x$prop.var.explained)), sep="", " Components"), round(x$prop.var.explained, 3))
dimnames(tab) <- list(1:length(x$prop.var.explained), c("", "Prop. Var. Explained"))
print(tab, quote=FALSE, sep="\t")
if(!is.null(x$meanx)) cat("Subtracted out mean of x: ", x$meanx, "\n")
if(verbose){
for(k in 1:x$K){
cat("\n Component ", k, ":\n")
v <- x$v[,k]
if(is.null(x$cnames)) x$cnames <- 1:length(v)
cat(fill=T)
vs <- cbind(x$cnames[v!=0], round(v[v!=0],3))
dimnames(vs) <- list(1:sum(v!=0), c("Feature Name", "Feature Weight"))
print(vs, quote=FALSE, sep="\t")
}
}
if(x$vpos) cat("Elements of v constrained to be positive.", fill=TRUE)
if(x$vneg) cat("Elements of v constrained to be negative.", fill=TRUE)
}
#' Get a penalized matrix decomposition for a data matrix.
#'
#' Performs a penalized matrix decomposition for a data matrix. Finds factors u
#' and v that summarize the data matrix well. u and v will both be sparse, and
#' v can optionally also be smooth.
#'
#' The criterion for the PMD is as follows: we seek vectors $u$ and $v$ such
#' that $u'Xv$ is large, subject to $||u||_2=1, ||v||_2=1$ and additional
#' penalties on $u$ and $v$. These additional penalties are as follows: If type
#' is "standard", then lasso ($L_1$) penalties (promoting sparsity) are placed
#' on u and v. If type is "ordered", then lasso penalty is placed on u and a
#' fused lasso penalty (promoting sparsity and smoothness) is placed on v.
#'
#' If type is "standard", then arguments sumabs OR sumabsu&sumabsv are used. If
#' type is "ordered", then sumabsu AND lambda are used. Sumabsu is the bound of
#' absolute value of elements of u. Sumabsv is bound of absolute value of
#' elements of v. If sumabs is given, then sumabsu is set to
#' sqrt(nrow(x))*sumabs and sumabsv is set to sqrt(ncol(x))*sumabs. $lambda$ is
#' the parameter for the fused lasso penalty on v when type is "ordered":
#' $lambda(||v||_1 + sum_j |v_j - v_(j-1))$.
#'
#' @param x Data matrix of dimension $n x p$, which can contain NA for missing
#' values.
#' @param type "standard" or "ordered": Do we want v to simply be sparse, or
#' should it also be smooth? If the columns of x are ordered (e.g. CGH spots
#' along a chromosome) then choose "ordered". Default is "standard". If
#' "standard", then the PMD function will make use of sumabs OR
#' sumabsu&sumabsv. If "ordered", then the function will make use of sumabsu
#' and lambda.
#' @param sumabs Used only if type is "standard". A measure of sparsity for u
#' and v vectors, between 0 and 1. When sumabs is specified, and sumabsu and
#' sumabsv are NULL, then sumabsu is set to $sqrt(n)*sumabs$ and sumabsv is set
#' to $sqrt(p)*sumabs$. If sumabs is specified, then sumabsu and sumabsv should
#' be NULL. Or if sumabsu and sumabsv are specified, then sumabs should be
#' NULL.
#' @param sumabsu Used for types "ordered" AND "standard". How sparse do you
#' want u to be? This is the sum of absolute values of elements of u. It must
#' be between 1 and the square root of the number of rows in data matrix. The
#' smaller it is, the sparser u will be.
#' @param sumabsv Used only if type is "standard". How sparse do you want v to
#' be? This is the sum of absolute values of elements of v. It must be between
#' 1 and square root of number of columns of data. The smaller it is, the
#' sparser v will be.
#' @param lambda Used only if type is "ordered". This is the tuning parameter
#' for the fused lasso penalty on v, which takes the form $lambda ||v||_1 +
#' lambda |v_j - v_(j-1)|$. $lambda$ must be non-negative. If NULL, then it is
#' chosen adaptively from the data.
#' @param niter How many iterations should be performed. It is best to run at
#' least 20 of so. Default is 20.
#' @param K The number of factors in the PMD to be returned; default is 1.
#' @param v The first right singular vector(s) of the data. (If missing data is
#' present, then the missing values are imputed before the singular vectors are
#' calculated.) v is used as the initial value for the iterative PMD algorithm.
#' If x is large, then this step can be time-consuming; therefore, if PMD is to
#' be run multiple times, then v should be computed once and saved.
#' @param trace Print out progress as iterations are performed? Default is
#' TRUE.
#' @param center Subtract out mean of x? Default is TRUE.
#' @param chrom If type is "ordered", then this gives the option to specify
#' that some columns of x (corresponding to CGH spots) are on different
#' chromosomes. Then v will be sparse, and smooth *within* each chromosome but
#' not *between* chromosomes. Length of chrom should equal number of columns of
#' x, and each entry in chrom should be a number corresponding to which
#' chromosome the CGH spot is on.
#' @param rnames An optional vector containing a name for each row of x.
#' @param cnames An optional vector containing a name for each column of x.
#' @param upos Constrain the elements of u to be positive? TRUE or FALSE.
#' @param uneg Constrain the elements of u to be negative? TRUE or FALSE.
#' @param vpos Constrain the elements of v to be positive? TRUE or FALSE.
#' Cannot be used if type is "ordered".
#' @param vneg Constrain the elements of v to be negative? TRUE or FALSE.
#' Cannot be used if type is "ordered."
#' @return \item{u}{u is output. If you asked for multiple factors then each
#' column of u is a factor. u has dimension nxK if you asked for K factors.}
#' \item{v}{v is output. If you asked for multiple factors then each column of
#' v is a factor. v has dimension pxK if you asked for K factors.} \item{d}{d
#' is output. Computationally, $d=u'Xv$ where $u$ and $v$ are the sparse
#' factors output by the PMD function and $X$ is the data matrix input to the
#' PMD function. When K=1, the residuals of the rank-1 PMD are given by $X -
#' duv'$.} \item{v.init}{The first right singular vector(s) of the data; these
#' are returned to save on computation time if PMD will be run again.}
#' \item{meanx}{Mean of x that was subtracted out before PMD was performed.}
#' @seealso \link{PMD.cv}, \link{SPC}
#' @references
#' Witten D. M., Tibshirani R., and Hastie, T. (2009)
#' \emph{A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis}, \emph{Biostatistics, Gol 10 (3), 515-534, Jul 2009}\cr
#' @examples
#'
#'
#' # Try PMD with L1 penalty on rows and columns: type="standard"
#' # A simple simulated example
#' set.seed(1)
#' # Our data is a rank-one matrix, plus noise. The underlying components
#' # contain 50 and 75 non-zero elements, respectively.
#' u <- matrix(c(rnorm(50), rep(0,150)),
#' ncol=1)
#' v <- matrix(c(rnorm(75),rep(0,225)), ncol=1)
#' x <- u%*%t(v)+
#' matrix(rnorm(200*300),ncol=300)
#' # We can use cross-validation to try to find optimal value of sumabs
#' cv.out <- PMD.cv(x, type="standard", sumabss=seq(0.1, 0.6, len=20))
#' print(cv.out)
#' plot(cv.out)
#' # The optimal value of sumabs is 0.4157, but we can get within one
#' # standard error of that CV error using sumabs=0.337, which corresponds to
#' # an average of 45.8 and 71.8 non-zero elements in each component - pretty
#' # close to the true model.
#' # We can fit the model corresponding to the lowest cross-validation error:
#' out <- PMD(x, type="standard", sumabs=cv.out$bestsumabs, K=1, v=cv.out$v.init)
#' print(out)
#' par(mfrow=c(2,2))
#' par(mar=c(2,2,2,2))
#' plot(out$u[,1], main="Est. u")
#' plot(out$v[,1], main="Est. v")
#' plot(u, main="True u")
#' plot(v, main="True v")
#' # And if we want to control sumabsu and sumabsv separately, we can do
#' # that too. Let's get 2 components while we're at it:
#' out2 <- PMD(x, type="standard", K=2, sumabsu=6, sumabsv=8, v=out$v.init,
#' cnames=paste("v", sep=" ", 1:ncol(x)), rnames=paste("u", sep=" ", 1:nrow(x)))
#' print(out2)
#'
#' # Now check out PMD with L1 penalty on rows and fused lasso penalty on
#' # columns: type="ordered". We'll use the Chin et al (2006) Cancer Cell
#' # data set; try "?breastdata" for more info.
#' \dontrun{
#' breastdata <- download_breast_data()
#' with(breastdata, {
#' # dna contains CGH data and chrom contains chromosome of each CGH spot;
#' # nuc contains position of each CGH spot.
#' dna <- t(dna) # Need samples on rows and CGH spots on columns
#' # First, look for shared regions of gain/loss on chromosome 1.
#' # Use cross-validation to choose tuning parameter value
#' par(mar=c(2,2,2,2))
#' ch1 = which(chrom == 1)
#' cv.out <- PMD.cv(dna[, ch1],type="ordered",chrom=chrom[ch1],
#' nuc=nuc[ch1],
#' sumabsus=seq(1, sqrt(nrow(dna)), len=15))
#' print(cv.out)
#' plot(cv.out)
#' out <- PMD(dna[,chrom==1],type="ordered",
#' sumabsu=cv.out$bestsumabsu,chrom=chrom[chrom==1],K=1,v=cv.out$v.init,
#' cnames=paste("Pos",sep="",
#' nuc[chrom==1]), rnames=paste("Sample", sep=" ", 1:nrow(dna)))
#' print(out, verbose=TRUE)
#' # Which samples actually have that region of gain/loss?
#' par(mfrow=c(3,1))
#' par(mar=c(2,2,2,2))
#' PlotCGH(dna[which.min(out$u[,1]),chrom==1],chrom=chrom[chrom==1],
#' main=paste(paste(paste("Sample ", sep="", which.min(out$u[,1])),
#' sep="; u=", round(min(out$u[,1]),3))),nuc=nuc[chrom==1])
#' PlotCGH(dna[88,chrom==1], chrom=chrom[chrom==1],
#' main=paste("Sample 88; u=", sep="", round(out$u[88,1],3)),
#' nuc=nuc[chrom==1])
#' PlotCGH(out$v[,1],chrom=chrom[chrom==1], main="V",nuc=nuc[chrom==1])
#' } )
#' }
#' @export PMD
PMD <- function(x, type=c("standard", "ordered"), sumabs=.4, sumabsu=5, sumabsv=NULL, lambda=NULL, niter=20, K=1, v=NULL, trace=TRUE, center=TRUE, chrom=NULL, rnames=NULL, cnames=NULL, upos=FALSE, uneg=FALSE, vpos=FALSE, vneg=FALSE){
if(upos&&uneg || vpos&&vneg) stop("Cannot contrain elements to be both positive and negative!")
if(type=="ordered" && (vpos||vneg)) stop("Cannot constrain signs of v if type is ordered.")
call <- match.call()
type <- match.arg(type)
if(type=="standard"){
out <- PMDL1L1(x, sumabs=sumabs, sumabsu=sumabsu, sumabsv=sumabsv, niter=niter, K=K, v=v, trace=trace, center=center, rnames=rnames, cnames=cnames, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg)
class(out) <- "PMDL1L1"
} else if(type=="ordered"){
out <- PMDL1FL(x,K=K,sumabsu=sumabsu,lambda=lambda,chrom=chrom,niter=niter, v=v, trace=trace, center=center, rnames=rnames, cnames=cnames, upos=upos, uneg=uneg)
class(out) <- "PMDL1FL"
}
out$upos <- upos
out$uneg <- uneg
out$vpos <- vpos
out$vneg <- vneg
out$call <- call
return(out)
}
#' Do tuning parameter selection for PMD via cross-validation
#'
#' Performs cross-validation to select tuning parameters for rank-1 PMD, the
#' penalized matrix decomposition for a data matrix.
#'
#' If type is "standard", then lasso ($L_1$) penalties (promoting sparsity) are
#' placed on u and v. If type is "ordered", then lasso penalty is placed on u
#' and a fused lasso penalty (promoting sparsity and smoothness) is placed on
#' v.
#'
#' Cross-validation of the rank-1 PMD is performed over sumabss (if type is
#' "standard") or over sumabsus (if type is "ordered"). If type is "ordered",
#' then lambda is chosen from the data without cross-validation.
#'
#' The cross-validation works as follows: Some percent of the elements of $x$
#' is removed at random from the data matrix. The PMD is performed for a range
#' of tuning parameter values on this partially-missing data matrix; then,
#' missing values are imputed using the decomposition obtained. The value of
#' the tuning parameter that results in the lowest sum of squared errors of the
#' missing values if "best".
#'
#' To do cross-validation on the rank-2 PMD, first the rank-1 PMD should be
#' computed, and then this function should be performed on the residuals, given
#' by $x-udv'$.
#'
#' @param x Data matrix of dimension $n x p$, which can contain NA for missing
#' values.
#' @param type "standard" or "ordered": Do we want v to simply be sparse, or
#' should it also be smooth? If the columns of x are ordered (e.g. CGH spots
#' along a chromosome) then choose "ordered". Default is "standard". If
#' "standard", then the PMD function will make use of sumabs OR
#' sumabsu&sumabsv. If "ordered", then the function will make use of sumabsu
#' and lambda.
#' @param sumabss Used only if type is "standard". A vector of sumabs values to
#' be used. Sumabs is a measure of sparsity for u and v vectors, between 0 and
#' 1. When sumabss is specified, and sumabsus and sumabsvs are NULL, then
#' sumabsus is set to $sqrt(n)*sumabss$ and sumabsvs is set at
#' $sqrt(p)*sumabss$. If sumabss is specified, then sumabsus and sumabsvs
#' should be NULL. Or if sumabsus and sumabsvs are specified, then sumabss
#' should be NULL.
#' @param sumabsus Used only for type "ordered". A vector of sumabsu values to
#' be used. Sumabsu measures sparseness of u - it is the sum of absolute values
#' of elements of u. Must be between 1 and sqrt(n).
#' @param lambda Used only if type is "ordered". This is the tuning parameter
#' for the fused lasso penalty on v, which takes the form $lambda ||v||_1 +
#' lambda |v_j - v_(j-1)|$. $lambda$ must be non-negative. If NULL, then it is
#' chosen adaptively from the data.
#' @param nfolds How many cross-validation folds should be performed? Default
#' is 5.
#' @param niter How many iterations should be performed. For speed, only 5 are
#' performed by default.
#' @param v The first right singular vector(s) of the data. (If missing data is
#' present, then the missing values are imputed before the singular vectors are
#' calculated.) v is used as the initial value for the iterative PMD algorithm.
#' If x is large, then this step can be time-consuming; therefore, if PMD is to
#' be run multiple times, then v should be computed once and saved.
#' @param chrom If type is "ordered", then this gives the option to specify
#' that some columns of x (corresponding to CGH spots) are on different
#' chromosomes. Then v will be sparse, and smooth *within* each chromosome but
#' not *between* chromosomes. Length of chrom should equal number of columns of
#' x, and each entry in chrom should be a number corresponding to which
#' chromosome the CGH spot is on.
#' @param nuc If type is "ordered", can specify the nucleotide position of each
#' CGH spot (column of x), to be used in plotting. If NULL, then it is assumed
#' that CGH spots are equally spaced.
#' @param trace Print out progress as iterations are performed? Default is
#' TRUE.
#' @param center Subtract out mean of x? Default is TRUE
#' @param upos Constrain the elements of u to be positive? TRUE or FALSE.
#' @param uneg Constrain the elements of u to be negative? TRUE or FALSE.
#' @param vpos Constrain the elements of v to be positive? TRUE or FALSE.
#' Cannot be used if type is "ordered".
#' @param vneg Constrain the elements of v to be negative? TRUE or FALSE.
#' Cannot be used if type is "ordered."
#' @return \item{cv}{Average sum of squared errors obtained over
#' cross-validation folds.} \item{cv.error}{Standard error of average sum of
#' squared errors obtained over cross-validation folds.} \item{bestsumabs}{If
#' type="standard", then value of sumabss resulting in smallest CV error is
#' returned.} \item{bestsumabsu}{If type="ordered", then value of sumabsus
#' resulting in smallest CV error is returned.} \item{v.init}{The first right
#' singular vector(s) of the data; these are returned to save on computation
#' time if PMD will be run again.}
#' @seealso \link{PMD}, \link{SPC}
#' @references
#' Witten D. M., Tibshirani R., and Hastie, T. (2009)
#' \emph{A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis}, \emph{Biostatistics, Gol 10 (3), 515-534, Jul 2009}\cr
#' @examples
#' # See examples in PMD help file
#' @export PMD.cv
PMD.cv <- function(x, type=c("standard", "ordered"), sumabss=seq(0.1,0.7,len=10), sumabsus=NULL, lambda=NULL, nfolds=5, niter=5, v=NULL, chrom=NULL, nuc=NULL, trace=TRUE, center=TRUE, upos=FALSE, uneg=FALSE, vpos=FALSE, vneg=FALSE){
if(upos&&uneg || vpos&&vneg) stop("Cannot contrain elements to be both positive and negative!")
if(type=="ordered" && (vpos||vneg)) stop("Cannot constrain signs of v if type is ordered.")
call <- match.call()
type <- match.arg(type)
if(type=="standard"){
out <- PMDL1L1.cv(x, sumabss=sumabss, nfolds=nfolds, niter=niter, v=v, trace=trace, center=center, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg)
class(out) <- "PMDL1L1.cv"
} else if(type=="ordered"){
out <- PMDL1FL.cv(x, nfolds=nfolds, niter=niter, lambda=lambda, sumabsus=sumabsus, chrom=chrom, v=v, trace=trace, nuc=nuc, center=center, upos=upos, uneg=uneg)
class(out) <- "PMDL1FL.cv"
}
out$upos <- upos
out$uneg <- uneg
out$vpos <- vpos
out$vneg <- vneg
out$call <- call
return(out)
}
PMDL1L1 <- function(x,sumabs=.4,sumabsu=NULL,sumabsv=NULL,niter=20,K=1,v=NULL, trace=TRUE, orth=FALSE, center=TRUE, rnames=NULL, cnames=NULL, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg){
if(center){
meanx <- mean_na(x)
x <- x-meanx
} else {
meanx <- NULL
}
if(orth){
if(is.null(sumabsu) || sumabsu < sqrt(nrow(x))){
orth <- FALSE
warning("Orth option ignored because sparse PCA results only when sumabsu equals sqrt(nrow(x))")
}
}
# if((is.null(sumabsu) && !is.null(sumabsv)) || (is.null(sumabsv) && !is.null(sumabsu))) warning("Sumabsu and sumabsv BOTH must be input when type=standard. Since only one was given, it was ignored and sumabs was used instead.")
if(is.null(sumabsu) || is.null(sumabsv)){
sumabsu <- sqrt(nrow(x))*sumabs
sumabsv <- sqrt(ncol(x))*sumabs
}
call <- match.call()
if(trace && abs(mean_na(x)) > 1e-15) warning("PMDL1L1 was run without first subtracting out the mean of x.")
if(!is.null(sumabsu) && (sumabsu<1 || sumabsu>sqrt(nrow(x)))) stop("sumabsu must be between 1 and sqrt(n)")
if(!is.null(sumabsv) && (sumabsv<1 || sumabsv>sqrt(ncol(x)))) stop("sumabsv must be between 1 and sqrt(p)")
v <- CheckPMDV(v,x,K)
if(K>1 && !orth) out <- (MultiSMD(x,sumabsu=sumabsu,sumabsv=sumabsv,niter=niter,K=K, trace=trace, v=v, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg))
if(K>1 && orth) out <- MultiSMDOrth(x,sumabsu=sumabsu,sumabsv=sumabsv,niter=niter,K=K, trace=trace, v=v, vpos=vpos, vneg=vneg)
if(K==1) out <- SMD(x,sumabsu=sumabsu,sumabsv=sumabsv,niter=niter, trace=trace, v=v, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg)
obj <- (list(u=out$u,v=out$v, d=out$d, v.init=out$v.init, call=call, meanx=meanx,sumabsu=sumabsu, sumabsv=sumabsv, rnames=rnames, cnames=cnames, K=K, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg))
class(obj) <- "PMDL1L1"
return(obj)
}
#' @method print PMDL1L1
#' @export
print.PMDL1L1 <- function(x,verbose=FALSE,...){
cat("Call: ")
dput(x$call)
cat("\n\n")
cat("Num non-zero u's: ", apply(x$u!=0, 2, sum), "\n")
cat("Num non-zero v's: ", apply(x$v!=0, 2, sum), "\n")
cat("Sumabsu used: ", x$sumabsu, "\n")
cat("Sumabsv used: ", x$sumabsv, "\n")
if(!is.null(x$meanx)) cat("Subtracted out mean of x: ", x$meanx, "\n")
if(verbose){
for(k in 1:x$K){
cat("\n Component ", k, ":\n")
u <- x$u[,k]
v <- x$v[,k]
if(is.null(x$rnames)) x$rnames <- 1:length(u)
if(is.null(x$cnames)) x$cnames <- 1:length(v)
cat(fill=T)
us <- cbind(x$rnames[u!=0], round(u[u!=0],3))
dimnames(us) <- list(1:sum(u!=0), c("Row Feature Name", "Row Feature Weight"))
vs <- cbind(x$cnames[v!=0], round(v[v!=0],3))
dimnames(vs) <- list(1:sum(v!=0), c("Column Feature Name", "Column Feature Weight"))
print(us, quote=FALSE, sep="\t")
cat(fill=T)
print(vs, quote=FALSE, sep="\t")
}
}
if(x$upos) cat("Elements of u constrained to be positive.", fill=TRUE)
if(x$vpos) cat("Elements of v constrained to be positive.", fill=TRUE)
if(x$uneg) cat("Elements of u constrained to be negative.", fill=TRUE)
if(x$vneg) cat("Elements of v constrained to be negative.", fill=TRUE)
}
#' @method print PMDL1FL
#' @export
print.PMDL1FL <- function(x,verbose=FALSE,...){
cat("Call: ")
dput(x$call)
cat("\n\n")
cat("Num non-zero u's: ", apply(x$u!=0, 2, sum), "\n")
cat("Num non-zero v's: ", apply(x$v!=0, 2, sum), "\n")
cat("Sumabsu used: ", x$sumabsu, "\n")
cat("Lambda used: ", x$lambda, "\n")
if(!is.null(x$meanx)) cat("Subtracted out mean of x: ", x$meanx, "\n")
if(verbose){
for(k in 1:x$K){
cat("\n Component ", k, ":\n")
u <- x$u[,k]
v <- x$v[,k]
if(is.null(x$rnames)) x$rnames <- 1:length(u)
if(is.null(x$cnames)) x$cnames <- 1:length(v)
cat(fill=T)
us <- cbind(x$rnames[u!=0], round(u[u!=0],3))
dimnames(us) <- list(1:sum(u!=0), c("Row Feature Name", "Row Feature Weight"))
vs <- cbind(x$cnames[v!=0], round(v[v!=0],3))
dimnames(vs) <- list(1:sum(v!=0), c("Column Feature Name", "Column Feature Weight"))
print(us, quote=FALSE, sep="\t")
cat(fill=T)
print(vs, quote=FALSE, sep="\t")
}
}
if(x$upos) cat("Elements of u constrained to be positive.", fill=TRUE)
if(x$uneg) cat("Elements of u constrained to be negative.", fill=TRUE)
}
MultiSMDOrth <- function(x,sumabsu,sumabsv,niter,K, trace, v, vpos, vneg){
if(sumabsu < sqrt(nrow(x))){
warning("sumabsu was less than sqrt(nrow(x)), so the orthogonal option was not implemented.")
return(MultiSMD(x,sumabsu=sumabsu,sumabsv=sumabsv,niter=niter,K=K, trace=trace, v=v, vpos=vpos, vneg=vneg))
}
nas <- is.na(x)
v.init <- v
xuse <- x
ds <- numeric(K)
us <- matrix(0,nrow=nrow(x),ncol=K)
vs <- matrix(0,nrow=ncol(x),ncol=K)
for(k in 1:K){
if(k==1) out <- SMD(xuse,sumabsu=sumabsu,sumabsv=sumabsv,niter=niter,v=matrix(v[,k],ncol=1), trace=trace, upos=FALSE, uneg=FALSE, vpos=vpos, vneg=vneg)
if(k>1) out <- SMDOrth(xuse,us[,1:(k-1)],sumabsv=sumabsv,niter=niter,v=matrix(v[,k],ncol=1), trace=trace, vpos=vpos, vneg=vneg)
us[,k] <- out$u
vs[,k] <- out$v
ds[k] <- out$d
}
return(list(u=us,v=vs,d=ds, v.init=v.init))
}
MultiSMD <- function(x, sumabsu, sumabsv, K=3, niter=20,v, trace=TRUE, upos, uneg, vpos, vneg){
nas <- is.na(x)
v.init <- v
xuse <- x
ds <- numeric(K)
us <- matrix(0,nrow=nrow(x),ncol=K)
vs <- matrix(0,nrow=ncol(x),ncol=K)
for(k in 1:K){
out <- SMD(xuse, sumabsu=sumabsu,sumabsv=sumabsv,niter=niter,v=matrix(v[,k],ncol=1), trace=trace, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg)
us[,k] <- out$u
vs[,k] <- out$v
ds[k] <- out$d
res <- xuse - out$d*out$u%*%t(out$v)
xuse[!nas] <- res[!nas] # [!nas] is new on July 24 2009
}
return(list(u=us,v=vs,d=ds, v.init=v.init))
}
PMDL1L1.cv <- function(x, sumabss=seq(0.1,0.7,len=10), nfolds=5, niter=5, v=NULL, trace=TRUE, center=TRUE, upos, uneg, vpos, vneg){
if(nfolds<2) stop("Must run at least 2 cross-validation folds.")
percentRemove <- min(.25, 1/nfolds)
call <- match.call()
if(max(sumabss)>1 || min(sumabss)<0) stop("sumabs must be between 0 and 1")
# PMD only cross-validates the single-factor model.
# and it requires just one sparsity measure.
xfill <- x
missing <- is.na(x)
v <- CheckPMDV(v,x,K=1)
errs <- matrix(NA, nrow=nfolds, ncol=length(sumabss))
nonzerous <- matrix(NA, nrow=nfolds, ncol=length(sumabss))
nonzerovs <- matrix(NA, nrow=nfolds, ncol=length(sumabss))
rands <- matrix(runif(nrow(x)*ncol(x)),ncol=ncol(x))
for(i in 1:nfolds){
if(trace) cat(" Fold ", i, " out of ", nfolds, "\n")
rm <- ((i-1)*percentRemove < rands)&(rands < i*percentRemove)
xrm <- x
xrm[rm] <- NA
for(j in 1:length(sumabss)){
out <- PMDL1L1(xrm, sumabs=sumabss[j], niter=niter, v=v, trace=FALSE, center=center,K=1, upos=upos, uneg=uneg, vpos=vpos, vneg=vneg)
xhat <- as.numeric(out$d)*out$u%*%t(out$v)
errs[i,j] <- sum(((xhat-x)[rm & !missing])^2)
nonzerous[i,j] <- sum(out$u!=0)
nonzerovs[i,j] <- sum(out$v!=0)
}
}
if(trace) cat(fill=TRUE)
err.means <- apply(errs, 2, mean)
err.sds <- apply(errs, 2, sd)/sqrt(nfolds)
nonzerous.mean <- apply(nonzerous,2,mean)
nonzerovs.mean <- apply(nonzerovs,2,mean)
bestsumabs <- sumabss[which.min(err.means)]
object <- (list(cv=err.means, cv.error=err.sds,bestsumabs=bestsumabs, nonzerous=nonzerous.mean, nonzerovs=nonzerovs.mean, v.init=v, call=call, sumabss=sumabss, nfolds=nfolds))
class(object) <- "PMDL1L1.cv"
return(object)
}
#' Perform cross-validation on sparse principal component analysis
#'
#' Selects tuning parameter for the sparse principal component analysis method
#' of Witten, Tibshirani, and Hastie (2008), which involves applying PMD to a
#' data matrix with lasso ($L_1$) penalty on the columns and no penalty on the
#' rows. The tuning parameter controls the sum of absolute values - or $L_1$
#' norm - of the elements of the sparse principal component.
#'
#' This method only performs cross-validation for the first sparse principal
#' component. It does so by performing the following steps nfolds times: (1)
#' replace a fraction of the data with missing values, (2) perform SPC on this
#' new data matrix using a range of tuning parameter values, each time getting
#' a rank-1 approximationg $udv'$ where $v$ is sparse, (3) measure the mean
#' squared error of the rank-1 estimate of the missing values created in step
#' 1.
#'
#' Then, the selected tuning parameter value is that which resulted in the
#' lowest average mean squared error in step 3.
#'
#' In order to perform cross-validation for the second sparse principal
#' component, apply this function to $X-udv'$ where $udv'$ are the output of
#' running SPC on the raw data $X$.
#'
#' @aliases SPC.cv plot.SPC.cv print.SPC.cv
#' @param x Data matrix of dimension $n x p$, which can contain NA for missing
#' values. We are interested in finding sparse principal components of
#' dimension $p$.
#' @param sumabsvs Range of sumabsv values to be considered in
#' cross-validation. Sumabsv is the sum of absolute values of elements of v. It
#' must be between 1 and square root of number of columns of data. The smaller
#' it is, the sparser v will be.
#' @param nfolds Number of cross-validation folds performed.
#' @param niter How many iterations should be performed. By default, perform
#' only 5 for speed reasons.
#' @param v The first right singular vector(s) of the data. (If missing data is
#' present, then the missing values are imputed before the singular vectors are
#' calculated.) v is used as the initial value for the iterative PMD($L_1$,
#' $L_1$) algorithm. If x is large, then this step can be time-consuming;
#' therefore, if PMD is to be run multiple times, then v should be computed
#' once and saved.
#' @param trace Print out progress as iterations are performed? Default is
#' TRUE.
#' @param orth If TRUE, then use method of Section 3.2 of Witten, Tibshirani
#' and Hastie (2008) to obtain multiple sparse principal components. Default is
#' FALSE.
#' @param center Subtract out mean of x? Default is TRUE
#' @param vpos Constrain elements of v to be positive? Default is FALSE.
#' @param vneg Constrain elements of v to be negative? Default is FALSE.
#' @return \item{cv}{Average sum of squared errors that results for each tuning
#' parameter value.} \item{cv.error}{Standard error of the average sum of
#' squared error that results for each tuning parameter value.}
#' \item{bestsumabsv}{Value of sumabsv that resulted in lowest CV error.}
#' \item{nonzerovs}{Average number of non-zero elements of v for each candidate
#' value of sumabsvs.} \item{v.init}{Initial value of v that was passed in. Or,
#' if that was NULL, then first right singular vector of X.}
#' \item{bestsumabsv1se}{The smallest value of sumabsv that is within 1
#' standard error of smallest CV error.}
#' @seealso \link{SPC}, \link{PMD}, \link{PMD.cv}
#' @references
#' Witten D. M., Tibshirani R., and Hastie, T. (2009)
#' \emph{A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis}, \emph{Biostatistics, Gol 10 (3), 515-534, Jul 2009}\cr
#' @examples
#'
#' #NOT RUN
#' ## A simple simulated example
#' #set.seed(1)
#' #u <- matrix(c(rnorm(50), rep(0,150)),ncol=1)
#' #v <- matrix(c(rnorm(75),rep(0,225)), ncol=1)
#' #x <- u%*%t(v)+matrix(rnorm(200*300),ncol=300)
#' ## Perform Sparse PCA - that is, decompose a matrix w/o penalty on rows
#' ## and w/ L1 penalty on columns
#' ## First, we perform sparse PCA and get 4 components, but we do not
#' ## require subsequent components to be orthogonal to previous components
#' #cv.out <- SPC.cv(x, sumabsvs=seq(1.2, sqrt(ncol(x)), len=6))
#' #print(cv.out)
#' #plot(cv.out)
#' #out <- SPC(x,sumabsv=cv.out$bestsumabs, K=4) # could use
#' ## cv.out$bestsumabvsv1se instead
#' #print(out,verbose=TRUE)
#' ## Now, we do sparse PCA using method in Section 3.2 of WT&H(2008) for getting
#' ## multiple components - that is, we require components to be orthogonal
#' #cv.out <- SPC.cv(x, sumabsvs=seq(1.2, sqrt(ncol(x)), len=6), orth=TRUE)
#' #print(cv.out)
#' #plot(cv.out)
#' #out.orth <- SPC(x,sumabsv=cv.out$bestsumabsv, K=4, orth=TRUE)
#' #print(out.orth,verbose=TRUE)
#' #par(mfrow=c(1,1))
#' #plot(out$u[,1], out.orth$u[,1], xlab="", ylab="")
#' #
#' #
#' @export SPC.cv
SPC.cv <- function(x, sumabsvs=seq(1.2,5,len=10), nfolds=5, niter=5, v=NULL, trace=TRUE, orth=FALSE, center=TRUE, vpos=FALSE, vneg=FALSE){
if(vpos&&vneg) stop("Cannot constrain elements of v to be both positive and negative.")
if(nfolds<2) stop("Must run at least 2 cross-validation folds.")
percentRemove <- min(.25, 1/nfolds)
call <- match.call()
if(max(sumabsvs)>sqrt(ncol(x)) || min(sumabsvs)<1) stop("sumabs must be between 1 and sqrt(ncol(x))")
xfill <- x
missing <- is.na(x)
v <- CheckPMDV(v,x,K=1)
errs <- matrix(NA, nrow=nfolds, ncol=length(sumabsvs))
nonzerovs <- matrix(NA, nrow=nfolds, ncol=length(sumabsvs))
rands <- matrix(runif(nrow(x)*ncol(x)),ncol=ncol(x))
for(i in 1:nfolds){
if(trace) cat(" Fold ", i, " out of ", nfolds, "\n")
rm <- ((i-1)*percentRemove < rands)&(rands < i*percentRemove)
xrm <- x
xrm[rm] <- NA
for(j in 1:length(sumabsvs)){
out <- SPC(xrm, sumabsv=sumabsvs[j], orth=orth, niter=niter, v=v, trace=FALSE, center=center,K=1, vpos=vpos, vneg=vneg)
xhat <- as.numeric(out$d)*out$u%*%t(out$v)
errs[i,j] <- sum(((xhat-x)[rm & !missing])^2)
nonzerovs[i,j] <- sum(out$v!=0)
}
}
if(trace) cat(fill=TRUE)
err.means <- apply(errs, 2, mean)
err.sds <- apply(errs, 2, sd)/sqrt(nfolds)
nonzerovs.mean <- apply(nonzerovs,2,mean)
bestsumabsv <- sumabsvs[which.min(err.means)]
bestsumabsv1se <- sumabsvs[min(which(err.means < min(err.means) + err.sds[which.min(err.means)]))]
object <- (list(cv=err.means, cv.error=err.sds,bestsumabsv=bestsumabsv,nonzerovs=nonzerovs.mean, v.init=v, call=call, sumabsvs=sumabsvs, nfolds=nfolds, bestsumabsv1se=bestsumabsv1se, vpos=vpos, vneg=vneg))
class(object) <- "SPC.cv"
return(object)
}
#' @method plot SPC.cv
#' @export
plot.SPC.cv <- function(x,...){
sumabsvs <- x$sumabsvs
err.means <- x$cv
err.sds <- x$cv.error
plot(sumabsvs, err.means, ylim=range(c(err.means+err.sds, err.means-err.sds)), main="CV Error for SPC", xlab="Value of sumabsv", ylab="Average SSE", type="l")
points(sumabsvs, err.means)
lines(sumabsvs, err.means+err.sds, lty="dashed")
lines(sumabsvs, err.means-err.sds, lty="dashed")
}
#' @method print SPC.cv
#' @export
print.SPC.cv <- function(x,...){
cat("Call:\n")
dput(x$call)
cat("\n \n")
cat('Cross-validation errors: \n')
mat <- round(cbind(round(x$sumabsvs, 3), x$cv, x$cv.error, x$nonzerovs),3)
dimnames(mat) <- list(1:length(x$sumabsvs), c("Sumabsvs", "CV Error", "CV S.E.", "# non-zero v's"))
print(mat, quote=F)
cat('\n Best sumabsv value (lowest CV error): ', x$bestsumabsv, "\n")
cat('\n Smallest sumabsv value that has CV error within 1 SE of best CV error: ', x$bestsumabsv1se, "\n")
if(x$vpos) cat("Elements of v constrained to be positive.", fill=TRUE)
if(x$vneg) cat("Elements of v constrained to be negative.", fill=TRUE)
}
#' @method print PMDL1L1.cv
#' @export
print.PMDL1L1.cv <- function(x,...){
cat("Call:\n")
dput(x$call)
cat("\n \n")
cat('Cross-validation errors: \n')
mat <- round(cbind(round(x$sumabss, 3), x$cv, x$cv.error, x$nonzerous, x$nonzerovs),3)
dimnames(mat) <- list(1:length(x$sumabss), c("Sumabss", "CV Error", "CV S.E.", "# non-zero u's", "# non-zero v's"))
print(mat, quote=F)
cat('\n Best sumabs value (lowest CV error): ', x$bestsumabs, "\n")
if(x$upos) cat("Elements of u constrained to be positive.", fill=TRUE)
if(x$uneg) cat("Elements of u constrained to be negative.", fill=TRUE)
if(x$vpos) cat("Elements of v constrained to be positive.", fill=TRUE)
if(x$vneg) cat("Elements of v constrained to be negative.", fill=TRUE)
}
#' @method plot PMDL1L1.cv
#' @export
plot.PMDL1L1.cv <- function(x,...){
sumabss <- x$sumabss
err.means <- x$cv
err.sds <- x$cv.error
plot(sumabss, err.means, ylim=range(c(err.means+err.sds, err.means-err.sds)), main="CV Error for PMD(L1,L1)", xlab="Value of sumabs", ylab="Average SSE", type="l")
points(sumabss, err.means)
lines(sumabss, err.means+err.sds, lty="dashed")
lines(sumabss, err.means-err.sds, lty="dashed")
}
PMDL1FL.cv <- function(x, nfolds=5, niter=5, lambda=NULL, sumabsus=NULL, chrom=NULL, v=NULL, trace=TRUE, nuc=NULL, center=TRUE, upos, uneg){
if(is.null(sumabsus)) sumabsus <- seq(1, .7*sqrt(nrow(x)),len=10)
percentRemove <- min(.25, 1/nfolds)
call <- match.call()
if(is.null(chrom)) chrom <- rep(1, ncol(x))
v <- CheckPMDV(v,x,K=1)
if(is.null(lambda)) lambda <- ChooseLambda1Lambda2(as.numeric(v[,1]))
errs <- matrix(NA, nrow=nfolds, ncol=length(sumabsus))
nonzerous <- matrix(NA, nrow=nfolds, ncol=length(sumabsus))
nonzerovs <- matrix(NA, nrow=nfolds, ncol=length(sumabsus))
rands <- matrix(runif(nrow(x)*ncol(x)), nrow=nrow(x))
missing <- is.na(x)
for(i in 1:nfolds){
if(trace) cat(" Fold ", i, " out of ", nfolds, "\n")
rm <- ((i-1)*percentRemove < rands)&(rands < i*percentRemove)
xrm <- x
xrm[rm] <- NA
for(j in 1:length(sumabsus)){
if(trace) cat(j,fill=F)
out <- PMDL1FL(xrm, K=1, lambda=lambda,sumabsu=sumabsus[j], niter=niter, chrom=chrom, v=v, trace=FALSE, center=center, upos=upos, uneg=uneg)
xhat <- as.numeric(out$d)*out$u%*%t(out$v)
errs[i,j] <- sum(((xhat-x)[rm & !missing])^2)
nonzerous[i,j] <- sum(out$u!=0)
nonzerovs[i,j] <- sum(out$v!=0)
}
}
if(trace) cat(fill=TRUE)
err.means <- apply(errs, 2, mean)
err.sds <- apply(errs, 2, sd)/sqrt(nfolds)
nonzerous.mean <- apply(nonzerous,2,mean)
nonzerovs.mean <- apply(nonzerovs,2,mean)
bestind <- which.min(err.means)
bestsumabsu <- sumabsus[bestind]
object <- (list(cv=err.means, lambda=lambda,cv.error=err.sds,bestsumabsu=bestsumabsu, nonzerous=nonzerous.mean, nonzerovs=nonzerovs.mean, v.init=v, call=call, sumabsus=sumabsus, nfolds=nfolds,x=x,chrom=chrom,nuc=nuc, niter=niter))
class(object) <- "PMDL1FL.cv"
return(object)
}
#' @method plot PMDL1FL.cv
#' @export
plot.PMDL1FL.cv <- function(x,...){
mat <- x$x
sumabsus <- x$sumabsus
err.means <- x$cv
err.sds <- x$cv.error
chrom <- x$chrom
niter <- x$niter
nuc <- x$nuc
lambda <- x$lambda
v <- x$v.init
bestsumabsu <- x$bestsumabsu
par(mfrow=c(2,1))
plot(sumabsus, err.means, ylim=range(c(err.means+err.sds, err.means-err.sds)), main="CV Error for PMD(L1, FL)", xlab="Value of sumabsu", ylab="Average SSE", type="l")
points(sumabsus, err.means)
lines(sumabsus, err.means+err.sds, lty="dashed")
lines(sumabsus, err.means-err.sds, lty="dashed")
out <- PMDL1FL(mat, sumabsu=bestsumabsu, K=1, chrom=chrom, niter=niter, v=v, trace=FALSE, lambda=lambda, upos=x$upos, uneg=x$uneg)
PlotCGH(out$v, main="Best V", chrom=chrom, nuc=nuc)
}
#' @method print PMDL1FL.cv
#' @export
print.PMDL1FL.cv <- function(x,...){
cat("Call:\n")
dput(x$call)
cat("\n \n")
cat('Cross-validation errors: \n')
mat <- round(cbind(round(x$sumabsus, 3), x$cv, x$cv.error, x$nonzerous, x$nonzerovs),3)
dimnames(mat) <- list(1:length(x$sumabsus), c("Sumabsus", "CV Error", "CV S.E.", "# non-zero u's", "# non-zero v's"))
print(mat, quote=F)
cat('\n Best sumabs value : ', x$bestsumabsu, "\n")
cat('\n Lambda Used : ', x$lambda, "\n")
if(x$upos) cat("Elements of u constrained to be positive.", fill=TRUE)
if(x$uneg) cat("Elements of u constrained to be negative.", fill=TRUE)
}
Try the PMA package in your browser
Any scripts or data that you put into this service are public.
R Package Documentation
Browse R Packages
We want your feedback!
Note that we can't provide technical support on individual packages. You should contact the package authors for that.
Embedding an R snippet on your website
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.