pca.scree.plot: Make scree plots for any PCA

In bigpca: PCA, Transpose and Multicore Functionality for 'big.matrix' Objects

Description

Make a scree plot using eigenvalues from princomp(), prcomp(), svd(), irlba(), big.PCA(), etc. Note that most these return values which need to be squared to be proper eigenvalues. There is also an option to use the estimate.eig.vpcs() function to estimate any missing eigenvalues (e.g, if using a function like irlba' to calculate PCA) and then to visualise the fitline of the estimate on the scree plot.

Usage

pca.scree.plot(eigenv, elbow = NA, printvar = TRUE, min.dim = NA,
  M = NULL, add.fit.line = FALSE, n.xax = max(30, length(eigenv)),
  linear = TRUE, verbose = FALSE, return.data = FALSE, ...)

Arguments

eigenv	the vector of eigenvalues actually calculated
elbow	the number of components which you think explain the important chunk of the variance of the dataset, so further components are modelled as reflecting noise or very subtle effects, e.g, often the number of components used is decided by the 'elbow' in a scree plot (see 'pca.scree.plot')
printvar	logical, whether to print summary of variance calculations
min.dim	the size of the smaller dimension of the matrix submitted to singular value decomposition, e.g, number of samples - i.e, the max number of possible eigenvalues, alternatively use 'M'.
M	optional enter the original dataset 'M'; simply used to derive the dimensions, alternatively use 'min.dim'.
add.fit.line	logical, if there is an existing scree plot, adds the fit line from this estimate to the plot ('pca.scree.plot' can use this option using the parameter of the same name)
n.xax	number of components to include on the x-axis
linear	whether to use a linear model to model the 'noise' eigenvalues; alternative is a 1/x model with no intercept.
verbose	logical, whether to display additional output
return.data	logical, whether to return the percentages of variance explained for each component, or nothing (just plot)
...	further arguments to the plot function

Value

Either a vector of variance percentages explained, or nothing (just a plot), depending on value of 'return.data'

See Also

pca.scree.plot

Examples

require(irlba)
nsamp <- 100; nvar <- 300; subset.size <- 25; elbow <- 6
mat <- matrix(rnorm(nsamp*nvar),ncol=nsamp) 
#this gives the full solution
pca <- svd(mat,nv=subset.size,nu=0)
pca2 <- irlba(mat,nv=subset.size,nu=0)
# show alternate fits for linear versus 1/x fit
pca.scree.plot((pca2$d^2)[1:subset.size],n.xax=100,add.fit.line=TRUE,
               min.dim=min(dim(mat)),linear=TRUE, elbow=6, ylim=c(0,1400))
pca.scree.plot((pca2$d^2)[1:subset.size],n.xax=100,add.fit.line=TRUE,
              min.dim=min(dim(mat)),linear=FALSE, elbow=40, ylim=c(0,1400))
subset.size <- 75
pca2 <- irlba(mat,nv=subset.size,nu=0)
pca.scree.plot((pca2$d^2)[1:subset.size],n.xax=100,add.fit.line=TRUE,
              min.dim=min(dim(mat)),linear=TRUE, elbow=6, ylim=c(0,1400))
pca.scree.plot((pca2$d^2)[1:subset.size],n.xax=100,add.fit.line=TRUE,
              min.dim=min(dim(mat)),linear=FALSE, elbow=40, ylim=c(0,1400))

bigpca documentation built on Nov. 22, 2017, 1:02 a.m.