Compute dimensionality reduction for subsets of the input data

subsetDimReduction(Y, seed, size = 0.8, nrSubsets = 10, method,
  optN = NULL, ndim = NULL, kmin = 1, kmax = 40, verbose = FALSE,
  parallel = FALSE, is.list.ellipsis = FALSE, ...)

Arguments

Y	\[N x P\] data matrix for which the dimensionality of P should be reduced,
seed	[integer] seed to initialise random number generator for drawing subsets of Y.
size	[double] proportion of samples from total number of samples to to choose for each subset.
nrSubsets	[integer] number of subsets to generate and apply dimensionality reduction to.
method	dimensionality reduction method [character] to be applied; one of DiffusionMap, DRR, ICA, LLE, Isomap, LaplacianEigenmap, MDS, PCA, kPCA, nMDS, tSNE and PEER.
optN	optimal number [integer] of neighbours to consider for dimensionality reduction; relevant for methods LLE, LaplacianEigenmaps, Isomap and tSNE. If not provided, will be estimated via `calc_k`.
ndim	maximum dimensionality [int] to retain in the data; large values can cause long computation times; if not provided max(P,N) is chosen.
kmin	if optN is not provided, kmin [int] specifies the minimum number of neighbours supplied to `calc_k`.
kmax	if optN is not provided, kmax [int] specifies the maximum number of neighbours supplied to `calc_k`.
verbose	[logical] If set, progress messages are printed to standard out.
parallel	if optN is not provided and parallel TRUE, parallel computation on multiple cpu cores is used with `calc_k`.
is.list.ellipsis	[logical] if ... arguments are provided as list, set TRUE.
...	Additional arguments passed to dimensionality reduction methods. For possible arguments, check function decomentation. See details for relevant packages and functions.

Value

list of size nrSubsets, containing at each entry a named list of results from computeDimReduction: Y_red: named list with dimensionality reduced phenotypes (reducedY) and object returned by specified dimensionality reduction method (results) with additional output M: vector [double] with Trustworthiness and Continuity estimates for the dimensionality reduction

Details

subsetDimReduction wraps around the following implementations of the dimensionality reduction methods it provides: * Diffusion Map: diffuse * Dimensionality reduction by regression (DRR): drr * Independent component analysis (ICA): fastICA * Local liner embedding (LLE): lle * Isomap: isomap * Laplacian Eigenmap: spec.emb and make.kNNG * Multi-dimensional scaling (MDS) : cmdscale * Principal component analysis (PCA): prcomp * Kernel PCA (kPCA): kpca * non-metrix MDS (nMDS): metaMDS * t- stochastic neighbourhood embedding (tSNE): Rtsne * Uniform manifold approximation and projection (umap): umap * PEER: PEER