PNMFO: Projective orthonormal nonnegative matrix factorization based...

In richardbeare/pNMF: Projective non-negative matrix factorization

Description

Projective orthonormal nonnegative matrix factorization based on euclidean distance.

Usage

PNMFO(
  X,
  nmfMod,
  tol = 1e-05,
  maxIter = 5000,
  verboseN = FALSE,
  zerotol = 1e-10
)

PNMFO2(
  X,
  nmfMod,
  tol = 1e-05,
  maxIter = 5000,
  verboseN = FALSE,
  zerotol = 1e-10
)

Arguments

X	Input data matrix
nmfMod	NMF model from the NMF package
tol	tolerance for stopping criteria
maxIter	Maximum number of iterations
verbose	Print status messages

Details

Implementation of "Linear and Nonlinear Projective Nonnegative Matrix Factorization", Z. Yang and E. Oja, IEEE Transactions on Neural Networks. Derived from matlab code by Z. Yang, https://sites.google.com/site/zhirongyangcs/pnmf.

The PNMFO2 version uses a different ordering of matrix operations that is slower, as more happens in the loop, but reduces the maximum matrix size. No need for a features x features matrix

Value

Fitted NMF model, as defined in NMF package.

Functions

• PNMFO2: Projective orthonormal nonnegative matrix factorization based on euclidean distance.

Examples

library(NMF)
setNMFMethod("PNMFO", pNMF::PNMFO)
mkD <- function(NOISE=TRUE) {
  n <- 1000 # rows
  counts <- c(30, 10, 20, 10, 15, 15) # samples
  syntheticNMF(n=n, r=counts, offset = NULL, noise = NOISE,
               factors = FALSE, seed = 99)
}
k<-mkD()
estim <- nmf(k, 6, method="PNMF", nrun=1)
## Not run: 
V.random <- randomize(k)
estim.r2 <- nmf(k, 2:20, method="PNMF", nrun=30)
estim.r2.random <- nmf(V.random, 2:20,  method="PNMF", nrun=30)

## End(Not run)

richardbeare/pNMF documentation built on June 30, 2020, 6:33 p.m.