gnmf (version 0.7.1)

gnmf: Generalized non-negative matrix factorization based on Renyi Divergence

Description

Performs generalized non-negative matrix factorization based on Renyi Divergence

Usage

gnmf(V, scheme, nsteps = 2000, repeats = 20, ranks = 2, cltarget = "PATTERN",
clscheme = "Binary", reffile = "", scaling = "F", normalizing = "F", alphas = 1,
runtype = "simulation", cstepsize = 20, idealization = 1)

Arguments

V

Input data matrix

scheme

KL, Renyi, or ED

nsteps

Update steps, default 2000

repeats

Repeats, default 20

ranks

The number of components into which matrix V is to be factored, default 2 (a scalar)

cltarget

Clustering target, default 'PATTERN' (H matrix) either PATTERN or ALTERNATE

clscheme

Clustering scheme, default 'Binary', could be 'PearsonHC'

reffile

Default none

scaling

Boolean, default F

normalizing

Boolean, H matrix normalization, default 'F'

alphas

Renyi parameter, default 1.0 (a scalar), ignored if scheme is not Renyi

runtype

simulation (default) or evaluation or whole

cstepsize

Convergence test step size, default 20

idealization

Default 1

Value

H

List of pattern matrices, one for each repetition

W

List of amplitude matrices, one for each repetition

References

Devarajan K. Nonnegative matrix factorization: an analytical and interpretive tool in computational biology. PLoS Comput Biol. 2008 Jul 25;4(7):e1000029.

Devarajan, K., Wang, G., Ebrahimi, N. (2011). A unified approach to nonnegative matrix factorization and probabilistic latent semantic indexing, (July 2011). Cobra Preprint Series. Working Paper 80. http://biostats.bepress.com/COBRA/Art80.

http://devarajan.fccc.edu

Examples

# Load sample data.
data(V)

# Compute NMF with 20 repeats.
result <- gnmf(V,scheme="KL")

# Extract H and W from the result.
# H and W are lists, each containing the result of 20 repeats.
H <- result$H
W <- result$W

# Get the H and W matrices of the first repeat.
H1 <- H[[1]]
W1 <- W[[1]]