Skip to contents

Simulate an NMF dataset

Source: R/simulateNMF.R
simulateNMF.Rd

Generate a random nonnegative matrix with known factor structure for benchmarking NMF recovery. Uses block-diagonal construction: each factor owns a disjoint subset of features (rows) and dominates a disjoint subset of samples (columns), with small cross-talk for realism. This produces clearly recoverable factors even at moderate noise levels. Inspired by NMF::syntheticNMF.

Usage

simulateNMF(nrow, ncol, k, noise = 0.5, dropout = 0, seed = NULL)

Arguments

nrow

number of rows (features)

ncol

number of columns (samples)

k

true rank (number of factors)

noise

noise level as a multiplier on the mean signal. A value of 1.0 means the noise standard deviation equals the mean signal value. Default: 0.5.

dropout

fraction of entries to set to zero (0 = no dropout). Default: 0.

seed

seed for random number generation

Value

list of dense matrix A and true w and h models

See also

simulateSwimmer, nmf

Examples

data <- simulateNMF(50, 30, k = 3, noise = 0.1, seed = 42)
dim(data$A)  # 50 x 3
#> [1] 50 30
dim(data$w)  # 50 x 3
#> [1] 50  3
dim(data$h)  # 3 x 30
#> [1]  3 30