I am doing Non-Negative Matrix Factorization (NMF) of a matrix A in R. It has Genes on rows and Samples on the columns. For NMF, I am using the CRAN package NMF. Once the basis matrix W and coefficient matrix H are computed, I want to check whether the factorisation was accurate enough. For that, I am trying to calculate the dot product of W and H so as to check whether I can get the original matrix A back.
Although the inner dimensions of basis matrix W and coefficient matrix H are same (rank = 3), I cannot multiply them. I keep getting the following error. This one line error is not very helpful.
Error in x * y : non-conformable arrays
Reproducible code
# Make a 10 * 10 matrix
ranVal= sample(1:10,100,replace =T)
M=matrix(ranVal,nrow=10)
# Rename the rows and columns
rownames(M) <- paste(rep("gene", 10),c(1:10), sep = "_")
colnames(M) <- paste(rep("sample", 10),c(1:10), sep = "_")
# See how it looks
M
# NMF
library(NMF)
tempRes <- nmf(x = M, rank = 3, .options = c("pv"))
# NMF implementation from Brunet et al. (2004)
result <- nmf_update.brunet_R(i = 1, v = M, x = tempRes, eps=.Machine$double.eps)
# Basis Matrix
W <- .basis(result);
# Coeff Matrix
H <- .coef(result)
# Checking dimensions
dim(W)
dim(H)
# Dot product
A <- W*H # Error: non-conformable arrays
A <- geometry::dot(W,H) # Error: non-conformable arrays
How should I go about it? Is there a different way to check it? Thanks!
What about
W %*% H?