Repex:
Suppose I have the dtm:
library(topicmodels)
data(AssociatedPress)
I am trying to assign a value of .001 to all those values that are 0
Use case:
I get this error when I run LDA on my matrix
Error in LDA(notSparse, k, method = "Gibbs", control = list(nstart = nstart, : Each row of the input matrix needs to contain at least one non-zero entry
I would like to see what happens if I turn the zeros into small values to reduce sparsity instead of using the dedicated function.
Changing your matrix values from 0 to 0.001 will not work with topicmodels::LDA.
There is a check in the code that expects all values to be integer values. Which means values of 0.001 are not allowed. See example below:
m_replaced_zero <- matrix(c(1, 1, 0.001, 0), nrow = 2)
LDA(m_replaced_zero)
Error in !all.equal(x$v, as.integer(x$v)) : invalid argument type
Weirdly the error you get means that you have a row in your matrix that only contains 0. Which shouldn't happen unless you removed some terms from your documenttermmatrix which resulted in a row where there is no value of 1 or more. See example below.
m_zero_row <- matrix(c(1, 0, 1, 0), nrow = 2)
[,1] [,2]
[1,] 1 1
[2,] 0 0
LDA(m_zero_row)
Error in LDA(m_zero_row) :
Each row of the input matrix needs to contain at least one non-zero entry
But if you are set on replacing the sparse entries in a documenttermmatrix, you first need to transform it into a matrix and then replace the 0's.
data("AssociatedPress")
m <- as.matrix(AssociatedPress)
m[m==0] <- 0.001