Factors to sparse matrix efficiently

fact_to_sparse.R

#' Creates sparse 0/1 matrix for factor variables
#' 
#' @param df dataframe containing factor columns that will be turned into sparse matrix.
#'    Note that the variables aren't required to be factor variables themselves. 
#'    Make sure not to include other variables in df
fact_to_sparse <- function(df) {

# Convert to factor variables
df[,names(df)] <- lapply(df[,names(df)] , factor)

# Number of levels for each variable
nlevels <- sapply(df, nlevels)

# Rows
i <- rep(seq_len(n), ncol(df))

# Columns
j <-
	# converts factor to number for each variable
	unlist(lapply(df, as.integer)) +
	# and shifts that number by the number of levels from previous variables
	rep(cumsum(c(0, head(nlevels, -1))), each = n)

# Value = 1
x <- 1

Matrix::sparseMatrix(i = i, j = j, x = x)

}



# ---- Example --------------------------------

n <- 1e6
df <- data.frame(
	x = factor(sample(c("A", "B", "C"), n, TRUE)), 
	y = factor(sample(c("D", "E"), n, TRUE))
)

mat <- fact_to_sparse(df)


# size comparison
pryr::object_size(mat)
pryr::object_size(matrix(mat))