I have a large data set and I would like to read specific columns or drop all the others.
data <- read.dta("file.dta")
I select the columns that I'm not interested in:
var.out <- names(data)[!names(data) %in% c("iden", "name", "x_serv", "m_serv")]
and than I'd like to do something like:
for(i in 1:length(var.out)) {
paste("data$", var.out[i], sep="") <- NULL
}
to drop all the unwanted columns. Is this the optimal solution?
Do not use -which() for this, it is extremely dangerous. Consider:
dat <- data.frame(x=1:5, y=2:6, z=3:7, u=4:8)
dat[ , -which(names(dat) %in% c("z","u"))] ## works as expected
dat[ , -which(names(dat) %in% c("foo","bar"))] ## deletes all columns! Probably not what you wanted...
Instead use subset or the ! function:
dat[ , !names(dat) %in% c("z","u")] ## works as expected
dat[ , !names(dat) %in% c("foo","bar")] ## returns the un-altered data.frame. Probably what you want
I have learned this from painful experience. Do not overuse which()!
First, you can use direct indexing (with booleans vectors) instead of re-accessing column names if you are working with the same data frame; it will be safer as pointed out by Ista, and quicker to write and to execute. So what you will only need is:
var.out.bool <- !names(data) %in% c("iden", "name", "x_serv", "m_serv")
and then, simply reassign data:
data <- data[,var.out.bool] # or...
data <- data[,var.out.bool, drop = FALSE] # You will need this option to avoid the conversion to an atomic vector if there is only one column left
Second, quicker to write, you can directly assign NULL to the columns you want to remove:
data[c("iden", "name", "x_serv", "m_serv")] <- list(NULL) # You need list() to respect the target structure.
Finally, you can use subset(), but it cannot really be used in the code (even the help file warns about it). Specifically, a problem to me is that if you want to directly use the drop feature of susbset() you need to write without quotes the expression corresponding to the column names:
subset( data, select = -c("iden", "name", "x_serv", "m_serv") ) # WILL NOT WORK
subset( data, select = -c(iden, name, x_serv, m_serv) ) # WILL
As a bonus, here is small benchmark of the different options, that clearly shows that subset is the slower, and that the first, reassigning method is the faster:
re_assign(dtest, drop_vec) 46.719 52.5655 54.6460 59.0400 1347.331
null_assign(dtest, drop_vec) 74.593 83.0585 86.2025 94.0035 1476.150
subset(dtest, select = !names(dtest) %in% drop_vec) 106.280 115.4810 120.3435 131.4665 65133.780
subset(dtest, select = names(dtest)[!names(dtest) %in% drop_vec