CODEWITHSUNDEEP
rdata.table
MYaseen208
MYaseen208

Reputation: 23898

data.table: .SD for the following code

This is the follow up of this question. Wonders how to use .SD in this problem (rather than doing computations separately for each variable, in this case for Y1 and Y2 separately). 

set.seed(12345)
A <- rep(x=paste0("A", 1:2), each=6)
B <- rep(x=paste0("B", 1:3), each=2, times=2)
Rep <- rep(x=1:2, times=3)
Y1 <- rnorm(n=12, mean = 50, sd = 5)
Y2 <- rnorm(n=12, mean = 50, sd = 10)

library(data.table)
dt <- data.table(A, B, Rep, Y1, Y2)

dt[, j = Eff1 := mean(Y1), .(A, B)][, j = Eff1 := Eff1 - mean(Y1), .(A)][, j = Eff1 := Eff1 - mean(Y1), .(B)][, j = Eff1 := Eff1 + mean(Y1)]
dt[, j = Eff2 := mean(Y2), .(A, B)][, j = Eff2 := Eff2 - mean(Y2), .(A)][, j = Eff2 := Eff2 - mean(Y2), .(B)][, j = Eff2 := Eff2 + mean(Y2)]

dt[, j = .(Eff1 = mean(Eff1), Eff2 = mean(Eff2)), by = .(A, B)]

Upvotes: 1

Views: 80

Answers (1)

Frank
Frank

Reputation: 66819

Personally, I would consider going outside the data.table syntax, using ave:

my_cols  = c("Y1", "Y2")
tmp_cols = c("Eff1", "Eff2")

dt[, (tmp_cols) :=
    lapply(.SD, function(x) mean(x) + ave(x, A, B) - ave(x, A) - ave(x, B))
, .SDcols = my_cols][,
    lapply(.SD, mean)    
, by=A:B, .SDcols = tmp_cols]

One long way is:

dtA  = dt[, lapply(.SD, mean), by=A, .SDcols = my_cols]
dtB  = dt[, lapply(.SD, mean), by=B, .SDcols = my_cols]
dtAB = dt[, lapply(.SD, mean), by=.(A,B), .SDcols = my_cols]

dt[,    (tmp_cols) := lapply(.SD, mean), .SDcols = my_cols]
dt[dtAB,(tmp_cols) := Map(`+`, mget(tmp_cols), mget(paste0("i.", my_cols))), on=c("A","B")]
dt[dtA, (tmp_cols) := Map(`-`, mget(tmp_cols), mget(paste0("i.", my_cols))), on="A"]
dt[dtB, (tmp_cols) := Map(`-`, mget(tmp_cols), mget(paste0("i.", my_cols))), on="B"]

dt[, lapply(.SD, mean), by=.(A,B), .SDcols=tmp_cols]

Upvotes: 2

Related Questions