Creating a simple by pipe operator

by.R

# function definition
`%by%` <- function(a,b){
    s <- substitute(a)
    l <- as.list(s)
    p <- parse(text=s)
    if(length(p)>2) {
        l <- l[3:length(l)]
        do.call("by", c(list(data=eval.parent(p[2]), 
                             INDICES=b, 
                             FUN=eval.parent(p[1])), 
                         l))
    } else
        by(eval.parent(p[2]), INDICES=b, FUN=eval.parent(p[1]))
}


# some simple tests
set.seed(1)
n <- 100
d <- data.frame(y = rnorm(n), 
                x1 = sample(1:2, n, TRUE),
                x2 = sample(1:2, n, TRUE))


# simple example
#by(d$y, list(d$x1), FUN=sd)
sd(d$y) %by% d$x1

# right-hand side list
#by(d$y, list(d$x1, d$x2), FUN=mean)
mean(d$y) %by% list(d$x1, d$x2)

# with subsetting
#by(d$y[1:30], list(d$x1[1:30]), FUN=sd)
sd(d$y[1:30]) %by% d$x1[1:30]

# missing data
d2 <- d
d2$y[sample(10, 1:n)] <- NA
#by(d2$y, list(d2$x1), FUN=mean, na.rm=TRUE)
mean(d2$y, na.rm=TRUE) %by% d2$x1

# using a `with` statement
#by(d2$x1, list(d2$x2), FUN=table)
table(d2$x1) %by% d2$x2

#with(d2, by(x1, list(x2), FUN=table))
with(d2, table(x1) %by% x2)