从海量数据框中选择元素的更有效方法

Question

I have a huge data frame:

library(gtools)
a<-permutations(2,20,v=c(0,1),repeats.allowed=TRUE)
a<-as.data.frame(a)

And I have 100 random strings:

set.seed(123)

b<-replicate(100,sample(c(0,1),20, replace=T))

I would like to identify the row numbers in in 'a' that corresponds to each column in 'b' .

Since 'a' is huge this process takes quite some time.

right now I am using the following method:

sapply(1:100, function(x)  which(colSums(t(a)==as.numeric(b[,x]))==20L))

This process takes a lot of time. I was wondering if there is a more efficient way to do this?

Answer 1

Represent the columns as digits by thinking of them as bit strings, then use %in% for fast look-up

library(gtools)
a <- permutations(2,20,v=c(0,1),repeats.allowed=TRUE)
a <- as.data.frame(a)

set.seed(123)
b <- replicate(100, sample(c(0, 1), 20, replace=TRUE))

a1 <- colSums(t(a) * 2^(0:19))
b1 <- colSums(b * 2^(0:19))

which produces

> head(which(a1 %in% b1))
[1]  1191  9434 14502 19812 30619 34313

To deal with duplicates, consider this smaller example

b1 <- c(1, 3, 3, 5, 4)
a1 <- c(3, 4, 8)

Discover the unique b1 values, and create a list that maps from the unique values to the index in the original values

ub1 <- unique(b1)
umap <- unname(split(seq_along(b1), match(b1, ub1)))

Now match the a1 to the unique b1, decide which to keep (are not NA), and look up the matches in the unique map

m <- match(a1, ub1)
keep <- which(!is.na(m))
keepmap <- umap[m[keep]]

Finally, use keepmap to figure out how many times each kept value needs to be replicated (because it maps to multiple original values) and create a data.frame of the results

len <- sapply(keepmap, length)
data.frame(ai=rep(keep, len),
           a1=rep(a1[keep], len),
           b1=unlist(unname(keepmap)))

So a complete function is

matchrows <-
    function(a, b)
{
    ## encode
    a1 <- colSums(t(a) * 2^(0:19))
    b1 <- colSums(b * 2^(0:19))

    ## match to unique values
    ub1 <- unique(b1)
    m <- match(a1, ub1)
    keep <- which(!is.na(m))

    ## expand unique matches to original coordinates
    umap <- unname(split(seq_along(b1), match(b1, ub1)))
    keepmap <- umap[m[keep]]

    len <- sapply(keepmap, length)
    data.frame(ai=rep(keep, len),
               bi=unlist(unname(keepmap)),
               value=rep(a1[keep], len))
}