从R中的矩阵中提取最大值（随机选择）

Question

Given a matrix, extracting the row names of the column with max values is a common problem.

sapply(mat,2,which.max)

mat<-matrix(list(20,0,0,80,80,0,
                 20,0,40,0,40,20,
                 40,0,40,20,20,0,
                 0,80,40,20,20,20),ncol=6,byrow=T)
rownames(mat)<-c("A","C","G","T")

But here, some columns have two similar max values (in the example matrix, col 3 and 4). By default the script chooses "A" has the row with the max column value in col 3 and 4. I am having trouble in writing a script to randomly select between two row names (A and T) wherein both have max values in column 3 and 4. Any help with the scripting is appreciated.

Answer 1

The rank function comes in handy:

> apply(mat,2,function(x) which(rank(-unlist(x), ties.method="random") == 1))
[1] 3 4 4 1 1 2
> apply(mat,2,function(x) which(rank(-unlist(x), ties.method="random") == 1))
[1] 3 4 3 1 1 2
> apply(mat,2,function(x) which(rank(-unlist(x), ties.method="random") == 1))
[1] 3 4 4 1 1 4

The ties.method="random" part is crucial for resolving the ties in a random fashion.

Answer 2

Consider reading the documentation for which.max , which suggests using which.is.max from nnet . Either borrow that algorithm or use that package.

> library(nnet)
> which.is.max
function (x) 
{
    y <- seq_along(x)[x == max(x)]
    if (length(y) > 1L) 
        sample(y, 1L)
    else y
}
<bytecode: 0x0000000013fda7c8>
<environment: namespace:nnet>

Answer 3

You could sample from those rownames which have values equal to the max value in that column:

mat<-matrix(c(20,0,0,80,80,0,
                 20,0,40,0,40,20,
                 40,0,40,20,20,0,
                 0,80,40,20,20,20),ncol=6,byrow=T)
rownames(mat)<-c("A","C","G","T")

set.seed(123)
apply( mat, 2 , function(x) sample( c( rownames(mat)[ which( x == max(x) ) ] ) , 1 ) )
#[1] "G" "T" "G" "A" "A" "C"

set.seed(1234)
apply( mat, 2 , function(x) sample( c( rownames(mat)[ which( x == max(x) ) ] ) , 1 ) )
#[1] "G" "T" "G" "A" "A" "T"

ps I'm not sure why you construct the matrix data usin a list object - matrices are vectors.