将一个函数列表应用于矩阵，并在R中返回一个矩阵

Question

I have a matrix of n columns and m rows and a list of f functions. Each function takes one row of the matrix and returns a single value, p .

What is the best way to generate a f columns by m rows matrix?

At present I am doing this:

# create a random 5x5 matrix
m <- matrix(rexp(25, rate=.1), ncol=5)

# example functions, in reality more complex but with the same signature
fs <- list(function(xs) { return(mean(xs)) }, function(xs) { return(min(xs)) } )

# create a function which takes a function and applies it to each row of m
g <- function(f) { return(apply(m, 1, f)) }

# use lapply to make a call for each function in fs
# use do.call and cbind to reshape the output from a list of lists to a matrix
do.call("cbind", lapply(fs, g))

Clarification edit: the above code does work, but I'm wondering if there is a more elegant approach.

Answer 1

With base R you can do it in one line:

cbind(apply(m, 1, mean), apply(m, 1, min))

#          [,1]      [,2]
#[1,] 13.287748 5.2172657
#[2,]  5.855862 1.8346868
#[3,]  8.077236 0.4162899
#[4,] 10.422803 1.5899831
#[5,] 10.283001 2.0444687

this is faster than the do.call approach:

microbenchmark::microbenchmark(
  do.call("cbind", lapply(fs, g)),
  cbind(apply(m, 1, mean), apply(m, 1, min))
)

which yelds:

#Unit: microseconds
#                                       expr    min     lq     mean
#            do.call("cbind", lapply(fs, g)) 66.077 67.210 88.75483
# cbind(apply(m, 1, mean), apply(m, 1, min)) 57.771 58.903 67.70094
# median     uq     max neval
# 67.965 71.741 851.446   100
# 59.658 60.036 125.735   100

Answer 2

This is how I've adapted @patL's answer to take a list of functions:

# create a random 5x5 matrix
m <- matrix(rexp(25, rate=.1), ncol=5)

# example functions, in reality more complex but with the same signature
fs <- list(function(xs) { return(mean(xs)) }, function(xs) { return(min(xs)) } )

# create a function which takes a function and applies it to each row of m
g <- function(f) { return(apply(m, 1, f)) }

# use sapply to make a call for each function in fs
# use cbind to reshape the output from a list of lists to a matrix
cbind(sapply(fs, g))

I am using this to score an ensemble of models, for example:

# models is a list of trained models and m is a matrix of input data
g <- function(model) { return(predict(model, m)) }
# produce a matrix of model scores
cbind(sapply(models, g))

Answer 3

With data:

          mean   median        sd      max       min      sum
[1,]  9.299471 3.531394 10.436391 26.37984 1.7293010 46.49735
[2,]  8.583419 2.904223 11.714482 28.75344 0.7925614 42.91709
[3,]  6.292835 4.578894  6.058633 16.92280 1.8387221 31.46418
[4,] 10.699276 5.688477 15.161685 36.91369 0.1049507 53.49638
[5,]  9.767307 2.748114 10.767438 24.66143 1.5677153 48.83653

You can do:

 sapply(fs, mapply, split(m, row(m)), USE.NAMES = T) 

Which returns:

  mean median sd max min sum [1,] 9.299471 3.531394 10.436391 26.37984 1.7293010 46.49735 [2,] 8.583419 2.904223 11.714482 28.75344 0.7925614 42.91709 [3,] 6.292835 4.578894 6.058633 16.92280 1.8387221 31.46418 [4,] 10.699276 5.688477 15.161685 36.91369 0.1049507 53.49638 [5,] 9.767307 2.748114 10.767438 24.66143 1.5677153 48.83653 

Note:

It is the slowest one, comparing to both approaches proposed above.