Error when using the apply function on a matrix in R-CodePudding

I'm trying to use apply to use the L1 norm on each column of a matrix in R. If we have a matrix

X=matrix(data=rnorm(100),nrow=10,ncol=10)

Then when I run

apply(X,2,function(x) norm(x,type="1"))

I get the error:

Error in base::norm(x, type, ...) : 'A' must be a numeric matrix

I've found that this is because when we index X, we lose the matrix type. For example, running

norm(X[,1],type="1")

we get the same error. Hence, I've run

  old <- `[`
`[` <- function(...) { old(..., drop=FALSE) }

To stop the matrix type being dropped when we subset X. Now, this works

norm(X[,1],type="1")

But I still get the same error for

apply(X,2,function(x) norm(x,type="1"))

CodePudding user response：

You could do apply(X, 2, function(x) norm(matrix(x, ncol = 1), type = "1"), but you could also just do colSums(abs(X))--there's not much point in using norm here.

CodePudding user response：

You should be aware that apply(X, 2, ...) is producing "vector"s like below, which are not the matrix format as norm requested

> set.seed(1)

> X <- matrix(data = rnorm(100), nrow = 10, ncol = 10)

> apply(X, 2, str)
 num [1:10] -0.626 0.184 -0.836 1.595 0.33 ...
 num [1:10] 1.512 0.39 -0.621 -2.215 1.125 ...
 num [1:10] 0.919 0.7821 0.0746 -1.9894 0.6198 ...
 num [1:10] 1.3587 -0.1028 0.3877 -0.0538 -1.3771 ...
 num [1:10] -0.165 -0.253 0.697 0.557 -0.689 ...
 num [1:10] 0.398 -0.612 0.341 -1.129 1.433 ...
 num [1:10] 2.4016 -0.0392 0.6897 0.028 -0.7433 ...
 num [1:10] 0.476 -0.71 0.611 -0.934 -1.254 ...
 num [1:10] -0.569 -0.135 1.178 -1.524 0.594 ...
 num [1:10] -0.543 1.208 1.16 0.7 1.587 ...
NULL

A quick fix might be using as.matrix or t(), e.g.,

> apply(X, 2, function(x) str(as.matrix(x)))
 num [1:10, 1] -0.626 0.184 -0.836 1.595 0.33 ...
 num [1:10, 1] 1.512 0.39 -0.621 -2.215 1.125 ...
 num [1:10, 1] 0.919 0.7821 0.0746 -1.9894 0.6198 ...
 num [1:10, 1] 1.3587 -0.1028 0.3877 -0.0538 -1.3771 ...
 num [1:10, 1] -0.165 -0.253 0.697 0.557 -0.689 ...
 num [1:10, 1] 0.398 -0.612 0.341 -1.129 1.433 ...
 num [1:10, 1] 2.4016 -0.0392 0.6897 0.028 -0.7433 ...
 num [1:10, 1] 0.476 -0.71 0.611 -0.934 -1.254 ...
 num [1:10, 1] -0.569 -0.135 1.178 -1.524 0.594 ...
 num [1:10, 1] -0.543 1.208 1.16 0.7 1.587 ...
NULL

> apply(X, 2, function(x) str(t(x)))
 num [1, 1:10] -0.626 0.184 -0.836 1.595 0.33 ...
 num [1, 1:10] 1.512 0.39 -0.621 -2.215 1.125 ...
 num [1, 1:10] 0.919 0.7821 0.0746 -1.9894 0.6198 ...
 num [1, 1:10] 1.3587 -0.1028 0.3877 -0.0538 -1.3771 ...
 num [1, 1:10] -0.165 -0.253 0.697 0.557 -0.689 ...
 num [1, 1:10] 0.398 -0.612 0.341 -1.129 1.433 ...
 num [1, 1:10] 2.4016 -0.0392 0.6897 0.028 -0.7433 ...
 num [1, 1:10] 0.476 -0.71 0.611 -0.934 -1.254 ...
 num [1, 1:10] -0.569 -0.135 1.178 -1.524 0.594 ...
 num [1, 1:10] -0.543 1.208 1.16 0.7 1.587 ...
NULL

such that

> apply(X, 2, function(x) norm(as.matrix(x), type = "1"))
 [1] 6.497906 8.282578 6.963624 6.011803 5.194332 8.010169 9.687043 5.383619
 [9] 6.336799 9.304195

or (be careful, we use type = "I" here)

> apply(X, 2, function(x) norm(t(x), type = "I"))
 [1] 6.497906 8.282578 6.963624 6.011803 5.194332 8.010169 9.687043 5.383619
 [9] 6.336799 9.304195

Cross-check with colSums(abs(X)) (as mentioned by Gregor Thomas)

> colSums(abs(X))
 [1] 6.497906 8.282578 6.963624 6.011803 5.194332 8.010169 9.687043 5.383619
 [9] 6.336799 9.304195