R use matrix to select row of multidimensional array

Question

I have a multidimensional array, created in the following way:

my_array <- array(seq(12), dim=my_dims, dimnames=my_dimnames)

where "my_dims" is a numeric vector and "my_dimnames" is a list of character vectors, with lengths corresponding to the dimensions specified in "mydims", for example:

my_dims <- c(2, 3, 2)
my_dimnames <- list(c("D_11", "D_12"), c("D_21", "D_22", "D_23"), c("D_31", "D_32"))

So that "my_array" looks something like this:

, , D_31

     D_21 D_22 D_23
D_11    1    3    5
D_12    2    4    6

, , D_32

     D_21 D_22 D_23
D_11    7    9   11
D_12    8   10   12

Additionally, I have a character vector containing specific values of each dimension:

my_values <- c("D_11", "D_21", "D_32")

Now, from this question: R - how to get a value of a multi-dimensional array by a vector of indices I learned that I can read and write values from and to specific cells in "my_array" using matrix indexing, like this:

my_array[matrix(my_values, 1)]

to get the value "7" and

my_array[matrix(my_values, 1)] <- 1

to set that same cell to value "1"

What I don't understand, though, is how I can get all values from a specific dimension from "my_array", given specific values for all other dimensions using this method.

For example: how can I get a vector containing all values of the first dimension with fixed values "D_21" and "D_32" for the second and third dimension? So in this case what I try to extract would be a vector of the form:

c(7, 8)

How do I need to adjust the matrix indexing to achieve this result?

Thanks a lot in advance!

Make use of the dimnames of your array? Something like: my_array[cbind(dimnames(my_array)[[1]], "D_21", "D_32")] perhaps? — A5C1D2H2I1M1N2O1R2T1
– A5C1D2H2I1M1N2O1R2T1, Commented Jan 14, 2016 at 17:07
How do "my_values" look in the case that you want all of a dimension? — alexis_laz
– alexis_laz, Commented Jan 14, 2016 at 17:59
@alexis_laz: that's exactly what I want to know. I hoped there was something similar to selecting all elements of a data.frame by leaving a dimension empty (like all rows for cells c1 and c2 in data frame df using "df[, c(c1, c2)]"). Is there a similar way to formulate a list or vector like "my_values" to generate a matrix for indexing? — scholt
– scholt, Commented Jan 15, 2016 at 15:49
@scholt : From the QA you've linked you can use do.call("[", ... To simulate the missing argument in a dimensions use substitute(). E.g. do.call("[", c(list(my_array), list("D_11", "D_23", "D_31"))); do.call("[", c(list(my_array), list(substitute(), "D_23", "D_31"))); do.call("[", c(list(my_array), list("D_12", substitute(), "D_31", drop = FALSE))) etc. Your arguments for do.call need to be in a "list"; e.g. for "my_values" you'd need as.list(my_values). — alexis_laz
– alexis_laz, Commented Jan 15, 2016 at 15:57

A5C1D2H2I1M1N2O1R2T1 · Accepted Answer · 2016-01-15 05:45:44Z

1

I don't know if something like this already exists, but as mentioned in the comments, you can make use of the dimnames of your array.

It would be best if your "my_values" object were a list, as that would give you a lot of flexibility that you won't have with a vector.

Here's a possible approach. Create a function that creates a matrix that can be used to subset from your array. The function would essentially look something like this:

mat_maker <- function(array, vals) {
  x <- sapply(vals, is.null)
  vals[x] <- dimnames(array)[x]
  as.matrix(expand.grid(vals))
}

The extraction function would be like this. (Most likely, you would just roll both of these functions together, but I thought I would share them here separately so that you can see what is going on.)

array_extractor <- function(array, vals) {
  array[mat_maker(array, vals)]
}

Now, try the following examples out. Notice that the "vals" argument is a list, and that NULL is used for dimensions that are not being specified.

## All values where rowname == "D_11"
mat_maker(my_array, list("D_11", NULL, NULL))
#      Var1   Var2   Var3  
# [1,] "D_11" "D_21" "D_31"
# [2,] "D_11" "D_22" "D_31"
# [3,] "D_11" "D_23" "D_31"
# [4,] "D_11" "D_21" "D_32"
# [5,] "D_11" "D_22" "D_32"
# [6,] "D_11" "D_23" "D_32"
array_extractor(my_array, list("D_11", NULL, NULL))
# [1]  1  3  5  7  9 11


## All first column values from the second array object
## This is your specific example
mat_maker(my_array, list(NULL, "D_21", "D_32"))
#      Var1   Var2   Var3  
# [1,] "D_11" "D_21" "D_32"
# [2,] "D_12" "D_21" "D_32"
array_extractor(my_array, list(NULL, "D_21", "D_32"))
# [1] 7 8

## First two columns from all array dimensions
array_extractor(my_array, list(NULL, c("D_21", "D_22"), NULL))
# [1]  1  2  3  4  7  8  9 10

An extension of this function would be to put the dimensional attributes (including the names) back into the output. It could look something like this:

extractMe <- function(array, vals) {
  x <- sapply(vals, is.null)
  vals[x] <- dimnames(array)[x]
  temp <- as.matrix(expand.grid(vals))
  `dimnames<-`(`dim<-`(array[temp], lengths(vals)), vals)
}

Here's the usage:

extractMe(my_array, list("D_11", NULL, NULL))
# , , D_31
# 
#      D_21 D_22 D_23
# D_11    1    3    5
# 
# , , D_32
# 
#      D_21 D_22 D_23
# D_11    7    9   11

extractMe(my_array, list(NULL, "D_21", "D_32"))
# , , D_32
# 
#      D_21
# D_11    7
# D_12    8

extractMe(my_array, list(NULL, c("D_21", "D_22"), NULL))
# , , D_31
# 
#      D_21 D_22
# D_11    1    3
# D_12    2    4
# 
# , , D_32
# 
#      D_21 D_22
# D_11    7    9
# D_12    8   10

And, for cases where you want just a vector of the values, just wrap it with c:

c(extractMe(my_array, list(NULL, "D_21", "D_32")))
# [1] 7 8

edited Jan 15, 2016 at 5:45

answered Jan 14, 2016 at 17:38

A5C1D2H2I1M1N2O1R2T1

194k30 gold badges416 silver badges496 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

scholt Over a year ago

Thank you very much for this solution. It seems to work for me, but actually I was wondering if it was possible to achieve the same goal with onboard functions that come with R. More specifically, I was wondering if there is a way to specify something like a wildcard - I think that is the "NULL" in your solution - in matrix indexing directly. I was always intrigued that you can select whole columns or rows of a data frame in R by just writing something like "df[, c("column_1", "column_2")]" and was wondering if there is something similar for matrix indexing.

A5C1D2H2I1M1N2O1R2T1 Over a year ago

@scholt, Yes, there is, but the question remains, how do you go from a flat vector ("my_values") to what you want. Since you're dealing with a 3-dimensional array here, you could do my_array[row, col, third-dimension]. Compare the results of my_array[1, , ]; my_array[, 1, ]; my_array[, , 1].

Collectives™ on Stack Overflow

R use matrix to select row of multidimensional array

1 Answer 1

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related