10

I have a data frame consisting entirely of integer64 columns that I'd like to convert to be a matrix.

library(bit64)
(dfr <- data.frame(x = as.integer64(10^(9:18))))
##                      x
## 1           1000000000
## 2          10000000000
## 3         100000000000
## 4        1000000000000
## 5       10000000000000
## 6      100000000000000
## 7     1000000000000000
## 8    10000000000000000
## 9   100000000000000000
## 10 1000000000000000000

Unfortunately, as.matrix doesn't give the correct answer.

(m <- as.matrix(dfr))
##                   x
##  [1,] 4.940656e-315
##  [2,] 4.940656e-314
##  [3,] 4.940656e-313
##  [4,] 4.940656e-312
##  [5,] 4.940656e-311
##  [6,] 4.940656e-310
##  [7,] 4.940656e-309
##  [8,] 5.431165e-308
##  [9,] 5.620396e-302
## [10,] 7.832953e-242

The problem seems to be that integer64 values are stored as numeric values with an "integer64" class attribute (plus some magic to make them print and do arithmetic correctly) that gets stripped by as.matrix.

I can't just do class(m) <- "integer64" because that changes the class of the matrix object not its contents.

Likewise, mode(m) <- "integer64" gives the wrong answer and typeof(m) <- "integer64" and storage.mode(m) <- "integer64" throw errors.

Of course I could just circumvent the problem by converting the columns to double (dfr$x <- as.double(dfr$x)) but it feels like there ought to be a way to do this properly.

How can I get a matrix of integer64 values?

Richie Cotton
  • 118,240
  • 47
  • 247
  • 360
  • 4
    Perhaps you're heading down a dark path: unless all the *functions* you're going to feed your matrix to have `integer64` methods, you'll still end up convering to doubles at some point. Does the `bit64` package have a collection of appropriate methods, and for that matter does it have some `as.matrix64` tool? If not you might as well convert to double or perhaps use `gmp` and `Rmpfr` packages. ETA: the word "matrix" doesn't even appear in the Help file for `bit64`, so I think you'll have to stick with dataframes and the available methods for that. – Carl Witthoft Feb 01 '15 at 12:24
  • 3
    @CarlWitthoft but ooh look at the shiny ball bouncing off into the dark, tangled woods from which there is no hope of return, maybe I should go chase it. – Richie Cotton Feb 01 '15 at 12:41

1 Answers1

5

For a raw vector, assigning the dim attribute directly seems to work:

> z <- as.integer64(1:10)
> z
integer64
 [1] 1  2  3  4  5  6  7  8  9  10
> dim(z) <- c(10, 1)
> z
integer64
      [,1]
 [1,] 1   
 [2,] 2   
 [3,] 3   
 [4,] 4   
 [5,] 5   
 [6,] 6   
 [7,] 7   
 [8,] 8   
 [9,] 9   
[10,] 10

For a data frame, cbinding the columns also works:

> df <- data.frame(x=as.integer64(1:5), y=as.integer64(6:10))
> df
  x  y
1 1  6
2 2  7
3 3  8
4 4  9
5 5 10
> cbind(df$x, df$y)
integer64
     [,1] [,2]
[1,] 1    6   
[2,] 2    7   
[3,] 3    8   
[4,] 4    9   
[5,] 5    10

So, for an arbitrary number of columns, do.call is the way to go:

> do.call(cbind, df)
integer64
     x y 
[1,] 1 6 
[2,] 2 7 
[3,] 3 8 
[4,] 4 9 
[5,] 5 10
Hong Ooi
  • 56,353
  • 13
  • 134
  • 187