Column referencing: [[i]] vs [,i] for matrix, dataframe, and data.table

问题

Could someone please explain to me the difference in column referencing between matrix, data.frame, and data.table? I'm getting my head around which syntax to use for each class, but I don't understand how/why they're different.

Take a 10x10 matrix

foo <- matrix( nrow = 10, ncol = 10 )

I'll just fill the 2nd column to demonstrate:

foo[,2] <- rnorm(10)
head( foo, 3 )

        [,1]       [,2] [,3] [,4] [,5] [,6] [,7] [,8] [,9] [,10]
  [1,]   NA -0.4688874   NA   NA   NA   NA   NA   NA   NA    NA
  [2,]   NA -1.0273370   NA   NA   NA   NA   NA   NA   NA    NA
  [3,]   NA -0.3981627   NA   NA   NA   NA   NA   NA   NA    NA

Now I can reference the 2nd column with foo[,2], but foo[[2]] returns only 1 cell, which in this case is NA:

foo[,2]
 [1]  0.18340527  0.46511236 -2.43277107  0.13260218  0.20227436 -0.57518392 -0.62211864  2.00239088  -0.09561907  0.67536428

foo[[2]]
  [1] NA

If I change the matrix to a dataframe, both referencing methods work:

foo <- data.frame( foo )
foo[,2]
  [1] -0.4688874 -1.0273370 -0.3981627 -0.2207062  0.5711004  1.1085851 -1.3343338  0.2337622  -1.0632469 -0.9783714


foo[[2]]
  [1] -0.4688874 -1.0273370 -0.3981627 -0.2207062  0.5711004  1.1085851 -1.3343338  0.2337622  -1.0632469 -0.9783714

Now if I convert to a data.table only the second method works, and the first method returns the value 2 (which isn't in the table at all):

foo[,2]
  [1] 2
foo[[2]]
  [1] -0.4688874 -1.0273370 -0.3981627 -0.2207062  0.5711004  1.1085851 -1.3343338  0.2337622  -1.0632469 -0.9783714

So my question is, why the different syntax for different classes? And is there a particular syntax that would work for all 3 classes, or do we need to know/check the tabular class before knowing how to call a reference?

EDIT: also interesting here is that row referencing is more consistent across classes.

For matrix, dataframe, and data.table respectively:

foo[2,]
 [1]        NA 0.4651124        NA        NA        NA        NA        NA        NA        NA        NA

foo <- data.frame( foo )
foo[2,]
  X1        X2 X3 X4 X5 X6 X7 X8 X9 X10
2 NA 0.4651124 NA NA NA NA NA NA NA  NA

setDT( foo )
foo[2,]
  X1        X2 X3 X4 X5 X6 X7 X8 X9 X10
1: NA 0.4651124 NA NA NA NA NA NA NA  NA

回答1:

A few things I've learned since posting this question here (with a lot of help from the commenters!), which have helped me to understand the differences I mentioned. If anyone could please clarify anything I'm getting wrong here, I'd appreciate it:

Objects of class matrix, data.frame, and data.table are all list objects under the hood, but they differ in an important way.
Each column of a data.frame or data.table object is an element of the list "under its hood", meaning that a column can be extracted in the same way as a list element would be, hence foo[[2]] works great for calling the second column in both of those classes.
A matrix differs in that every cell is an element of the list, meaning that foo[[2]] will only retrieve one cell, rather than a column (which brings us to...).
Those list items making up the matrix are "filled" column-wise (top-to-bottom, left-to-right), so the call foo[[2]] is retrieving the second item, which here resides in row 2 of column 1.
Since the matrix does also have dimensions, foo[,2] is accepted as referring to the second column, as it does for a data.frame object.
A data.table object (up until recently, see next point) didn't have a particularly logical response to the call foo[,2], and returned the value 2 regardless of the data to which it was referring, for no good reason I can find.
As of very recent updates to the data.table package (as of 1.9.8 I think? Thank you maintainers!) the syntax foo[,2] is now logically accepted as per a data.frame, so some of the confusion which lead to my question has been superseded!

So in conclusion:

All of the objects I mentioned in my question are actually lists (which means I now get @N8TRO's joke in the comments, to which I was naively oblivious before), with both data.table and data.frame containing a list element for each column, and a matrix containing a list element for each cell (this makes the [[ call make sense to me now).
All of the objects mentioned have dimensions, which means (as of recent data.table package updates) the foo[,2] syntax works the same for all 3 classes. YAY!

Thank you so much to the commenters for pointing me in the right direction (and making jokes that I now get). I hope this might help someone in the future who comes across the same confusion I did.

来源：https://stackoverflow.com/questions/38263734/column-referencing-i-vs-i-for-matrix-dataframe-and-data-table

标签

matrix

dataframe

data.table