tiledb 0.9.1 drops rows and cols when attrs= is specified

[LTLA]

I'll use the example generated by ex_1.R. (Note that I needed to add names to the list at:

TileDB-R/inst/examples/ex_1.R

Lines 31 to 33 in c76c783

	data <- list(array(seq(1:50), dim = c(10,5)),
	array(as.double(seq(101,by=0.5,length=50)), dim = c(10,5)),
	array(c(letters[1:26], "brown", "fox", LETTERS[1:22]), dim = c(10,5)))

Otherwise it complains Error: No column names supplied.)

Anyway, once ex_1/ is generated, we can run the following code in 0.9.1 (8cca43b):

> library(tiledb)
> obj <- tiledb_array("ex_1/", as.data.frame=TRUE)
> head(obj[])
  rows cols  a     b c
1    1    1  1 101.0 a
2    1    2 11 106.0 k
3    1    3 21 111.0 u
4    1    4 31 116.0 C
5    1    5 41 121.0 M
6    2    1  2 101.5 b
> obj <- tiledb_array("ex_1/", attrs="a", as.data.frame=TRUE)
> head(obj[])
[1]  1 11 21 31 41  2
> obj <- tiledb_array("ex_1/", attrs=c("a", "b"), as.data.frame=TRUE)
> head(obj[])
   a     b
1  1 101.0
2 11 106.0
3 21 111.0
4 31 116.0
5 41 121.0
6  2 101.5

You can see how specifying attrs= causes the rows and cols columns to be removed, and eventually for the data frame structure to be dropped altogether if attrs is of length 1. This is pretty surprising for downstream code that expects to see the rows and cols columns for further manipulation of the indices.

For comparison, this is what I see on 0.9.0 (20e775c):

> obj <- tiledb_array("ex_1/", as.data.frame=TRUE)
> head(obj[])
  rows cols  a     b c
1    1    1  1 101.0 a
2    1    2 11 106.0 k
3    1    3 21 111.0 u
4    1    4 31 116.0 C
5    1    5 41 121.0 M
6    2    1  2 101.5 b
> obj <- tiledb_array("ex_1/", attrs="a", as.data.frame=TRUE)
> head(obj[])
  rows cols  a
1    1    1  1
2    1    2 11
3    1    3 21
4    1    4 31
5    1    5 41
6    2    1  2
> obj <- tiledb_array("ex_1/", attrs=c("a", "b"), as.data.frame=TRUE)
> head(obj[])
  rows cols  a     b
1    1    1  1 101.0
2    1    2 11 106.0
3    1    3 21 111.0
4    1    4 31 116.0
5    1    5 41 121.0
6    2    1  2 101.5

[eddelbuettel]

Thanks for the report. I can reproduce both the shortcoming in the example (missing, as you noted, the names for the write assignment) and the dimension drop. I will take a closer look. There is single decision in

TileDB-R/R/TileDBArray.R

Line 607 in c76c783

if (!x@extended || length(sel) > 0) {

and I think I may have looked at all the repurcussions mostly for sparse matrices. So am I understanding you correctly that you would prefer to have the selected attributes (which may be one) but always retain rows and columns (as in the 0.9.0 example) ?

[LTLA]

So am I understanding you correctly that you would prefer to have the selected attributes (which may be one) but always retain rows and columns (as in the 0.9.0 example) ?

Yes, that 0.9.0 behavior is what I would have expected.

[eddelbuettel]

BTW we also added as.matrix return so for there same data set and attr argument:

$ Rscript -e 'library(tiledb); uri <- "ex_1"; obj <- tiledb_array(uri, attrs=c("a"), as.matrix=TRUE); obj[])'
      [,1] [,2] [,3] [,4] [,5]
 [1,]    1   11   21   31   41
 [2,]    2   12   22   32   42
 [3,]    3   13   23   33   43
 [4,]    4   14   24   34   44
 [5,]    5   15   25   35   45
 [6,]    6   16   26   36   46
 [7,]    7   17   27   37   47
 [8,]    8   18   28   38   48
 [9,]    9   19   29   39   49
[10,]   10   20   30   40   50
$ 

Edit: And without setting attrs you get a list of three matrices on you would expect.