mibble() / frame_matrix()

[lionel-]

I think it would make sense to have a matrix equivalent of nibble() / frame_data(). It would be useful for defining small covariance and correlation matrices in a readable way. Such small matrices are probably a more common use case than small data frames.

It should support column names:

frame_matrix(
  ~col1, ~col2,
     10,     3,
      3,     2
)

As well as row names:

frame_matrix(
  ~row1, 10, 3,
  ~row2,  3, 2
)

Or both:

frame_matrix(
  NULL,  ~col1, ~col2,
  ~row1,    10,     3,
  ~row2,     3,     2
)

Most of the time matrices don't have column names so it'd be useful to provide an alternative way of figuring out the size of the matrix, with an argument:

Sigma <- frame_matrix(ncol = 2,
  10, 3
  3,  2
)

These are all a lot more readable in a script than:

matrix(c(10, 3, 3, 2), 2)

[lionel-]

And then we could have other verbs for symmetric matrices? They would work in the same way but would assume the input is triangular:

frame_lower(ncol = 2,
  10,
   3, 2
)

frame_upper(ncol = 2,
  10, 3,
      2
)

frame_lower(
  NULL,  ~var1, ~var2,
  ~var1,    10,
  ~var2,     3,     2
)

[hadley]

I think we'd accept a PR that implements this, but it's not high enough priority for us to do.

[anhqle]

I could contribute a PR, but unsure how much more convenient this is.

Comparing:

frame_matrix(
  ~col1, ~col2,
     10,     3,
      3,     2
)

and

matrix(c(10, 3,
         3, 2), 2)

I would love to see the appeal of the frame_matrix more clearly.

[lionel-]

You are right that this is already pretty close, though arguably not aesthetic:

c(
  1, 2,
  3, 4
) %>% matrix(2)

I'd say the appeal of frame_matrix() would be:

• To provide a consistent way of specifying small tabular data structures in R.
• With the triangular version there is less chance of data entry mistakes.
• The rownames and colnames can be specified inline alongside the data.

And maybe also the dimnames? E.g. you could substitute the NULL in

frame_lower(
  NULL,  ~var1, ~var2,
  ~var1,    10,
  ~var2,     3,     2
)

with a rows ~ cols specification:

frame_lower(dim1 ~ dim2,
         ~var1, ~var2,
  ~var1,    10,
  ~var2,     3,     2
)

[anhqle]

@lionel- Would you be worry that something like

frame_lower(dim1 ~ dim2,
         ~var1, ~var2,
  ~var1,    10,
  ~var2,     3,     2
)

would require extensive white-space editing to make things line up? A single change in the cell value or row/column name means a whole lot of editing to make things align again. I'm a little concerned that it would actually cause more data entry problems.

[lionel-]

I think that's a general problem with these functions, they are more readable once done but require some manual editing to get spacing right.

[krlmlr]

@LaDilettante: The reformatting problem is alleviated with #127. Happy to review a PR.