It strikes me that in some cases it might be convenient to be able to extract a column from a gretl matrix that has column-names attached by name rather than column-number. The primary case I have in mind is a matrix returned by a function (or in a bundle returned by a function), where the columns basically represent series. So, given matrix M = some_func(some_args) series some_name = M[,4] one could do, say, matrix M = some_func(some_args) series some_name = M[,"forecast"] Any thoughts on this? If the idea seems attractive I can see about implementing it.

On 14/04/2026 21:14, Sven Schreiber wrote: Am 13.04.2026 um 22:50 schrieb Cottrell, Allin: It strikes me that in some cases it might be convenient to be able to extract a column from a gretl matrix that has column-names attached by name rather than column-number. The primary case I have in mind is a matrix returned by a function (or in a bundle returned by a function), where the columns basically represent series. So, given matrix M = some_func(some_args) series some_name = M[,4] one could do, say, matrix M = some_func(some_args) series some_name = M[,"forecast"] Any thoughts on this? If the idea seems attractive I can see about implementing it. Yes it does seem attractive. A follow-up question would be: what's the error if the name doesn't exist? Index out of bounds? Another thought: Instead of having named columns collected in a matrix, the same columns could already be stored inside a bundle with the same names, and then would be accessible by name without the square brackets and the quotes, like B.forecast. On the other hand, if the columns are in the same matrix, that would ensure that their lengths are equal, which might be an advantage. I find the idea _very_ attractive too. Beside Sven's remakes, there are other two points that are IMO worth considering: why not rows too? how about string arrays for multiple selection?

There is a more general point that IMO is worth considering. Suppose we're slicing a matrix with row/column names (via traditional methods or via this cool new feature). Should names be carried over to the resulting objects? To be more explicit (quoting from above): <hansl> set seed 765545431 matrix M = mnormal(4,3) cnameset(M, "one two three") matrix c = M[,"two"] </hansl> should the "c" vector be endowed with the column name "two" too? My spontaneous reaction would be "no" (on the grounds of efficiency), but others may disagree.