arg_max for 2d arrays?

[artsiomkaltovich]

First of all, I apologize if this should be an issue for the libminizinc project or if it shouldn’t be an issue at all.

What does $$E mean in the following code?

function $$E: arg_max(array [$$E] of bool: x)
function $$E: arg_max(array [$$E] of int: x)
function $$E: arg_max(array [$$E] of float: x)

Is it possible to use arg_max for 2D arrays or arrays with higher dimensions? Could you provide a working example of how to use it?

Thank you in advance.

[Dekker1]

In the function declarations you state, $$E would be a type-inst variable that (because of the double dollar signs) can be instated with enumerated types or just integers. In this case this ensures that if a enumerated type is used to form the index set of the argument for arg_max, then the returned value is also of that enumerated type (and not just an int).

Creating a version of arg_max that works for higher dimensions is certainly possible. A decomposition can be created using the 1d version of arg_max and an element constraint:

function var tuple($$E, $$F): arg_max(array[$$E, $$F] of var int: x) = let {
	any: idxs = [(i, j) | i in index_set_1of2(x), j in index_set_2of2(x)];
	any: sel = arg_max(array1d(x));
} in idxs[sel];

Note that it seems that the any type doesn't work for idxs in the current release. You instead can use the array[int] of tuple(int, int) for idxs, although it is technically not fully correct.