Hi all,
I'm looking for a d-ish way to solve a basic "split-apply-combine" workflow. The idea is described (and solved) here:

https://stackoverflow.com/questions/39922986/pandas-group-by-and-sum

So, given a structure with some fields, say

´´´
struct S
{
	string s;
	int i;
}
´´´

I create an array of them, like

´´´
void main()
{
        import std.experimental.all;

	S[] sarr;
	sarr.length = 6;
	sarr[0].s = "a";
	sarr[1].s = "a";
	sarr[2].s = "b";
	sarr[3].s = "b";
	sarr[4].s = "c";
	sarr[5].s = "c";
	sarr[0].i = 1;
	sarr[1].i = 2;
	sarr[2].i = 4;
	sarr[3].i = 8;
	sarr[4].i = 16;
	sarr[5].i = 32;
	auto res = sarr.group!((a, b) => a.s == b.s);
        //writeln(res);
}
´´´

I'm also able to group them by a field, see last line.

But now the problems begin:
- The group operation tries to use the structure itself as a key, despite I provide a custom binary predicate.
- I could ignore the fact above, but how, given the result of the grouping operation I can merge by some function (like sum) the group results?

At this moment, I assume, that I'm approaching the problem from the wrong end, and simply don't see something trivial... Anyway, does anybody has a hint for me?

On 23/07/2018 11:49 PM, Alex wrote:
> 
> At this moment, I assume, that I'm approaching the problem from the wrong end, and simply don't see something trivial... Anyway, does anybody has a hint for me?

Map the range to the integer value, then sum it.

On Monday, 23 July 2018 at 11:49:58 UTC, Alex wrote:
> Hi all,
> I'm looking for a d-ish way to solve a basic "split-apply-combine" workflow. The idea is described (and solved) here:
>
> https://stackoverflow.com/questions/39922986/pandas-group-by-and-sum
>
> So, given a structure with some fields, say
>
> ´´´
> struct S
> {
> 	string s;
> 	int i;
> }
> ´´´
>
> I create an array of them, like
>
> ´´´
> void main()
> {
>         import std.experimental.all;
>
> 	S[] sarr;
> 	sarr.length = 6;
> 	sarr[0].s = "a";
> 	sarr[1].s = "a";
> 	sarr[2].s = "b";
> 	sarr[3].s = "b";
> 	sarr[4].s = "c";
> 	sarr[5].s = "c";
> 	sarr[0].i = 1;
> 	sarr[1].i = 2;
> 	sarr[2].i = 4;
> 	sarr[3].i = 8;
> 	sarr[4].i = 16;
> 	sarr[5].i = 32;
> 	auto res = sarr.group!((a, b) => a.s == b.s);
>         //writeln(res);
> }
> ´´´
>
> I'm also able to group them by a field, see last line.
>
> But now the problems begin:
> - The group operation tries to use the structure itself as a key, despite I provide a custom binary predicate.
> - I could ignore the fact above, but how, given the result of the grouping operation I can merge by some function (like sum) the group results?
>
> At this moment, I assume, that I'm approaching the problem from the wrong end, and simply don't see something trivial... Anyway, does anybody has a hint for me?

You could use chunkBy:

auto res = sarr.chunkBy!((a, b) => a.s == b.s).map!(a => tuple(a.front.s, a.map!(b => b.i).sum));
https://run.dlang.io/is/TJOEmf

chunkBy:

---
[S("a", 1), S("a", 2)]
[S("b", 4), S("b", 8)]
[S("c", 16), S("c", 32)]
---

group:

---
Tuple!(S, uint)(const(S)("a", 1), 2)
Tuple!(S, uint)(const(S)("b", 4), 2)
Tuple!(S, uint)(const(S)("c", 16), 2)
---

On Monday, 23 July 2018 at 12:07:37 UTC, Seb wrote:
>
> You could use chunkBy:
>
> auto res = sarr.chunkBy!((a, b) => a.s == b.s).map!(a => tuple(a.front.s, a.map!(b => b.i).sum));
> https://run.dlang.io/is/TJOEmf
>

Ha... This helps! Thanks a lot! :)