www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - Mir Slice.shape is not consistent with the actual array shape

reply Pavel Shkadzko <p.shkadzko gmail.com> writes:
I am confused by the return value of Mir shape.
Consider the following example.

///////////////////////////////////////////////////////////////////////////
import std.stdio;
import std.conv;
import std.array: array;
import std.range: chunks;
import mir.ndslice;

int[] getShape(T : int)(T obj, int[] dims = null)
{
     return dims;
}

// return arr shape
int[] getShape(T)(T obj, int[] dims = null)
{
     dims ~= obj.length.to!int;
     return getShape!(typeof(obj[0]))(obj[0], dims);
}

void main() {
     int[] arr = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 
15, 16];
     int[][][] a = arr.chunks(4).array.chunks(2).array;

     writeln(arr);
     writeln(arr.shape);

     auto arrSlice = arr.sliced;
     writeln(arrSlice);
     writeln(arrSlice.shape);

}
///////////////////////////////////////////////////////////////////////////

[[[1, 2, 3, 4], [5, 6, 7, 8]], [[9, 10, 11, 12], [13, 14, 15, 
16]]]
[2, 2, 4] <-- correct shape
[[[1, 2, 3, 4], [5, 6, 7, 8]], [[9, 10, 11, 12], [13, 14, 15, 
16]]]
[2] <-- which shape is that?

I would expect sliced to create a Slice with the same dims. Well, 
sliced returns a shell over the array, but why does it return its 
own shape instead of the shape of the array it provides view 
into? This makes it even more confusing once you print both 
representations.
What's the rationale here?
May 24 2020
next sibling parent reply Pavel Shkadzko <p.shkadzko gmail.com> writes:
On Sunday, 24 May 2020 at 14:17:33 UTC, Pavel Shkadzko wrote:

 I am confused by the return value of Mir shape.
 Consider the following example.

 [...]


Sorry for the typo. It should be "auto arrSlice = a.sliced;"
May 24 2020
next sibling parent reply jmh530 <john.michael.hall gmail.com> writes:
On Sunday, 24 May 2020 at 14:21:26 UTC, Pavel Shkadzko wrote:

 [snip]

 Sorry for the typo. It should be "auto arrSlice = a.sliced;"


Try using fuse

/+dub.sdl:
dependency "mir-algorithm" version="*"
+/
import std.stdio;
import std.conv;
import std.array: array;
import std.range: chunks;
import mir.ndslice;

int[] getShape(T : int)(T obj, int[] dims = null)
{
     return dims;
}

// return arr shape
int[] getShape(T)(T obj, int[] dims = null)
{
     dims ~= obj.length.to!int;
     return getShape!(typeof(obj[0]))(obj[0], dims);
}

void main() {
     int[] arr = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 
15, 16];
     int[][][] a = arr.chunks(4).array.chunks(2).array;

     int err;
     writeln(arr);
     writeln(a.shape(err));

     auto aSlice = a.fuse;
     writeln(aSlice);
     writeln(aSlice.shape);

}
May 24 2020
parent Pavel Shkadzko <p.shkadzko gmail.com> writes:
On Sunday, 24 May 2020 at 14:35:33 UTC, jmh530 wrote:

 On Sunday, 24 May 2020 at 14:21:26 UTC, Pavel Shkadzko wrote:

 [snip]

 Sorry for the typo. It should be "auto arrSlice = a.sliced;"


 Try using fuse

 /+dub.sdl:
 dependency "mir-algorithm" version="*"
 +/
 import std.stdio;
 import std.conv;
 import std.array: array;
 import std.range: chunks;
 import mir.ndslice;

 int[] getShape(T : int)(T obj, int[] dims = null)
 {
     return dims;
 }

 // return arr shape
 int[] getShape(T)(T obj, int[] dims = null)
 {
     dims ~= obj.length.to!int;
     return getShape!(typeof(obj[0]))(obj[0], dims);
 }

 void main() {
     int[] arr = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 
 15, 16];
     int[][][] a = arr.chunks(4).array.chunks(2).array;

     int err;
     writeln(arr);
     writeln(a.shape(err));

     auto aSlice = a.fuse;
     writeln(aSlice);
     writeln(aSlice.shape);

 }


Yes, this works. It doesn't explain why sliced behaves like this 
though. Also, isn't sliced the default way of converting D arrays 
to Mir arrays? Above all arr.fuse.field flattens the array. So, 
fuse works as D join if combined with field but as slice 
allocator is used on a D array...
May 24 2020
prev sibling parent Pavel Shkadzko <p.shkadzko gmail.com> writes:
On Sunday, 24 May 2020 at 14:21:26 UTC, Pavel Shkadzko wrote:

 On Sunday, 24 May 2020 at 14:17:33 UTC, Pavel Shkadzko wrote:

 I am confused by the return value of Mir shape.
 Consider the following example.

 [...]


 Sorry for the typo. It should be "auto arrSlice = a.sliced;"


And another typo "writeln(arr.getShape);", too rushy.
May 24 2020
prev sibling next sibling parent 9il <ilyayaroshenko gmail.com> writes:
On Sunday, 24 May 2020 at 14:17:33 UTC, Pavel Shkadzko wrote:

 I am confused by the return value of Mir shape.
 Consider the following example.

 [...]


`sliced` returns a view on the array data, a 1-dimensional slice 
composed of common D arrays. Try to use `fuse` instead of 
`sliced`.
May 24 2020
prev sibling parent reply 9il <ilyayaroshenko gmail.com> writes:
On Sunday, 24 May 2020 at 14:17:33 UTC, Pavel Shkadzko wrote:

 I am confused by the return value of Mir shape.
 Consider the following example.

 ///////////////////////////////////////////////////////////////////////////
 import std.stdio;
 import std.conv;
 import std.array: array;
 import std.range: chunks;
 import mir.ndslice;

 int[] getShape(T : int)(T obj, int[] dims = null)
 {
     return dims;
 }

 // return arr shape
 int[] getShape(T)(T obj, int[] dims = null)
 {
     dims ~= obj.length.to!int;
     return getShape!(typeof(obj[0]))(obj[0], dims);
 }

 void main() {
     int[] arr = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 
 15, 16];
     int[][][] a = arr.chunks(4).array.chunks(2).array;

     writeln(arr);
     writeln(arr.shape);

     auto arrSlice = arr.sliced;
     writeln(arrSlice);
     writeln(arrSlice.shape);

 }
 ///////////////////////////////////////////////////////////////////////////

 [[[1, 2, 3, 4], [5, 6, 7, 8]], [[9, 10, 11, 12], [13, 14, 15, 
 16]]]
 [2, 2, 4] <-- correct shape
 [[[1, 2, 3, 4], [5, 6, 7, 8]], [[9, 10, 11, 12], [13, 14, 15, 
 16]]]
 [2] <-- which shape is that?

 I would expect sliced to create a Slice with the same dims. 
 Well, sliced returns a shell over the array, but why does it 
 return its own shape instead of the shape of the array it 
 provides view into? This makes it even more confusing once you 
 print both representations.
 What's the rationale here?


BTW, the code example above doesn't compiles.

OT:
Instead of


     int[] arr = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 
 15, 16];
     int[][][] a = arr.chunks(4).array.chunks(2).array;


you can generate the same common D array using Mir:

     auto a = [2, 2, 4].iota!int(1).ndarray;
May 24 2020
parent Pavel Shkadzko <p.shkadzko gmail.com> writes:
On Sunday, 24 May 2020 at 15:24:14 UTC, 9il wrote:

 On Sunday, 24 May 2020 at 14:17:33 UTC, Pavel Shkadzko wrote:

 [...]


 BTW, the code example above doesn't compiles.

 OT:
 Instead of


     [...]


 you can generate the same common D array using Mir:

     auto a = [2, 2, 4].iota!int(1).ndarray;


I posted in a rush. There should be "auto arrSlice = a.sliced;" 
and "writeln(a.getShape);".
May 24 2020