www.digitalmars.com         C & C++   DMDScript  

digitalmars.D - Lazy concatenation and padding utilities

reply Ilya Yaroshenko <ilyayaroshenko gmail.com> writes:
Hello

Ndslice got [1] lazy multidimensional concatenation and padding 
utilities:
  1. stack
  2. pad
  2. padEdge
  3. padSymmetric
  4. padWrap


`stack` can be used for ndslices instead of `chain`.

---
import mir.ndslice.allocation: slice;
import mir.ndslice.stack: stack;

auto d = stack(a, b, c).slice;
---

ndslice allocation and assign utilities are optimised to work 
with stacks. For example the code above is significantly faster 
then

---
import std.array: array;
import std.range: chain;

auto d = chain(a, b, c).array;
---

because `slice` uses external iteration for `Stack` data 
structure.

[1] http://docs.algorithm.dlang.io/latest/mir_ndslice_stack.html

Best regards,
Ilya
Mar 09 2017
next sibling parent reply =?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:
On Thursday, 9 March 2017 at 08:59:28 UTC, Ilya Yaroshenko wrote:

 Ndslice got [1] lazy multidimensional concatenation and padding 
 utilities:


Nice. Is this dependent on choosing either RC- or GC-based 
allocation?
Mar 09 2017
next sibling parent Ilya Yaroshenko <ilyayaroshenko gmail.com> writes:
On Thursday, 9 March 2017 at 09:10:47 UTC, Nordlöw wrote:

 On Thursday, 9 March 2017 at 08:59:28 UTC, Ilya Yaroshenko 
 wrote:

 Ndslice got [1] lazy multidimensional concatenation and 
 padding utilities:


 Nice. Is this dependent on choosing either RC- or GC-based 
 allocation?


No, they are completely generic.

The new ndslice uses iterators under the hood. An iterator for 
manually allocated memory (with `makeSlice!T`) and GC allocated 
memory (with `slice!T`) is just a pointer type of T*.

You can use `slicedField` [1] to create ndslice on top of a 
RC-array. slicedField will create iterator on top of the rc-array 
using FieldIterator [2], and return ndslice based on the 
iterator. This is quite simple

RC arrays can be very simple: only `auto ref opIndex(size_t 
index)`  and `length` are required. `front`, `opSlice` and other 
RAR primitives are not used by FieldIterator.

[1] 
http://docs.algorithm.dlang.io/latest/mir_ndslice_slice.html#slicedField
[2] 
http://docs.algorithm.dlang.io/latest/mir_ndslice_iterator.html#FieldIterator
Mar 09 2017
prev sibling parent Ilya Yaroshenko <ilyayaroshenko gmail.com> writes:
On Thursday, 9 March 2017 at 09:10:47 UTC, Nordlöw wrote:

 On Thursday, 9 March 2017 at 08:59:28 UTC, Ilya Yaroshenko 
 wrote:

 Ndslice got [1] lazy multidimensional concatenation and 
 padding utilities:


 Nice. Is this dependent on choosing either RC- or GC-based 
 allocation?


... continue prev. reply:

stack and pad* functions are 100% lazy and do not allocate 
anything.
If you have allocated slice of the same shape you can do the 
following:

preallocated_slice[] = stack(a, b, c);

where preallocated_slice was created on top of RC-array using 
`slicedField`.
Mar 09 2017
prev sibling next sibling parent Ilya Yaroshenko <ilyayaroshenko gmail.com> writes:
On Thursday, 9 March 2017 at 08:59:28 UTC, Ilya Yaroshenko wrote:


 [1] http://docs.algorithm.dlang.io/latest/mir_ndslice_stack.html


renamed:
http://docs.algorithm.dlang.io/latest/mir_ndslice_concatenation.html
Mar 09 2017
prev sibling parent reply Ilya Yaroshenko <ilyayaroshenko gmail.com> writes:
On Thursday, 9 March 2017 at 08:59:28 UTC, Ilya Yaroshenko wrote:

 Hello

 Ndslice got [1] lazy multidimensional concatenation and padding 
 utilities:
  1. stack
  2. pad
  2. padEdge
  3. padSymmetric
  4. padWrap


 `stack` can be used for ndslices instead of `chain`.

 ---
 import mir.ndslice.allocation: slice;
 import mir.ndslice.stack: stack;

 auto d = stack(a, b, c).slice;
 ---

 ndslice allocation and assign utilities are optimised to work 
 with stacks. For example the code above is significantly faster 
 then

 ---
 import std.array: array;
 import std.range: chain;

 auto d = chain(a, b, c).array;
 ---

 because `slice` uses external iteration for `Stack` data 
 structure.

 [1] http://docs.algorithm.dlang.io/latest/mir_ndslice_stack.html

 Best regards,
 Ilya


`stack` was renamed to `concatenation` to much numpy and Matlab
Mar 09 2017
parent jmh530 <john.michael.hall gmail.com> writes:
On Thursday, 9 March 2017 at 13:46:57 UTC, Ilya Yaroshenko wrote:

 `stack` was renamed to `concatenation` to much numpy and Matlab


Numpy does have stack, but this is more similar to the 
concatenate function (I'd prefer this to concatenation). Matlab 
just calls it cat, which is easier to type, but less informative.

https://docs.scipy.org/doc/numpy-1.10.0/reference/generated/numpy.stack.html
https://docs.scipy.org/doc/numpy-1.10.0/reference/generated/numpy.concatenate.html#numpy.concatenate
https://www.mathworks.com/help/matlab/ref/cat.html
Mar 09 2017