Features/178 tile

[ClaudiaComito]

Description

Implementation of heat version of np.tile():
https://numpy.org/doc/stable/reference/generated/numpy.tile.html

• process-local operations via torch.repeat()
• distributed tiling via Alltoallv
  - also, introducing sanitation module Implement sanitation module #468 (with only a few basic functions), the idea is to expand it on the fly as needed

Issue/s resolved: #178

Changes proposed:

• Implemented new function: manipulations.tile(), after https://numpy.org/doc/stable/reference/generated/numpy.tile.html

Type of change

• New feature (non-breaking change which adds functionality)

Due Diligence

• All split configurations tested
• Multiple dtypes tested in relevant functions
• Documentation updated (if needed)
• Updated changelog.md under the title "Pending Additions"

Does this change modify the behaviour of other functions? If so, which?

no

[codecov]

Codecov Report

Merging #673 (d4f450e) into master (e0af0e5) will increase coverage by 0.00%.
The diff coverage is 94.02%.

@@           Coverage Diff            @@
##           master     #673    +/-   ##
========================================
  Coverage   95.40%   95.40%            
========================================
  Files          64       64            
  Lines        9246     9380   +134     
========================================
+ Hits         8821     8949   +128     
- Misses        425      431     +6     

Flag	Coverage Δ
gpu	94.57% <94.02%> (+0.01%)	⬆️
unit	91.17% <94.02%> (+0.06%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
heat/core/manipulations.py	98.56% <94.02%> (-0.52%)	⬇️
heat/core/indexing.py	100.00% <0.00%> (+4.76%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update e0af0e5...d4f450e. Read the comment docs.

[ClaudiaComito]

Is the change of the split axis intended?

>>> ht.tile(ht.array([1,2,3],split=0), [1,1])
DNDarray([[1, 2, 3]], dtype=ht.int64, device=cpu:0, split=1)

@mtar belated thanks for commenting, yes you are tiling a 1-D array with 2-dimensional repetitions, in this case tile() will prepend the missing dimension (just like numpy.tile()), so the split axis nominally increases by one in this case.

[coquelin77]

rerun tests

[coquelin77]

the code itself looks fine, but i found a bug in the alltoallv logic:

x = ht.arange(3*3).reshape((3, 3)).resplit(1)
print(x)
>>> DNDarray([[0, 1, 2],
              [3, 4, 5],
              [6, 7, 8]], dtype=ht.int32, device=cpu:0, split=1)
y = ht.tile(x, (2, 3))
print(y)
>>> DNDarray([[        0,         4,         2,         0,         4,         2,         0,         4,         2],
              [        1,         1,         0,         1,     22065,     32628,         1,     32628,         0],
              [        3,         0,         0,         3,         0, 724118576,         3,         0,         0],
              [        0,         4,         2,         0,         4,         2,         0,         4,         2],
              [        1,         1,         0,         1,     22065,     32628,         1,     32628,         0],
              [        3,         0,         0,         3,         0, 724118576,         3,         0,         0]], dtype=ht.int32, device=cpu:0, split=1)

[ClaudiaComito]

the code itself looks fine, but i found a bug in the alltoallv logic:

x = ht.arange(3*3).reshape((3, 3)).resplit(1)
print(x)
>>> DNDarray([[0, 1, 2],
              [3, 4, 5],
              [6, 7, 8]], dtype=ht.int32, device=cpu:0, split=1)
y = ht.tile(x, (2, 3))
print(y)
>>> DNDarray([[        0,         4,         2,         0,         4,         2,         0,         4,         2],
              [        1,         1,         0,         1,     22065,     32628,         1,     32628,         0],
              [        3,         0,         0,         3,         0, 724118576,         3,         0,         0],
              [        0,         4,         2,         0,         4,         2,         0,         4,         2],
              [        1,         1,         0,         1,     22065,     32628,         1,     32628,         0],
              [        3,         0,         0,         3,         0, 724118576,         3,         0,         0]], dtype=ht.int32, device=cpu:0, split=1)

Good catch! Fixed

[coquelin77]

the code itself looks fine, but i found a bug in the alltoallv logic:

x = ht.arange(3*3).reshape((3, 3)).resplit(1)
print(x)
>>> DNDarray([[0, 1, 2],
              [3, 4, 5],
              [6, 7, 8]], dtype=ht.int32, device=cpu:0, split=1)
y = ht.tile(x, (2, 3))
print(y)
>>> DNDarray([[        0,         4,         2,         0,         4,         2,         0,         4,         2],
              [        1,         1,         0,         1,     22065,     32628,         1,     32628,         0],
              [        3,         0,         0,         3,         0, 724118576,         3,         0,         0],
              [        0,         4,         2,         0,         4,         2,         0,         4,         2],
              [        1,         1,         0,         1,     22065,     32628,         1,     32628,         0],
              [        3,         0,         0,         3,         0, 724118576,         3,         0,         0]], dtype=ht.int32, device=cpu:0, split=1)

Good catch! Fixed

then this is ready as far as i can tell