Features/179 repeat

[lenablind]

Description

Implementation of function repeat() based on np.repeat()
For process-local operations, I use torch.repeat_interleave

Docs numpy: https://numpy.org/doc/stable/reference/generated/numpy.repeat.html
Docs pytorch: https://pytorch.org/docs/stable/generated/torch.repeat_interleave.html

Strategy

In the following, only the distributed case (<=> a.split != None) will be explained more specifically as the algorithm otherwise results mainly in a direct call of the torch function.

axis is None

Implies a.split = 0 as the (total) result would be in the wrong order otherwise.
To assure the correct distribution of repeats and syntactical consistency with numpy (repeats must be 1-dimensional), repeats has to be reshaped to the global shape of a, redistributed along axis 0 and be flattened afterwards.
The last step is necessary due to compatibility with torch (1-dimensionality required as in numpy).

axis is not None

Depending on whether a.split == axis, repeats has to be either split along axis 0 (case True) or gathered on all processes (case False). The eventually resulting redistribution of data is needed to assure correct local and therefore global results.

Issue/s resolved: #179

Changes proposed:

• Implemented new function: manipulations.repeat()
• Implemented corresponding tests

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

[mtar]

GPU cluster tests are currently disabled on this Pull Request.

[codecov]

Codecov Report

Merging #674 into master will increase coverage by 0.05%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #674      +/-   ##
==========================================
+ Coverage   97.47%   97.53%   +0.05%     
==========================================
  Files          87       87              
  Lines       17231    17643     +412     
==========================================
+ Hits        16796    17208     +412     
  Misses        435      435              

Impacted Files	Coverage Δ
heat/core/manipulations.py	99.28% <100.00%> (+0.06%)	⬆️
heat/core/tests/test_factories.py	100.00% <100.00%> (ø)
heat/core/tests/test_manipulations.py	99.94% <100.00%> (+0.01%)	⬆️

---

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 5cd2b42...229585a. Read the comment docs.

[coquelin77]

looks pretty good. good job!

[lenablind]

looks pretty good. good job!

Thank you!

[mtar]

ok to test

[coquelin77]

very very minor changes and only a couple questions. good work!

[lenablind]

very very minor changes and only a couple questions. good work!

@coquelin77 Thank you!