Bug/squeeze split semantics

[ClaudiaComito]

Description

Changes to manipulations.squeeze() to follow split semantics. In the previous version, the output tensor was being "gathered" and the function returned a non-distributed tensor.

Split semantics for squeeze() are now as follows: a distributed tensor will keep its original split dimension after "squeezing", which, depending on the squeeze axis, may result in a lower numerical 'split' value, as in:

    >>> x.shape
    (10, 1, 12, 13)
    >>> x.split
    2
    >>> x.squeeze().shape
    (10, 12, 13)
    >>> x.squeeze().split
    1

The edge case x.split = axis returns a non-distributed tensor.

Issue/s resolved: #561, #273

Changes proposed:

• I removed the Allgather part and made sure the global output shape was set correctly in returning the DNDarray. I also reinstated the test_squeeze() portion on unevenly distributed tensors.

Type of change

Bug fix, breaking change.

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?

yes if they rely on ht.squeeze() returning a non-distributed tensor.

[codecov]

Codecov Report

Merging #562 into master will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #562      +/-   ##
==========================================
+ Coverage   96.44%   96.45%   +0.01%     
==========================================
  Files          75       75              
  Lines       15130    15145      +15     
==========================================
+ Hits        14592    14608      +16     
+ Misses        538      537       -1     

Impacted Files	Coverage Δ
heat/core/manipulations.py	99.51% <100.00%> (+0.15%)	⬆️
heat/core/tests/test_manipulations.py	99.39% <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 a1c3792...6ca24b8. Read the comment docs.

[coquelin77]

approval pending the addition of 1 more test (or a small change of one). can you make one of them axis=None? just for completeness

[coquelin77]

sorry, these tests are already included. merging now