Bug/median

[ClaudiaComito]

Description

ht.median(a, axis) (aka ht.percentile(a, q=50, axis)) was returning erroneous results, or sometimes an empty DNDarray, when a.split == axis and the number of mpi processes was larger than 2.

Problems:

• line 1437, perc_chunk was being calculated based on the wrong rank value, which led to perc_slice being off and percentile() returning wrong (or "empty") values. Fixed.
• indices on ranks > 0 not being corrected for the presence of halo data. This affects the result when the relevant data aren't on rank 0, which is always the case for median() on more than 2 processes. Fixed at lines 1441 and ff.

Not occurring in #623 but also a problem:

• dtype of output was promote_type(data.dtype, q.dtype), but we might have integer data and integer q (e.g. q=50) and still need a float result. Fixed at lines 1326-1336.

Issue/s resolved: #623

Changes proposed:

• see above, plus:
• implemented explicit test for median in test_percentile()

Type of change

• Bug fix (non-breaking change which fixes an issue)

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 #624 into master will increase coverage by 0.02%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #624      +/-   ##
==========================================
+ Coverage   97.47%   97.49%   +0.02%     
==========================================
  Files          77       77              
  Lines       15862    15869       +7     
==========================================
+ Hits        15461    15472      +11     
+ Misses        401      397       -4     

Impacted Files	Coverage Δ
heat/core/statistics.py	97.32% <100.00%> (+0.64%)	⬆️
heat/core/tests/test_statistics.py	100.00% <100.00%> (ø)

---

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 89bfd29...82c7afb. Read the comment docs.