Features/343 qr

[coquelin77]

Description

Please include a summary of the change and which issue/s is/are fixed.
Please also include relevant motivation and context. List any dependencies that are required for this change.

Fixes: #343

Changes proposed

• added tiling class
• added tiles property to DNDarray
• implemented QR factorization
• abstracted the c_block_setter function from matmul
• reorder functions in DNDarray

Type of change

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

Are all split configurations tested and accounted for?
[x] yes [ ] no
Does this change require a documentation update outside of the changes proposed?
[ ] yes [x] no
Does this change modify the behaviour of other functions?
[ ] yes [x] no
Are there code practices which require justification?
[x] yes [ ] no
-> There are multiple instances of >3 indentations here. Very often this occurs because of rank selection.

[codecov]

Codecov Report

Merging #429 into master will decrease coverage by 0.17%.
The diff coverage is 94.73%.

@@            Coverage Diff            @@
##           master    #429      +/-   ##
=========================================
- Coverage   96.77%   96.6%   -0.18%     
=========================================
  Files          63      65       +2     
  Lines       12872   13991    +1119     
=========================================
+ Hits        12457   13516    +1059     
- Misses        415     475      +60

Impacted Files	Coverage Δ
heat/core/__init__.py	100% <100%> (ø)	⬆️
heat/core/tests/test_communication.py	97.2% <100%> (ø)	⬆️
heat/core/tests/test_tiling.py	100% <100%> (ø)
heat/core/communication.py	89.04% <100%> (+0.06%)	⬆️
heat/core/tests/test_linalg.py	96.29% <71.27%> (-3.06%)	⬇️
heat/core/dndarray.py	96.59% <95%> (-0.11%)	⬇️
heat/core/linalg.py	96.23% <95.34%> (-0.9%)	⬇️
heat/core/tiling.py	96.28% <96.28%> (ø)
... and 1 more

---

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 8b438d2...96649c9. Read the comment docs.

[coquelin77]

I ran the tests successful with 1 to 8 GPUs and CPUs. I have also included an extended tests feature for QR. the loops for this equate to the following:

st_whole = torch.randn(70, 70, device=device)
for sp in range(0, 1):
        for m in range(50, 71, 1):
            for n in range(50, 71, 1):
                for t in range(1, 3):
                    st = st_whole[:m, :n].clone()
                    a_comp = ht.array(st, split=0, device=ht_device)
                    a = ht.array(st, split=sp, device=ht_device)
                    qr = a.qr(tiles_per_proc=t)
                    self.assertTrue(ht.allclose(a_comp, qr.Q @ qr.R, rtol=1e-5, atol=1e-5))
                    self.assertTrue(
                        ht.allclose(
                            qr.Q.T @ qr.Q, ht.eye(m, device=ht_device), rtol=1e-5, atol=1e-5
                        )
                    )
                    self.assertTrue(
                        ht.allclose(
                            ht.eye(m, device=ht_device), qr.Q @ qr.Q.T, rtol=1e-5, atol=1e-5
                        )
                    )

there tests were successfully run on 4,7, and 8 GPUs and CPUs, however they do take about 100x as long. I have successfully run the extended tests for range(30, 100) as well; the change to range(50, 71) was made to reduce the time required. To activate these tests one must add EXTENDED_TESTS=True before the command starting the tests. It is the same syntax as the DEVICE parameter as implemented by @mtar for running the tests on GPUs. As these functions have been shown to work, I propose that they be merged in 2 weeks provided that there are no objections.