Support batched QR decomposition

[mrfh92]

Due Diligence

• General:
  • title of the PR is suitable to appear in the Release Notes
• Implementation:
  • unit tests: all split configurations tested
  • unit tests: multiple dtypes tested
  • benchmarks: created for new functionality
  • (within the error margin) benchmarks: performance improved or maintained
  • documentation updated where needed

Description

Batched QR

Issue/s resolved: necessary for #1696

Changes proposed:

ht.linalg.qr now is able to compute QR for batches of matrices, as PyTorch's batched linear algebra does

Type of change

(not really) new feature

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

yes (ht.linalg.qr)

[github-actions]

Thank you for the PR!

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 92.21%. Comparing base (c23428e) to head (c8702a8).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1720      +/-   ##
==========================================
- Coverage   92.21%   92.21%   -0.01%     
==========================================
  Files          83       83              
  Lines       12318    12316       -2     
==========================================
- Hits        11359    11357       -2     
  Misses        959      959              

Flag	Coverage Δ
unit	92.21% <100.00%> (-0.01%)	⬇️

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

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[github-actions]

Thank you for the PR!

[JuanPedroGHM]

Benchmarks results - Sponsored by perun

function	mpi_ranks	device	metric	value	ref_value	std	% change	type	alert	lower_quantile	upper_quantile
matmul_split_0	4	CPU	RUNTIME	0.13726	0.181238	0.0256314	-24.265	jump-detection	True	nan	nan
apply_inplace_normalizer	4	CPU	RUNTIME	0.00320091	0.00104532	0.00649931	206.213	jump-detection	True	nan	nan
matmul_split_0	4	GPU	RUNTIME	0.033585	0.0562632	0.0285393	-40.3074	jump-detection	True	nan	nan
matmul_split_1	4	GPU	RUNTIME	0.0259047	0.0343244	0.0123812	-24.5298	jump-detection	True	nan	nan
concatenate	4	GPU	RUNTIME	0.173058	0.142046	0.0596402	21.8327	jump-detection	True	nan	nan
qr_split_1	4	CPU	RUNTIME	0.197579	0.186863	0.0262833	5.73458	trend-deviation	True	0.179407	0.196056
hierachical_svd_tol	4	CPU	RUNTIME	0.0513506	0.0521396	0.000182696	-1.51317	trend-deviation	True	0.0515701	0.0531136
reshape	4	CPU	RUNTIME	0.155652	0.160234	0.00165965	-2.85945	trend-deviation	True	0.156578	0.167456
resplit	4	CPU	RUNTIME	1.07068	1.0918	0.00795151	-1.93423	trend-deviation	True	1.0711	1.11149
apply_inplace_robust_scaler_and_inverse	4	CPU	RUNTIME	2.36339	2.45182	0.0576843	-3.60671	trend-deviation	True	2.38057	2.56004
matmul_split_0	4	GPU	RUNTIME	0.033585	0.0501268	0.0285393	-33	trend-deviation	True	0.0339485	0.0687051
lanczos	4	GPU	RUNTIME	0.618193	0.595779	0.00583022	3.76207	trend-deviation	True	0.581043	0.617789
hierachical_svd_tol	4	GPU	RUNTIME	0.123252	0.119621	0.000163931	3.03512	trend-deviation	True	0.117207	0.122445
kmeans	4	GPU	RUNTIME	0.680437	0.646702	0.00515695	5.21645	trend-deviation	True	0.623644	0.677172
apply_inplace_robust_scaler_and_inverse	4	GPU	RUNTIME	6.21877	5.8251	0.0352397	6.75803	trend-deviation	True	5.51667	6.17805

Grafana Dashboard
Last updated: 2024-11-15T08:33:51Z

[github-actions]

Thank you for the PR!

[ClaudiaComito]

Thanks a lot @mrfh92! I only have 2 tiny changes, otherwise it's ready as far as I can tell. 👍🏼

[github-actions]

Thank you for the PR!

[github-actions]

Thank you for the PR!

[mrfh92]

@JuanPedroGHM looks like the BM failed because the gitlab runner had to wait more than 01:30h for getting the job allocated. Is the cluster currently busy or out-of-order?

[mrfh92]

@ClaudiaComito I turned off the benchmark_pr as it does not work at the moment; except of that everything should be fine now (the previour run of the benchmark already has shown that there is no performance problem associated with the changes)

[github-actions]

Thank you for the PR!

[github-actions]

Thank you for the PR!

[ClaudiaComito]

Thanks again @mrfh92 !