π¨ Breaking Changes
π Bug Fixes
- Fix ucx-py update-version script (#4603) @raydouglass
- add setuptools to host requirements for conda packages that need it (#4582) @jameslamb
- Add pylibcugraph dependency on pylibraft. (#4570) @bdice
- Fix build error with NO_CUGRAPH_OPS (#4563) @seunghwak
- [BUG] Fix Failing WholeGraph Tests (#4560) @alexbarghi-nv
- Temporarily Disable Feature Store Tests with WholeGraph (#4559) @alexbarghi-nv
- Fix MG
katz_centrality
: Check if DataFrame Arg is Not None (#4555) @nv-rliu - nx-cugraph: fix
from_pandas_edgekey
given edgekey but not edgeattr (#4550) @eriknw - Fix triangle count test bug (#4549) @jnke2016
- [BUG] Use the Correct WG Communicator (#4548) @alexbarghi-nv
- Add Additional Check For SSSP Source Vertex & Fix SSSP Benchmark (#4541) @nv-rliu
- Fix OOM Bug for Jaccard, Sorensen, and Overlap benchmarks (#4524) @nv-rliu
- Distribute start_list across ranks (#4519) @jnke2016
- [FIX] Skip Distributed Sampler Tests if PyTorch with CUDA is not Available (#4518) @alexbarghi-nv
- [BUG] Fix a hang issue in MG triangle counts (when invoked with a small number of vertices to update triangle counts) (#4517) @seunghwak
- Update MG Benchmark List (#4516) @nv-rliu
- Fix TensorProductConv test and improve docs (#4480) @tingyu66
- Test nx-cugraph package instead of editable install (#4442) @KyleFromNVIDIA
π Documentation
- DOC: typo in nx_transition.rst (#4491) @raybellwaves
- Doc cleanup for nx-cugraph: fixed typos, cleaned up various descriptions, renamed notebook to match naming convetion. (#4478) @rlratzel
- [DOC] Minor Improvements to cuGraph-PyG Documentation (#4460) @alexbarghi-nv
π New Features
- Use MNMG version of ECG in python layer instead, and remove legacy ECG and Louvain (#4514) @naimnv
- c_api and plc binding for lookup src dst using edge ids and type(s) (#4494) @naimnv
- Forward merge branch-24.06 into branch-24.08 (#4489) @nv-rliu
- [FEA] New Graph Interface and Loaders for Distributed Sampling in DGL (#4486) @alexbarghi-nv
- compute cosine similarity for vertex pairs (#4482) @naimnv
- Define heterogeneous renumbering API (#4463) @seunghwak
- Lookup edge src dst using edge id and type (#4449) @naimnv
- Biased sampling (#4443) @seunghwak
π οΈ Improvements
- nx-cugraph: check networkx version (#4571) @eriknw
- nx-cugraph: add
G.__networkx_cache__
to enable graph conversion caching (#4567) @eriknw - split up CUDA-suffixed dependencies in dependencies.yaml (#4552) @jameslamb
- Use workflow branch 24.08 again (#4544) @KyleFromNVIDIA
- Support non p2p configuration when initializing the comms (#4543) @jnke2016
- Fix Warning from
simpleDistributedGraph.py
(#4540) @nv-rliu - Create a graph from the edge list in multiple chunks (#4539) @seunghwak
- nx-cugraph: add dijkstra sssp functions (#4538) @eriknw
- nx-cugraph: add
from_dict_of_lists
andto_dict_of_lists
(#4537) @eriknw - Ensure
get_test_data.sh
doesn't re-download datasets (#4536) @trxcllnt - Define and Implement C API for biased sampling (#4535) @ChuckHastings
- Build and test with CUDA 12.5.1 (#4534) @KyleFromNVIDIA
- Refactor C++ unit tests to allow finer grained filtering (#4533) @ChuckHastings
- [IMP] Set the Default WG Memory Type to 'distributed' for the MNMG PyG Example (#4532) @alexbarghi-nv
- nx-cugraph: add
relabel_nodes
andconvert_node_labels_to_integers
(#4531) @eriknw - Add
-cuXX
suffixed versions of cugraph-service-client dependency to pyproject.toml's project.dependencies list (#4530) @trxcllnt - Further optimize
from_pandas_edgelist
with cudf (#4528) @eriknw - Performance optimize BFS (including direction optimizing BFS implementation, mainly for single-GPU) (#4527) @seunghwak
- Add CUDA_STATIC_MATH_LIBRARIES (#4526) @KyleFromNVIDIA
- Better handle cudf.pandas in
from_pandas_edgelist
(#4525) @eriknw - Skip the benchmark ctests within CI (#4522) @ChuckHastings
- remove thriftpy2 ceiling (#4521) @jameslamb
- Avoid --find-links in wheel jobs (#4509) @jameslamb
- Refactor code base to reduce memory requirement for building libcugraph (#4506) @naimnv
- Tweak rmm configuration for C++ unit tests (#4503) @ChuckHastings
- Expose new all-pairs Similarity algorithms (#4502) @jnke2016
- remove openmpi ceiling (#4496) @jameslamb
- Cut peak memory footprint in per_v_transform_reduce_dst_key_aggregated_outgoing_e (#4484) @seunghwak
- Skip MG
dgl_uniform_sampler
test in nightlies (#4479) @nv-rliu - Remove text builds of documentation (#4468) @vyasr
- [IMP] Limit the Test Data Size when Running CI in
gcn_dist_sg.py
(#4461) @alexbarghi-nv - Forward Merge branch-24.06 into branch-24.08 (#4454) @nv-rliu
- Properly clean up python directories (#4453) @ChuckHastings
- Fixes for On-Going MG Test Failures (#4450) @nv-rliu
- remove unnecessary 'setuptools' and 'wheel' dependencies (#4448) @jameslamb
- MG Implementation K-Truss (#4438) @jnke2016
- Overhaul ops-codeowners (#4409) @raydouglass
- Use rapids-build-backend (#4393) @KyleFromNVIDIA
- Optimize K-Truss (#4375) @jnke2016