* [Performance] improve coo2csr space complexity when row is not sorted
* [Perf] replace std::vector<> by NDArray
* keep both impl of unsorted coo to csr and choose according to graph density dynamically
* refine criteria to choose btw Unsorted algos
Co-authored-by: Ubuntu <ubuntu@ip-172-31-34-27.us-west-2.compute.internal>
* fix.
* fix.
* fix.
* fix.
* Fix test
* Deprecate old DistEmbedding impl, use synchronized embedding impl
* Basic imple of heterogeneous on homogenenous sampling
* make pass
* Pass C++ test
* Add python test code
* lint
* lint
* Add MultiLayerEtypeNeighborSampler
* Add unitest for single machine dataloader
* Add dist dataloader test for edge type sampler
* Fix lint
* fix
* support for per etype sample
* Fix some bug and enable distributed training with per edge sample
* fix
* Now distributed training works
* turn off some mxnet
* turn off mxnet for some dist test
* fix
* upd
* upd according to the comments
* Fix
* Fix test and now distributed works.
* upd
* upd
* Fix
* Fix bug
* remove dead code.
* upd
* Fix
* upd
* Fix
Co-authored-by: Ubuntu <ubuntu@ip-172-31-71-112.ec2.internal>
Co-authored-by: Ubuntu <ubuntu@ip-172-31-2-66.ec2.internal>
Co-authored-by: Da Zheng <zhengda1936@gmail.com>
* Split from NCCL PR
* Fix type in comment
* Expand documentation for sparse_all_to_all_push
* Restore previous behavior in example
* Re-work optimizer to use NCCL based on gradient location
* Allow for running with embedding on CPU but using NCCL for gradient exchange
* Optimize single partition case
* Fix pylint errors
* Add missing include
* fix gradient indexing
* Fix line continuation
* Migrate 'first_step'
* Skip tests without enough GPUs to run NCCL
* Improve empty tensor handling for pytorch 1.5
* Fix indentation
* Allow multiple NCCL communicator to coexist
* Improve handling of empty message
* Update python/dgl/nn/pytorch/sparse_emb.py
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>
* Update python/dgl/nn/pytorch/sparse_emb.py
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>
* Keepy empty tensor dimensionaless
* th.empty -> th.tensor
* Preserve shape for empty non-zero dimension tensors
* Use shared state, when embedding is shared
* Add support for gathering an embedding
* Fix typo
* Fix more typos
* Fix backend call
* Use NodeDataLoader to take advantage of ddp
* Update training script to share memory
* Only squeeze last dimension
* Better handle empty message
* Keep embedding on the target device GPU if dgl_sparse if false in RGCN example
* Fix typo in comment
* Add asserts
* Improve documentation in example
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>
* Split NCCL wrapper from sparse optimizer and sparse embedding
* Add more unit tests for single node nccl
* Fix unit test for tf
* Switch to device histogram
* Fix histgram issues
* Finish migration to histogram
* Handle cases with zero send/recieve data
* Start on partition object
* Get compiling
* Updates
* Add unit tests
* Switch to partition object
* Fix linting issues
* Rename partition file
* Add python doc
* Fix python assert and finish doxygen comments
* Remove stubs for range based partition to satisfy pylint
* Wrap unit test in GPU only
* Wrap explicit cuda call in ifdef
* Merge with partition.py
* update docstrings
* Cleanup partition_op
* Add Workspace object
* Switch to using workspace object
* Move last remainder based function out of nccl_api
* Add error messages
* Update docs with examples
* Fix linting erros
Co-authored-by: xiang song(charlie.song) <classicxsong@gmail.com>
* Remove double-checking sorted
* Remove sorting of CSR by default
* Update unit test to use unsorted matix
* delete whitespace
* Expand unit tests
* Replace cusparse sort
* Fix row column sorting
* Explicitly don't sort columns
* Fix linting errors
* Fix bit-width calculation
* Fix sorting assertion and unit test
* Fix linting
* Improve CPU COO2CSR
* Remove references
* Rename and add documentation to edge encoding/decoding funcionts
* Fix sorting keys as 64 bit
* Revert cosmetic changes to unit tests
* Update documentation
* Update complexity documentation for coo to csr conversion
* Remove COOIsSorted check in CPU implementation too
* start
* coo csr union partition
* lint
* lint
* lint
* Add matrix->data transform
* update
* Fix window compile
* Add CSR support for DisjointPartition
* lint
* Fix
* Use IdArray Op
* Concat ready
* Fix and all pass
* resolve comments
* Add union COO C++ test
* Add C++ test for csr
* lint
* triger
* Update include
* Fix merge
* test
Co-authored-by: Ubuntu <ubuntu@ip-172-31-51-214.ec2.internal>
* add cuda utils; change g.to; add g.device
* split array.h into several headers
* cuda index select
* file
* three cuda kernels
* add cuda elementwise arith and several others
* cuda CSRIsNonZero
* fix lint
* lint
* lint
* fix bug in changing ctx to property
* address comments
* remove unused codes
* address comments
* Deprecate multi-graph
* Handle heterograph and edge_ids
* lint
* Fix
* Remove multigraph in C++ end
* Fix lint
* Add some test and fix something
* Fix
* Fix
* upd
* Fix some test case
* Fix
* Fix
Co-authored-by: Ubuntu <ubuntu@ip-172-31-51-214.ec2.internal>
Co-authored-by: Jinjing Zhou <VoVAllen@users.noreply.github.com>
Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>
* Fix coo_sort
* Add COOSort test
* Fix
* omp
* coo2csr can benefit from sorted coo
* Add more test code
* Add LIBCXX_ENABLE_PARALLEL_ALGORITHMS to Makefile
* upd
* Turn off LIBCXX_ENABLE_PARALLEL_ALGORITHMS by default
Co-authored-by: Ubuntu <ubuntu@ip-172-31-51-214.ec2.internal>
* remove edge and to bipartite and graphsage with sampling
* fixes
* fixes
* fixes
* reenable multigpu training
* fixes
* compatibility in DGLGraph
* rename to compact_as_bipartite
* bugfix
* lint
* add offline inference
* skip GPU tests
* fix
* addresses comments
* fix
* fix
* fix
* more tests
* more docs and unit tests
* workaround for empty slice on empty data
* improve performance of sample_neighbors
* some more improve
* test script
* benchmarks
* multi process
* update more tests
* WIP
* adding two API for state saving
* add create from state
* upd test
* missing file
* wip: pickle/unpickle
* more c apis
* find the problem of empty data array
* add null array; pickling speed is bad
* still bad perf
* still bad perf
* wip
* fix the pickle speed test; now everything looks good
* minor fix
* bugfix
* some lint fix
* address comments
* more fix
* fix lint
* add utest for random.choice
* add utest for dgl.rand_graph
* fix cpp utests
* try fix ci
* fix bug in TF backend
* upd choice docstring
* address comments
* upd
* try fix compile
* add comment
* graph format
* fix lint
* lint
* fix
* unit test
* lint
* add magic num
* move serialize out of struct
* lint
Co-authored-by: zhoujinjing09 <zhoujinjing09@users.noreply.github.com>
* trying to refactor IndexSelect
* partial implementation
* add index select and assign for floats as well
* move to random choice source
* more updates
* fixes
* fixes
* more fixes
* adding python impl
* fixes
* unit test
* lint
* lint x2
* lint x3
* update metapath2vec
* debugging performance
* still debugging for performance
* tuning
* switching to succvec
* redo
* revert non-uniform sampler to use vector
* still not fast
* why does this crash with OpenMP???
* because there was a data race!!!
* add documentations and remove assign op
* lint
* lint x2
* lol what have i done
* lint x3
* fix and disable gpu testing
* bugfix
* generic random walk
* reorg the random walk source code
* Update randomwalks.h
* Update randomwalks_cpu.cc
* rename file
* move internal function to anonymous ns
* reorg & docstrings
* constant restart probability
* docstring fix
* more commit
* random walk with restart, tested
* some fixes
* switch to using NDArray for choice
* massive fix & docstring
* lint x?
* lint x??
* fix
* export symbols
* skip gpu test
* addresses comments
* replaces another VecToIdArray
* add randomwalks.h to include
* replace void * with template
Rhett Ying