make the logger customized with set logger function

we have two options to make customized logger:

1. client makes changes to LOGGING
2. create new set_logger function which allows client to pass customized logger

in this PR, we are implementing the second one.

Bumps numpy from 1.21.0 to 1.22.0.

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

Presently, neigh_nodes are sampled twice (once on Line 74 and a second time when query_func is invoked on Line 79), this results in a disconnected graph.

Before this change:

>>> numpy.array_equal(context['encoder']['node_feats']['neighbor_feats'], context['encoder']['neighbor_feats']['node_feats'])
False

After this change:

>>> numpy.array_equal(context['encoder']['node_feats']['neighbor_feats'], context['encoder']['neighbor_feats']['node_feats'])
True

• [x] Forked repo is synced with upstream -> github shows no code delta outside of the desired. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork
• [x] Tests are passing? https://github.com/microsoft/DeepGNN/blob/main/CONTRIBUTING.md#run-tests
• [x] Documentation is added or updated to reflect new code in the same format as the rest of the repo.
• [x] PR is labeled using the label menu on the right side.

New Behavior

• Add a new converter input format called "Linear" with nodes and edges on separate lines.
• Built in datasets now use linear for speed gains.
• Add to unit tests - convert_tests, e2e_tests and sparse_features_test.

Notes

• No checks for if the file is sorted properly or not.
• Linear is similar to the TSV format, but different enough where they cant really be merged(neighbor edges are on the same line as node, feature encoding).

• [x] Forked repo is synced with upstream -> github shows no code delta outside of the desired. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork
• [x] Tests are passing? https://github.com/microsoft/DeepGNN/blob/main/CONTRIBUTING.md#run-tests
• [x] Changelog and documentation updated.
• [x] PR is labeled using the label menu on the right side.

New Behavior

Add example for what pytorch examples will look like after migrate script.

Bumps setuptools from 59.6.0 to 65.5.1.

Dependabot will resolve any conflicts with this PR as long as you don't alter it yourself. You can also trigger a rebase manually by commenting @dependabot rebase.

• [ ] Issue is labeled using the label menu on the right side.

Environment

• Python version: (python -V) 3.7.7
• deepgnn-ge Version: (python -m pip show deepgnn-ge) 0.1.55.1
• deepgnn-torch Version: (python -m pip show deepgnn-torch) 0.1.55.1
• deepgnn-tf Version: (python -m pip show deepgnn-tf) not installed
• OS: (Windows, Linux, ...) Windows 10 Enterprise

Issue Details

• What you did - code sample or commands run

I installed deepgnn-torch via pip in a virtual environment. Then I cloned the deepgnn repository, cd-ed into the examples/pytorch/gat/ and then ran bash run.sh

• Expected behavior

I expected the training script to run without issues.

• Actual behavior

I see the error File "c:\users\myid\appdata\local\programs\python\python37\lib\multiprocessing\reduction.py", line 60, in dump ForkingPickler(file, protocol).dump(obj) AttributeError: Can't pickle local object 'CDLL.__init__.<locals>._FuncPtr'

Full stack trace:

$ bash run.sh
+ DEVICE=cpu
++ dirname run.sh
+ DIR_NAME=.
+ GRAPH=/tmp/cora
+ python -m deepgnn.graph_engine.data.citation --data_dir /tmp/cora
c:\users\myid\appdata\local\programs\python\python37\lib\runpy.py:125: RuntimeWarning: 'deepgnn.graph_engine.data.citation' found in sys.modules after import of package 'deepgnn.graph_engine.data', but prior to execution of 'deepgnn.graph_engine.data.citation'; this may result in unpredictable behaviour
  warn(RuntimeWarning(msg))
[2022-09-09 14:46:04,150] {convert.py:100} INFO - worker 0 try to generate partition: 0 - 1
[2022-09-09 14:46:04,151] {_adl_reader.py:124} INFO - [1,0] Input files: ['C:/Users/myid/AppData/Local/Temp/cora\\graph.json']
[2022-09-09 14:46:04,782] {dispatcher.py:143} INFO - record processed: 1000
[2022-09-09 14:46:05,257] {dispatcher.py:143} INFO - record processed: 2000
[2022-09-09 14:46:05,657] {local.py:44} INFO - Graph data path: C:/Users/myid/AppData/Local/Temp/cora. Partitions [0]. Storage type 0. Config path . Stream False.
[2022-09-09 14:46:05,707] {local.py:52} INFO - Loaded snark graph. Node counts: [140, 500, 1000, 1068]. Edge counts: [10556]
graph data: C:/Users/myid/AppData/Local/Temp/cora
+ MODEL_DIR=/tmp/model_fix
+ rm -rf /tmp/model_fix
+ [[ cpu == \g\p\u ]]
+ python ./main.py --data_dir /tmp/cora --mode train --seed 123 --backend snark --graph_type local --converter skip --batch_size 140 --learning_rate 0.005 --num_epochs 180 --sample_file /tmp/cora/train.nodes --node_type 0 --model_dir /tmp/model_fix --metric_dir /tmp/model_fix --save_path /tmp/model_fix --eval_file /tmp/cora/test.nodes --eval_during_train_by_steps 1 --feature_idx 0 --feature_dim 1433 --label_idx 1 --label_dim 1 --head_num 8,1 --num_classes 7 --neighbor_edge_types 0 --attn_drop 0.6 --ffd_drop 0.6 --log_by_steps 1 --use_per_step_metrics
[2022-09-09 14:46:08,646] {factory.py:38} INFO - GE_OMP_NUM_THREADS=1
[2022-09-09 14:46:08,647] {factory.py:38} INFO - apex_opt_level=O2
[2022-09-09 14:46:08,647] {factory.py:38} INFO - attn_drop=0.6
[2022-09-09 14:46:08,647] {factory.py:38} INFO - backend=snark
[2022-09-09 14:46:08,647] {factory.py:38} INFO - batch_size=140
[2022-09-09 14:46:08,647] {factory.py:38} INFO - client_rank=None
[2022-09-09 14:46:08,647] {factory.py:38} INFO - clip_grad=False
[2022-09-09 14:46:08,647] {factory.py:38} INFO - config_path=
[2022-09-09 14:46:08,647] {factory.py:38} INFO - converter=skip
[2022-09-09 14:46:08,647] {factory.py:38} INFO - data_dir=C:/Users/myid/AppData/Local/Temp/cora
[2022-09-09 14:46:08,647] {factory.py:38} INFO - data_parallel_num=2
[2022-09-09 14:46:08,647] {factory.py:38} INFO - dim=256
[2022-09-09 14:46:08,647] {factory.py:38} INFO - disable_ib=False
[2022-09-09 14:46:08,647] {factory.py:38} INFO - enable_adl_uploader=False
[2022-09-09 14:46:08,647] {factory.py:38} INFO - enable_ssl=False
[2022-09-09 14:46:08,647] {factory.py:38} INFO - eval_during_train_by_steps=1
[2022-09-09 14:46:08,647] {factory.py:38} INFO - eval_file=C:/Users/myid/AppData/Local/Temp/cora/test.nodes
[2022-09-09 14:46:08,647] {factory.py:38} INFO - fanouts=[10, 10]
[2022-09-09 14:46:08,647] {factory.py:38} INFO - featenc_config=None
[2022-09-09 14:46:08,647] {factory.py:38} INFO - feature_dim=1433
[2022-09-09 14:46:08,648] {factory.py:38} INFO - feature_idx=0
[2022-09-09 14:46:08,648] {factory.py:38} INFO - feature_type=float
[2022-09-09 14:46:08,648] {factory.py:38} INFO - ffd_drop=0.6
[2022-09-09 14:46:08,648] {factory.py:38} INFO - fp16=amp
[2022-09-09 14:46:08,648] {factory.py:38} INFO - ge_start_timeout=30
[2022-09-09 14:46:08,648] {factory.py:38} INFO - gpu=False
[2022-09-09 14:46:08,648] {factory.py:38} INFO - grad_max_norm=1.0
[2022-09-09 14:46:08,648] {factory.py:38} INFO - graph_type=local
[2022-09-09 14:46:08,648] {factory.py:38} INFO - head_num=[8, 1]
[2022-09-09 14:46:08,648] {factory.py:38} INFO - hidden_dim=8
[2022-09-09 14:46:08,648] {factory.py:38} INFO - job_id=aa812d6f
[2022-09-09 14:46:08,648] {factory.py:38} INFO - l2_coef=0.0005
[2022-09-09 14:46:08,648] {factory.py:38} INFO - label_dim=1
[2022-09-09 14:46:08,648] {factory.py:38} INFO - label_idx=1
[2022-09-09 14:46:08,648] {factory.py:38} INFO - learning_rate=0.005
[2022-09-09 14:46:08,648] {factory.py:38} INFO - local_rank=0
[2022-09-09 14:46:08,648] {factory.py:38} INFO - log_by_steps=1
[2022-09-09 14:46:08,648] {factory.py:38} INFO - max_id=None
[2022-09-09 14:46:08,648] {factory.py:38} INFO - max_samples=0
[2022-09-09 14:46:08,648] {factory.py:38} INFO - max_saved_ckpts=0
[2022-09-09 14:46:08,648] {factory.py:38} INFO - meta_dir=
[2022-09-09 14:46:08,648] {factory.py:38} INFO - metric_dir=C:/Users/myid/AppData/Local/Temp/model_fix
[2022-09-09 14:46:08,649] {factory.py:38} INFO - mode=train
[2022-09-09 14:46:08,649] {factory.py:38} INFO - model_args=
[2022-09-09 14:46:08,649] {factory.py:38} INFO - model_dir=C:/Users/myid/AppData/Local/Temp/model_fix
[2022-09-09 14:46:08,649] {factory.py:38} INFO - neighbor_count=10
[2022-09-09 14:46:08,649] {factory.py:38} INFO - neighbor_edge_types=[0]
[2022-09-09 14:46:08,649] {factory.py:38} INFO - node_type=0
[2022-09-09 14:46:08,649] {factory.py:38} INFO - num_classes=7
[2022-09-09 14:46:08,649] {factory.py:38} INFO - num_epochs=180
[2022-09-09 14:46:08,649] {factory.py:38} INFO - num_ge=0
[2022-09-09 14:46:08,649] {factory.py:38} INFO - num_negs=5
[2022-09-09 14:46:08,649] {factory.py:38} INFO - num_parallel=2
[2022-09-09 14:46:08,649] {factory.py:38} INFO - partitions=[0]
[2022-09-09 14:46:08,649] {factory.py:38} INFO - prefetch_factor=2
[2022-09-09 14:46:08,649] {factory.py:38} INFO - prefetch_size=16
[2022-09-09 14:46:08,649] {factory.py:38} INFO - sample_file=C:/Users/myid/AppData/Local/Temp/cora/train.nodes
[2022-09-09 14:46:08,649] {factory.py:38} INFO - save_ckpt_by_epochs=1
[2022-09-09 14:46:08,649] {factory.py:38} INFO - save_ckpt_by_steps=0
[2022-09-09 14:46:08,649] {factory.py:38} INFO - save_path=C:/Users/myid/AppData/Local/Temp/model_fix
[2022-09-09 14:46:08,649] {factory.py:38} INFO - seed=123
[2022-09-09 14:46:08,649] {factory.py:38} INFO - server_idx=None
[2022-09-09 14:46:08,649] {factory.py:38} INFO - servers=
[2022-09-09 14:46:08,649] {factory.py:38} INFO - skip_ge_start=False
[2022-09-09 14:46:08,649] {factory.py:38} INFO - sort_ckpt_by_mtime=False
[2022-09-09 14:46:08,649] {factory.py:38} INFO - ssl_cert=
[2022-09-09 14:46:08,650] {factory.py:38} INFO - storage_type=0
[2022-09-09 14:46:08,650] {factory.py:38} INFO - strategy=RandomWithoutReplacement
[2022-09-09 14:46:08,650] {factory.py:38} INFO - stream=False
[2022-09-09 14:46:08,650] {factory.py:38} INFO - sync_dir=
[2022-09-09 14:46:08,650] {factory.py:38} INFO - trainer=base
[2022-09-09 14:46:08,650] {factory.py:38} INFO - uploader_process_num=1
[2022-09-09 14:46:08,650] {factory.py:38} INFO - uploader_store_name=
[2022-09-09 14:46:08,650] {factory.py:38} INFO - uploader_threads_num=12
[2022-09-09 14:46:08,650] {factory.py:38} INFO - use_per_step_metrics=True
[2022-09-09 14:46:08,650] {factory.py:38} INFO - user_name=10.0.0.200
[2022-09-09 14:46:08,650] {factory.py:38} INFO - warmup=0.0002
[2022-09-09 14:46:08,654] {local.py:44} INFO - Graph data path: C:/Users/myid/AppData/Local/Temp/cora. Partitions [0]. Storage type 0. Config path . Stream False.
[2022-09-09 14:46:08,666] {local.py:52} INFO - Loaded snark graph. Node counts: [140, 500, 1000, 1068]. Edge counts: [10556]
[2022-09-09 14:46:08,666] {main.py:37} INFO - Creating GAT model with seed:123.
[2022-09-09 14:46:08,668] {base_model.py:39} INFO - [BaseModel] feature_type: FeatureType.FLOAT, feature_idx:0, feature_dim:0.
[2022-09-09 14:46:08,672] {trainer.py:472} INFO - [1,0] Max steps per epoch:-1
[2022-09-09 14:46:08,672] {utils.py:107} INFO - 0, input_layer.att_head-0.bias: torch.Size([8]), cpu
[2022-09-09 14:46:08,672] {utils.py:107} INFO - 1, input_layer.att_head-0.w.weight: torch.Size([8, 1433]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 2, input_layer.att_head-0.attn_l.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 3, input_layer.att_head-0.attn_l.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 4, input_layer.att_head-0.attn_r.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 5, input_layer.att_head-0.attn_r.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 6, input_layer.att_head-1.bias: torch.Size([8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 7, input_layer.att_head-1.w.weight: torch.Size([8, 1433]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 8, input_layer.att_head-1.attn_l.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 9, input_layer.att_head-1.attn_l.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 10, input_layer.att_head-1.attn_r.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 11, input_layer.att_head-1.attn_r.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 12, input_layer.att_head-2.bias: torch.Size([8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 13, input_layer.att_head-2.w.weight: torch.Size([8, 1433]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 14, input_layer.att_head-2.attn_l.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 15, input_layer.att_head-2.attn_l.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 16, input_layer.att_head-2.attn_r.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 17, input_layer.att_head-2.attn_r.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 18, input_layer.att_head-3.bias: torch.Size([8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 19, input_layer.att_head-3.w.weight: torch.Size([8, 1433]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 20, input_layer.att_head-3.attn_l.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 21, input_layer.att_head-3.attn_l.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 22, input_layer.att_head-3.attn_r.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 23, input_layer.att_head-3.attn_r.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 24, input_layer.att_head-4.bias: torch.Size([8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 25, input_layer.att_head-4.w.weight: torch.Size([8, 1433]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 26, input_layer.att_head-4.attn_l.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 27, input_layer.att_head-4.attn_l.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 28, input_layer.att_head-4.attn_r.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,673] {utils.py:107} INFO - 29, input_layer.att_head-4.attn_r.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 30, input_layer.att_head-5.bias: torch.Size([8]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 31, input_layer.att_head-5.w.weight: torch.Size([8, 1433]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 32, input_layer.att_head-5.attn_l.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 33, input_layer.att_head-5.attn_l.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 34, input_layer.att_head-5.attn_r.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 35, input_layer.att_head-5.attn_r.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 36, input_layer.att_head-6.bias: torch.Size([8]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 37, input_layer.att_head-6.w.weight: torch.Size([8, 1433]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 38, input_layer.att_head-6.attn_l.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 39, input_layer.att_head-6.attn_l.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 40, input_layer.att_head-6.attn_r.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 41, input_layer.att_head-6.attn_r.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 42, input_layer.att_head-7.bias: torch.Size([8]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 43, input_layer.att_head-7.w.weight: torch.Size([8, 1433]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 44, input_layer.att_head-7.attn_l.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 45, input_layer.att_head-7.attn_l.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 46, input_layer.att_head-7.attn_r.weight: torch.Size([1, 8]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 47, input_layer.att_head-7.attn_r.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 48, out_layer.att_head-0.bias: torch.Size([7]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 49, out_layer.att_head-0.w.weight: torch.Size([7, 64]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 50, out_layer.att_head-0.attn_l.weight: torch.Size([1, 7]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 51, out_layer.att_head-0.attn_l.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 52, out_layer.att_head-0.attn_r.weight: torch.Size([1, 7]), cpu
[2022-09-09 14:46:08,674] {utils.py:107} INFO - 53, out_layer.att_head-0.attn_r.bias: torch.Size([1]), cpu
[2022-09-09 14:46:08,675] {utils.py:116} INFO - parameter count: 92391
[2022-09-09 14:46:08,675] {logging_utils.py:84} INFO - Training worker started. Model: GAT.
Traceback (most recent call last):
  File "./main.py", line 126, in <module>
    _main()
  File "./main.py", line 121, in _main
    init_args_fn=init_args,
  File "C:\Users\myid\Downloads\DeepGNN\venv\lib\site-packages\deepgnn\pytorch\training\factory.py", line 134, in run_dist
    eval_dataloader_for_training,
  File "C:\Users\myid\Downloads\DeepGNN\venv\lib\site-packages\deepgnn\pytorch\training\trainer.py", line 100, in run
    self._train(model)
  File "C:\Users\myid\Downloads\DeepGNN\venv\lib\site-packages\deepgnn\pytorch\training\trainer.py", line 171, in _train
    self._train_one_epoch(model, epoch)
  File "C:\Users\myid\Downloads\DeepGNN\venv\lib\site-packages\deepgnn\pytorch\training\trainer.py", line 174, in _train_one_epoch
    for i, data in enumerate(self.dataset):
  File "C:\Users\myid\Downloads\DeepGNN\venv\lib\site-packages\torch\utils\data\dataloader.py", line 444, in __iter__
    return self._get_iterator()
  File "C:\Users\myid\Downloads\DeepGNN\venv\lib\site-packages\torch\utils\data\dataloader.py", line 390, in _get_iterator
    return _MultiProcessingDataLoaderIter(self)
  File "C:\Users\myid\Downloads\DeepGNN\venv\lib\site-packages\torch\utils\data\dataloader.py", line 1077, in __init__
    w.start()
  File "c:\users\myid\appdata\local\programs\python\python37\lib\multiprocessing\process.py", line 112, in start
    self._popen = self._Popen(self)
  File "c:\users\myid\appdata\local\programs\python\python37\lib\multiprocessing\context.py", line 223, in _Popen
    return _default_context.get_context().Process._Popen(process_obj)
  File "c:\users\myid\appdata\local\programs\python\python37\lib\multiprocessing\context.py", line 322, in _Popen
    return Popen(process_obj)
  File "c:\users\myid\appdata\local\programs\python\python37\lib\multiprocessing\popen_spawn_win32.py", line 89, in __init__
    reduction.dump(process_obj, to_child)
  File "c:\users\myid\appdata\local\programs\python\python37\lib\multiprocessing\reduction.py", line 60, in dump
    ForkingPickler(file, protocol).dump(obj)
AttributeError: Can't pickle local object 'CDLL.__init__.<locals>._FuncPtr'

Bug Description: Previously, indices idx (output of numpy.unique) of 1-hop neighbors (i.e. neigh_feats_unique computed on Line 83) were being used to reconstruct features of 2-hop neighbors (i.e. neigh_feats_unique["neighbor_feats"] computed on Line 90); this results in slicing of feature matrix for 2-hop neighbors.

This change drops the indexing on Line 90.

Steps to reproduce: (Dataset: Cora; Fanout: [10, 10]; Batch-Size: 140; Sampling: RandomWithoutReplacement)

Before this code change:

>>> context['encoder']['node_feats']['node_feats'].shape
(140, 50)

>>> context['encoder']['node_feats']['neighbor_feats'].shape
(1400, 50)

>>> context['encoder']['neighbor_feats']['node_feats'].shape
(1400, 50)

>>> context['encoder']['neighbor_feats']['neighbor_feats'].shape
(1400, 50)

After this code change:

>>> context['encoder']['node_feats']['node_feats'].shape
(140, 50)

>>> context['encoder']['node_feats']['neighbor_feats'].shape
(1400, 50)

>>> context['encoder']['neighbor_feats']['node_feats'].shape
(1400, 50)

>>> context['encoder']['neighbor_feats']['neighbor_feats'].shape
(4160, 50)

• [ ] Forked repo is synced with upstream -> github shows no code delta outside of the desired. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork
• [ ] Tests are passing? https://github.com/microsoft/DeepGNN/blob/main/CONTRIBUTING.md#run-tests
• [ ] Documentation is added or updated to reflect new code in the same format as the rest of the repo.
• [ ] PR is labeled using the label menu on the right side.

Previous Behavior

• Provide relevant issue number if applicable, eg #44.

New Behavior

• [X ] Forked repo is synced with upstream -> github shows no code delta outside of the desired. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork
• [X ] Tests are passing? https://github.com/microsoft/DeepGNN/blob/main/CONTRIBUTING.md#run-tests
• [X ] Documentation is added or updated to reflect new code in the same format as the rest of the repo.
• [ X] PR is labeled using the label menu on the right side.

Previous Behavior

• Provide relevant issue number if applicable, eg #44.

New Behavior

• Adds neighbor count (node degree) functionality for both memory graph and distributed graphs
• Adds testing for neighbor count feature for memory/distributed graph (base cases, single/multiple partition, invalid nodes and types)
• Adds Python handling of c functions leveraging ctypes library

hi,

I am trying GAT example and I see that the model only returns loss, pred and labels, how can I get node id and node embedding returned?

    def forward(self, inputs):
        """Evaluate model, calculate loss, predictions and extract labels."""
        # fmt: off
        nodes, feat, mask, labels, edges, edges_value, adj_shape = inputs
        nodes = torch.squeeze(nodes)                # [N], N: num of nodes in subgraph
        feat = torch.squeeze(feat)                  # [N, F]
        mask = torch.squeeze(mask)                  # [N]
        labels = torch.squeeze(labels)              # [N]
        edges = torch.squeeze(edges)                # [X, 2], X: num of edges in subgraph
        edges_value = torch.squeeze(edges_value)    # [X]
        adj_shape = torch.squeeze(adj_shape)        # [2]
        # fmt: on

        sp_adj = torch.sparse_coo_tensor(edges, edges_value, adj_shape.tolist())
        h_1 = self.input_layer(feat, sp_adj)
        scores = self.out_layer(h_1, sp_adj)

        labels = labels.type(torch.int64)
        labels = labels[mask]  # [batch_size]
        scores = scores[mask]  # [batch_size]
        pred = scores.argmax(dim=1)
        loss = self.xent(scores, labels)
        return loss, pred, labels

Externalize graph format

• Make easier to convert from parquet/orvo to our input format.
• Support neighbor inversion // sorting by dst nodes.
• JSON can't handle high number of edges well.

Patching CVE-2007-4559

Hi, we are security researchers from the Advanced Research Center at Trellix. We have began a campaign to patch a widespread bug named CVE-2007-4559. CVE-2007-4559 is a 15 year old bug in the Python tarfile package. By using extract() or extractall() on a tarfile object without sanitizing input, a maliciously crafted .tar file could perform a directory path traversal attack. We found at least one unsantized extractall() in your codebase and are providing a patch for you via pull request. The patch essentially checks to see if all tarfile members will be extracted safely and throws an exception otherwise. We encourage you to use this patch or your own solution to secure against CVE-2007-4559. Further technical information about the vulnerability can be found in this blog.

If you have further questions you may contact us through this projects lead researcher Kasimir Schulz.

• [x] Forked repo is synced with upstream -> github shows no code delta outside of the desired. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork
• [x] Tests are passing? https://github.com/microsoft/DeepGNN/blob/main/CONTRIBUTING.md#run-tests
• [x] Changelog and documentation updated.
• [x] PR is labeled using the label menu on the right side.

Previous Behavior

• We always use most recent release in bazel install.

New Behavior

Fix to version 5.4.0 (latest working) bazel version in install instructions, tested on new machine.

• [x] Forked repo is synced with upstream -> github shows no code delta outside of the desired. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork
• [x] Tests are passing? https://github.com/microsoft/DeepGNN/blob/main/CONTRIBUTING.md#run-tests
• [x] Changelog and documentation updated.
• [x] PR is labeled using the label menu on the right side.

New Behavior

Add documentation for example on how to use Ray with Tensorflow. -> This is a repeat of a previous pull request, this is meant to merge in with the torch ray_usage docs before the migration.

• [x] Forked repo is synced with upstream -> github shows no code delta outside of the desired. https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork
• [x] Tests are passing? https://github.com/microsoft/DeepGNN/blob/main/CONTRIBUTING.md#run-tests
• [x] Changelog and documentation updated.
• [x] PR is labeled using the label menu on the right side.

Previous Behavior

• We implement custom trainers and use them in all examples.

New Behavior

• Add migrate script from custom trainers to Ray Train Torch.
• Remove all custom trainers.
• Next PR will migrate all examples to train torch.
• Migrate all models to use Ray using migrate script.
• Unit tests leverage same training loop using Ray as run.sh.
• Removed redundant // overly complex tests for some models.

• [x] Tests are passing? https://github.com/microsoft/DeepGNN/blob/main/CONTRIBUTING.md#run-tests
• [x] Changelog and documentation updated.
• [x] PR is labeled using the label menu on the right side.

Previous Behavior

1. Only static features supported.
2. Feature extraction relied on using node types/edge flags to identify missing features.

New Behavior

1. Every node/edge has an optional parameter timestamp associated with it. If parameter is present, then we are going to do a binary search to find the latest feature. Otherwise return raw feature data.
2. Use feature timestamps buffer to identify missing features and fill with zeros if it is missing.

• [x] Tests are passing? https://github.com/microsoft/DeepGNN/blob/main/CONTRIBUTING.md#run-tests
• [x] Changelog and documentation updated.
• [x] PR is labeled using the label menu on the right side.

Previous Behavior

Static graphs without any time dependencies across its components.

New Behavior

Add dynamic behavior to graphs in 4 parts:

Documentation Feature extraction Neighbor sampling Converters, python E2E tests