Stochastic Training on Batched Hetero-Graphs

Overview

Sup! I’ve been working through the tutorial on node classification with neighborhood sampling with my own data. I generate a few thousand bidirectional heterographs (G_1, G_2, …, G_n) with two node types (a & b), which I batch using dgl.batch.

The output for one of these graphs is below.

  Graph(num_nodes={'a': 94, 'b': 500},
  num_edges={('a', 'link', 'b'): 2000, ('b', 'link', 'a'): 2000},
  metagraph=[('a', 'b', 'link'), ('b', 'a', 'link')])

I want to use the batched graph and the MultiLayerFullNeighborSampler, to generate the blocks for a Stochastic 2 Layer GCN. However at the moment I’m just trying to get the sampler to work on a single graph. I’m still using the dgl.dataloading.MultiLayerFullNeighborSampler and dgl.dataloading.NodeDataLoader defaults from the tutorial, execpt I’m testing on just a single GNN layer.

The sampler and dataloader returns without error: the input_nodes, output_nodes, and blocks as below.

  [
  {'a': tensor([19, 43, 67, 90], dtype=torch.int32),'b': tensor([389], dtype=torch.int32)},
  {'a': tensor([], dtype=torch.int32), 'b': tensor([389], dtype=torch.int32)},
  [Block(num_src_nodes={'a': 4, 'b': 1},num_dst_nodes={'a': 0, 'b': 1},num_edges={('a', 'link', 'b'): 4},metagraph=[('a', 'b', 'link')])]
  ]

Problem

All set to do some prediction I set up a single layer to test on:

  test_layer = dglnn.HeteroGraphConv({rel : dglnn.GraphConv(1, 2, norm='right')for rel in ['link']})

Calling forward on this test layer with the block and srcdata gives the following error.

AssertionError: Current HeteroNodeDataView has multiple node types, can not be iterated.

What am I doing wrong?

Things I’ve tried and it hasn’t helped

  • uniquely identifying the two connections when generating the heterograph ie {(a,link_a,b),(b,link_b,a)}
  • having just a one directional heterograph
  • different batch sizes greater than 1
  • did a deep trace to ensure all the pytorch tensors were formatted correctly

What I think is wrong

  • could be that I’ve structured my nid_training_dict incorrectly for the dataloader(see below it’s just all nodes atm)
  • maybe I’ve built my graph incorrectly or I’m missing something there?

Potentially Useful Outputs

This is the format of my nid_training_dict

	{'a': tensor([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16, 17,
			 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35,
			 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53,
			 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71,
			 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89,
			 90, 91, 92, 93], dtype=torch.int32),
	 'b': tensor([  0,   1,   2,   3,   4,   5,   6,   7,   8,   9,  10,  11,  12,  13,
			  14,  15,  16,  17,  18,  19,  20,  21,  22,  23,  24,  25,  26,  27,
			  28,  29,  30,  31,  32,  33,  34,  35,  36,  37,  38,  39,  40,  41,
			  42,  43,  44,  45,  46,  47,  48,  49,  50,  51,  52,  53,  54,  55,
			  56,  57,  58,  59,  60,  61,  62,  63,  64,  65,  66,  67,  68,  69,
			  70,  71,  72,  73,  74,  75,  76,  77,  78,  79,  80,  81,  82,  83,
			  84,  85,  86,  87,  88,  89,  90,  91,  92,  93,  94,  95,  96,  97,
			  98,  99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111,
			 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125,
			 126, 127, 128, 129, 130, 131, 132, 133, 134, 135, 136, 137, 138, 139,
			 140, 141, 142, 143, 144, 145, 146, 147, 148, 149, 150, 151, 152, 153,
			 154, 155, 156, 157, 158, 159, 160, 161, 162, 163, 164, 165, 166, 167,
			 168, 169, 170, 171, 172, 173, 174, 175, 176, 177, 178, 179, 180, 181,
			 182, 183, 184, 185, 186, 187, 188, 189, 190, 191, 192, 193, 194, 195,
			 196, 197, 198, 199, 200, 201, 202, 203, 204, 205, 206, 207, 208, 209,
			 210, 211, 212, 213, 214, 215, 216, 217, 218, 219, 220, 221, 222, 223,
			 224, 225, 226, 227, 228, 229, 230, 231, 232, 233, 234, 235, 236, 237,
			 238, 239, 240, 241, 242, 243, 244, 245, 246, 247, 248, 249, 250, 251,
			 252, 253, 254, 255, 256, 257, 258, 259, 260, 261, 262, 263, 264, 265,
			 266, 267, 268, 269, 270, 271, 272, 273, 274, 275, 276, 277, 278, 279,
			 280, 281, 282, 283, 284, 285, 286, 287, 288, 289, 290, 291, 292, 293,
			 294, 295, 296, 297, 298, 299, 300, 301, 302, 303, 304, 305, 306, 307,
			 308, 309, 310, 311, 312, 313, 314, 315, 316, 317, 318, 319, 320, 321,
			 322, 323, 324, 325, 326, 327, 328, 329, 330, 331, 332, 333, 334, 335,
			 336, 337, 338, 339, 340, 341, 342, 343, 344, 345, 346, 347, 348, 349,
			 350, 351, 352, 353, 354, 355, 356, 357, 358, 359, 360, 361, 362, 363,
			 364, 365, 366, 367, 368, 369, 370, 371, 372, 373, 374, 375, 376, 377,
			 378, 379, 380, 381, 382, 383, 384, 385, 386, 387, 388, 389, 390, 391,
			 392, 393, 394, 395, 396, 397, 398, 399, 400, 401, 402, 403, 404, 405,
			 406, 407, 408, 409, 410, 411, 412, 413, 414, 415, 416, 417, 418, 419,
			 420, 421, 422, 423, 424, 425, 426, 427, 428, 429, 430, 431, 432, 433,
			 434, 435, 436, 437, 438, 439, 440, 441, 442, 443, 444, 445, 446, 447,
			 448, 449, 450, 451, 452, 453, 454, 455, 456, 457, 458, 459, 460, 461,
			 462, 463, 464, 465, 466, 467, 468, 469, 470, 471, 472, 473, 474, 475,
			 476, 477, 478, 479, 480, 481, 482, 483, 484, 485, 486, 487, 488, 489,
			 490, 491, 492, 493, 494, 495, 496, 497, 498, 499], dtype=torch.int32)}

Could I see how you invoked the test_layer and the stacktrace of the raised error?

Invoking the test_layer looks like this:

test_layer = dglnn.HeteroGraphConv({ 'link' : dglnn.GraphConv(1, 2, norm='right')},aggregate='mean')
test_layer(blocks[0],blocks[0].srcdata)

I’ve never had to do a stacktrace before :grimacing: but I took a swing. Is this what you need?

Traceback (most recent call last):
  File "c:/Users/22566465/Desktop/UWA_GIT/ginn/notebooks/Graph NB/testing.py", line 42, in <module>
    test_layer(blocks[0],blocks[0].srcdata)
  File "C:\Users\22566465\Anaconda3\envs\ginn\lib\site-packages\torch\nn\modules\module.py", line 1051, in _call_impl
    return forward_call(*input, **kwargs)
  File "C:\Users\22566465\Anaconda3\envs\ginn\lib\site-packages\dgl\nn\pytorch\hetero.py", line 168, in forward
    dst_inputs = {k: v[:g.number_of_dst_nodes(k)] for k, v in inputs.items()}
  File "C:\Users\22566465\Anaconda3\envs\ginn\lib\site-packages\dgl\nn\pytorch\hetero.py", line 168, in <dictcomp>
    dst_inputs = {k: v[:g.number_of_dst_nodes(k)] for k, v in inputs.items()}
  File "C:\Users\22566465\Anaconda3\envs\ginn\lib\_collections_abc.py", line 743, in __iter__
    for key in self._mapping:
  File "C:\Users\22566465\Anaconda3\envs\ginn\lib\site-packages\dgl\view.py", line 100, in __iter__
    'Current HeteroNodeDataView has multiple node types, ' \
AssertionError: Current HeteroNodeDataView has multiple node types, can not be iterated.
PS C:\Users\22566465\Desktop\UWA_GIT\ginn>

blocks[0].srcdata will be a dictionary of dictionary of tensors, whereas HeteroGraphConv takes in a dictionary of tensors. So you need something like:

test_layer(blocks[0], {'typeA': A, 'typeB': B})