def setUp(self) -> None:
self.batch_size = 8
self.input_size = 100
self.n_node_properties = 20
self.model = NSMCell(self.input_size, self.n_node_properties)
self.graph_batch = collate_graphs(
list(
islice(
infinite_graphs(
self.input_size,
self.n_node_properties,
node_distribution=(16.4, 8.2),
density_distribution=(0.2, 0.4),
),
self.batch_size,
)))
self.instruction_batch = torch.rand(self.batch_size, self.input_size)
self.distribution = (
1.0 /
self.graph_batch.nodes_per_graph)[self.graph_batch.node_indices]
all_prop_similarities = F.softmax(
self.instruction_batch @ torch.rand(self.input_size,
self.n_node_properties + 1),
dim=1,
)
self.node_prop_similarities = all_prop_similarities[:, :-1]
self.relation_similarity = all_prop_similarities[:, -1]
def setUp(self) -> None:
graph_gen = infinite_graphs(
300,
77,
node_distribution=(16.4, 8.2),
density_distribution=(0.2, 0.4),
)
self.batch = collate_graphs(list(islice(graph_gen, 8)))
def collate_fn(batch):
graphs, questions, targets, instructionss = zip(*batch)
return (
collate_graphs(graphs),
torch.nn.utils.rnn.pack_sequence(questions,
enforce_sorted=False),
vocab.concept_embeddings,
vocab.property_embeddings,
torch.tensor(targets),
torch.stack(instructionss),
)
def test_collate_graphs_inverse(self) -> None:
batch_lens = random.sample(range(1, 21), 7)
gen = infinite_graphs()
for b_len in batch_lens:
original_graphs = list(islice(gen, b_len))
reconstructed_graphs = split_batch(collate_graphs(original_graphs))
self.assertEqual(len(original_graphs), len(reconstructed_graphs))
for og, ng in zip(original_graphs, reconstructed_graphs):
for key in og._fields:
with self.subTest(batch_len=b_len, graph_attr=key):
self.assertTrue(
(getattr(og, key) == getattr(ng, key)).all())
def collate_fn(batch):
graphs, questions, targets = zip(*batch)
# return a 6-tuple, thats what the last iteration of
# NSMLightningModule expects
return (
collate_graphs(graphs),
torch.nn.utils.rnn.pack_sequence(questions,
enforce_sorted=False),
vocab.concept_embeddings,
vocab.property_embeddings,
torch.tensor(targets),
# dummy tensor, gold_instructions not used
torch.empty(1),
)
def test_edge_indices_shift_edge_attrs(self) -> None:
graphs = list(islice(self.graph_gen, 8))
batch = collate_graphs(graphs)
for _ in range(5):
# Get 5 random nodes and test that their adjacency list match
g_idx = random.randrange(len(graphs))
graph = graphs[g_idx]
node_idx = random.randrange(graph.node_attrs.size(0))
# breakpoint()
correct_edge_attrs = graph.edge_attrs[graph.edge_indices[0] ==
node_idx]
new_node_idx = sum(
batch.nodes_per_graph.tolist()[:g_idx]) + node_idx
new_edge_attrs = batch.edge_attrs[batch.edge_indices[0] ==
new_node_idx]
# breakpoint()
self.assertEqual(new_edge_attrs.size(), correct_edge_attrs.size())
self.assertTrue(new_edge_attrs.eq(correct_edge_attrs).all())
def test_output_shapes(self) -> None:
batch_size = 8
graphs = list(islice(self.graph_gen, batch_size))
batch = collate_graphs(graphs)
with self.subTest("node_attrs"):
total_nodes = sum(g.node_attrs.size(0) for g in graphs)
self.assertEqual(
batch.node_attrs.size(),
(total_nodes, self.n_properties, self.hidden_size),
)
with self.subTest("edges"):
total_edges = sum(g.edge_attrs.size(0) for g in graphs)
with self.subTest("edge_attrs"):
self.assertEqual(batch.edge_attrs.size(),
(total_edges, self.hidden_size))
with self.subTest("edge_indices"):
self.assertEqual(batch.edge_indices.size(), (2, total_edges))
with self.subTest("nodes_per_graph"):
self.assertEqual(batch.nodes_per_graph.size(0), batch_size)
def setUp(self) -> None:
# These tests should be using expected dimensions
self.batch_size = 64
self.output_size = 2000
n_node_properties = 78
input_size = 300
computation_steps = 8
self.vocab = torch.rand(1335, input_size)
self.prop_embeds = torch.rand(n_node_properties + 1, input_size)
self.model = NSM(
input_size,
n_node_properties,
computation_steps,
self.output_size,
)
self.graph_batch = collate_graphs(
list(
islice(
infinite_graphs(
input_size,
n_node_properties,
node_distribution=(16.4, 8.2),
density_distribution=(0.2, 0.4),
),
self.batch_size,
)))
self.question_batch = pack_sequence(
sorted(
[
torch.rand(random.randint(10, 20), input_size)
for _ in range(self.batch_size)
],
key=len,
reverse=True,
))
def test_nodes_per_graph(self) -> None:
graphs = list(islice(self.graph_gen, 8))
nodes_per_graph = [g.node_attrs.size(0) for g in graphs]
batch = collate_graphs(graphs)
self.assertEqual(batch.nodes_per_graph.tolist(), nodes_per_graph)
self.assertEqual(batch.node_attrs.size(0), sum(nodes_per_graph))
def test_output_type(self) -> None:
batch = collate_graphs(list(islice(self.graph_gen, 8)))
self.assertIsInstance(batch, Batch)