def setUp(self):
actions_counter = Counter()
for action in [
"IN:A",
"IN:B",
"IN:UNSUPPORTED",
"REDUCE",
"SHIFT",
"SL:C",
"SL:D",
]:
actions_counter[action] += 1
actions_vocab = Vocab(actions_counter, specials=[])
self.parser = RNNGParser(
ablation=RNNGParser.Config.AblationParams(),
constraints=RNNGParser.Config.RNNGConstraints(),
lstm_num_layers=2,
lstm_dim=20,
max_open_NT=10,
dropout=0.2,
beam_size=3,
top_k=3,
actions_vocab=actions_vocab,
shift_idx=4,
reduce_idx=3,
ignore_subNTs_roots=[2],
valid_NT_idxs=[0, 1, 2, 5, 6],
valid_IN_idxs=[0, 1, 2],
valid_SL_idxs=[5, 6],
embedding=EmbeddingList(
embeddings=[
WordEmbedding(
num_embeddings=5,
embedding_dim=20,
embeddings_weight=None,
init_range=[-1, 1],
unk_token_idx=4,
mlp_layer_dims=[],
),
DictEmbedding(
num_embeddings=4, embed_dim=10, pooling_type=PoolingType.MEAN
),
],
concat=True,
),
p_compositional=CompositionalNN(lstm_dim=20),
)
self.parser.train()
class Config(Task.Config):
model: RNNGParser.Config = RNNGParser.Config()
trainer: HogwildTrainer.Config = HogwildTrainer.Config()
data_handler: CompositionalDataHandler.Config = CompositionalDataHandler.Config(
)
labels: Optional[WordLabelConfig] = None
metric_reporter: CompositionalMetricReporter.Config = CompositionalMetricReporter.Config(
)
class Config(NewTask.Config):
model: RNNGParser.Config = RNNGParser.Config()
trainer: HogwildTrainer.Config = HogwildTrainer.Config()
metric_reporter: CompositionalMetricReporter.Config = (
CompositionalMetricReporter.Config())
class RNNGParserTest(unittest.TestCase):
def setUp(self):
actions_counter = Counter()
for action in [
"IN:A",
"IN:B",
"IN:UNSUPPORTED",
"REDUCE",
"SHIFT",
"SL:C",
"SL:D",
]:
actions_counter[action] += 1
actions_vocab = Vocab(actions_counter, specials=[])
self.parser = RNNGParser(
ablation=RNNGParser.Config.AblationParams(),
constraints=RNNGParser.Config.RNNGConstraints(),
lstm_num_layers=2,
lstm_dim=20,
max_open_NT=10,
dropout=0.2,
beam_size=3,
top_k=3,
actions_vocab=actions_vocab,
shift_idx=4,
reduce_idx=3,
ignore_subNTs_roots=[2],
valid_NT_idxs=[0, 1, 2, 5, 6],
valid_IN_idxs=[0, 1, 2],
valid_SL_idxs=[5, 6],
embedding=EmbeddingList(
embeddings=[
WordEmbedding(
num_embeddings=5,
embedding_dim=20,
embeddings_weight=None,
init_range=[-1, 1],
unk_token_idx=4,
mlp_layer_dims=[],
),
DictEmbedding(
num_embeddings=4, embed_dim=10, pooling_type=PoolingType.MEAN
),
],
concat=True,
),
p_compositional=CompositionalNN(lstm_dim=20),
)
self.parser.train()
def populate_buffer(self):
state = ParserState(self.parser)
for _ in range(2):
state.buffer_stackrnn.push(torch.zeros(1, 30), Element("Token"))
return state
def check_valid_actions(self, state, actions):
self.assertSetEqual(set(self.parser.valid_actions(state)), set(actions))
def test_valid_actions_unconstrained(self):
self.parser.constraints_intent_slot_nesting = False
self.parser.constraints_no_slots_inside_unsupported = False
state = self.populate_buffer()
# Valid Actions at beginning: all nonterminals
self.check_valid_actions(state, self.parser.valid_NT_idxs)
# After pushing IN:A: all nonterminals, SHIFT
self.parser.push_action(state, self.parser.actions_vocab.stoi["IN:A"])
self.check_valid_actions(
state, self.parser.valid_NT_idxs + [self.parser.shift_idx]
)
# After pushing SL:C and SHIFT: all nonterminals, SHIFT, REDUCE
self.parser.push_action(state, self.parser.actions_vocab.stoi["SL:C"])
self.parser.push_action(state, self.parser.actions_vocab.stoi["SHIFT"])
self.check_valid_actions(
state,
self.parser.valid_NT_idxs
+ [self.parser.shift_idx]
+ [self.parser.reduce_idx],
)
# After all SHIFTs: only REDUCE
self.parser.push_action(state, self.parser.actions_vocab.stoi["SHIFT"])
self.check_valid_actions(state, [self.parser.reduce_idx])
# After all REDUCEs, no valid actions
self.parser.push_action(state, self.parser.actions_vocab.stoi["REDUCE"])
self.parser.push_action(state, self.parser.actions_vocab.stoi["REDUCE"])
self.check_valid_actions(state, [])
def test_valid_actions_constraint_insl(self):
self.parser.constraints_intent_slot_nesting = True
self.parser.constraints_no_slots_inside_unsupported = False
state = self.populate_buffer()
# Valid Actions at beginning: all intents
self.check_valid_actions(state, self.parser.valid_IN_idxs)
# After pushing IN:A: all slots, SHIFT
self.parser.push_action(state, self.parser.actions_vocab.stoi["IN:A"])
self.check_valid_actions(
state, self.parser.valid_SL_idxs + [self.parser.shift_idx]
)
# After pushing SL:C: all intents, SHIFT
self.parser.push_action(state, self.parser.actions_vocab.stoi["SL:C"])
self.check_valid_actions(
state, self.parser.valid_IN_idxs + [self.parser.shift_idx]
)
# After all SHIFTs: only REDUCE
self.parser.push_action(state, self.parser.actions_vocab.stoi["SHIFT"])
self.parser.push_action(state, self.parser.actions_vocab.stoi["SHIFT"])
self.check_valid_actions(state, [self.parser.reduce_idx])
# After all REDUCEs, no valid actions
self.parser.push_action(state, self.parser.actions_vocab.stoi["REDUCE"])
self.parser.push_action(state, self.parser.actions_vocab.stoi["REDUCE"])
self.check_valid_actions(state, [])
def test_valid_actions_constraint_unsupported(self):
self.parser.constraints_intent_slot_nesting = True
self.parser.constraints_no_slots_inside_unsupported = True
state = self.populate_buffer()
# Valid Actions at beginning: all intents
self.check_valid_actions(state, self.parser.valid_IN_idxs)
# Needed to make test logic work
state.predicted_actions_idx.append(
self.parser.actions_vocab.stoi["IN:UNSUPPORTED"]
)
# After pushing IN:UNSUPPORTED: SHIFT
self.parser.push_action(state, self.parser.actions_vocab.stoi["IN:UNSUPPORTED"])
self.check_valid_actions(state, [self.parser.shift_idx])
def test_forward_shapes(self):
self.parser.eval(Stage.EVAL)
tokens = torch.tensor([[0, 1, 2, 3]])
seq_lens = torch.tensor([tokens.shape[1]])
dict_feat = (
torch.tensor([[1, 1, 1, 1, 1, 1, 3, 1]]),
torch.tensor([[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0]]),
torch.tensor([1, 1, 2, 1]),
)
actions, scores = self.parser(
tokens=tokens, seq_lens=seq_lens, dict_feat=dict_feat
)[0]
self.assertGreater(actions.shape[1], tokens.shape[1])
self.assertEqual(actions.shape[0:2], scores.shape[0:2])
self.assertEqual(scores.shape[2], len(self.parser.actions_vocab.itos))
# Beam Search Test
self.parser.eval(Stage.TEST)
results = self.parser(tokens=tokens, seq_lens=seq_lens, dict_feat=dict_feat)
self.assertEqual(len(results), 3)
for actions, scores in results:
self.assertGreater(actions.shape[1], tokens.shape[1])
self.assertEqual(actions.shape[0:2], scores.shape[0:2])
self.assertEqual(scores.shape[2], len(self.parser.actions_vocab.itos))
