def test_encode_invalid_sequence(self):
funcdef = ExpandTreeRule(
NodeType("def", NodeConstraint.Node, False),
[("name", NodeType("value", NodeConstraint.Token, True)),
("body", NodeType("expr", NodeConstraint.Node, True))])
expr = ExpandTreeRule(
NodeType("expr", NodeConstraint.Node, False),
[("op", NodeType("value", NodeConstraint.Token, False)),
("arg0", NodeType("value", NodeConstraint.Token, True)),
("arg1", NodeType("value", NodeConstraint.Token, True))])
encoder = ActionSequenceEncoder(
Samples([funcdef, expr], [
NodeType("def", NodeConstraint.Node, False),
NodeType("value", NodeConstraint.Token, True),
NodeType("expr", NodeConstraint.Node, True)
], [("", "f")]), 0)
action_sequence = ActionSequence()
action_sequence.eval(ApplyRule(funcdef))
action_sequence.eval(GenerateToken("", "f"))
action_sequence.eval(GenerateToken("", "1"))
action_sequence.eval(ApplyRule(CloseVariadicFieldRule()))
assert encoder.encode_action(action_sequence,
[Token("", "2", "2")]) is None
def test_eval(self):
entries = [Environment(
{"text_query": "foo bar", "ground_truth": "y = x + 1"},
set(["ground_truth"])
)]
dataset = ListDataset(entries)
d = get_samples(dataset, MockParser())
aencoder = ActionSequenceEncoder(d, 0)
action_sequence = GroundTruthToActionSequence(MockParser())(
"y = x + 1"
)
transform = AddActions(aencoder)
action_tensor = transform(
reference=[Token(None, "foo", "foo"), Token(None, "bar", "bar")],
action_sequence=action_sequence,
train=False
)
assert np.array_equal(
[
[2, 2, 0], [4, 3, 1], [6, 4, 2], [6, 4, 2], [5, 3, 1],
[6, 5, 5], [6, 5, 5], [5, 5, 5], [6, 4, 8], [6, 4, 8],
[5, 5, 5], [9, 6, 11], [9, 6, 11], [-1, -1, -1]
],
action_tensor.numpy()
)
def test_encode_parent(self):
funcdef = ExpandTreeRule(
NodeType("def", NodeConstraint.Node, False),
[("name", NodeType("value", NodeConstraint.Token, True)),
("body", NodeType("expr", NodeConstraint.Node, True))])
expr = ExpandTreeRule(
NodeType("expr", NodeConstraint.Node, False),
[("op", NodeType("value", NodeConstraint.Token, True)),
("arg0", NodeType("value", NodeConstraint.Token, True)),
("arg1", NodeType("value", NodeConstraint.Token, True))])
encoder = ActionSequenceEncoder(
Samples([funcdef, expr], [
NodeType("def", NodeConstraint.Node, False),
NodeType("value", NodeConstraint.Token, True),
NodeType("expr", NodeConstraint.Node, False)
], [("", "f"), ("", "2")]), 0)
action_sequence = ActionSequence()
action_sequence.eval(ApplyRule(funcdef))
action_sequence.eval(GenerateToken("", "f"))
action_sequence.eval(GenerateToken("", "1"))
action_sequence.eval(GenerateToken("", "2"))
action_sequence.eval(ApplyRule(CloseVariadicFieldRule()))
parent = encoder.encode_parent(action_sequence)
assert np.array_equal([[-1, -1, -1, -1], [1, 2, 0, 0], [1, 2, 0, 0],
[1, 2, 0, 0], [1, 2, 0, 0], [1, 2, 0, 1]],
parent.numpy())
def test_encode_empty_sequence(self):
funcdef = ExpandTreeRule(
NodeType("def", NodeConstraint.Node, False),
[("name", NodeType("value", NodeConstraint.Token, False)),
("body", NodeType("expr", NodeConstraint.Node, True))])
expr = ExpandTreeRule(
NodeType("expr", NodeConstraint.Node, False),
[("op", NodeType("value", NodeConstraint.Token, False)),
("arg0", NodeType("value", NodeConstraint.Token, False)),
("arg1", NodeType("value", NodeConstraint.Token, False))])
encoder = ActionSequenceEncoder(
Samples([funcdef, expr], [
NodeType("def", NodeConstraint.Node, False),
NodeType("value", NodeConstraint.Token, False),
NodeType("expr", NodeConstraint.Node, False)
], [("", "f")]), 0)
action_sequence = ActionSequence()
action = encoder.encode_action(action_sequence, [Token("", "1", "1")])
parent = encoder.encode_parent(action_sequence)
d, m = encoder.encode_tree(action_sequence)
assert np.array_equal([[-1, -1, -1, -1]], action.numpy())
assert np.array_equal([[-1, -1, -1, -1]], parent.numpy())
assert np.array_equal(np.zeros((0, )), d.numpy())
assert np.array_equal(np.zeros((0, 0)), m.numpy())
def test_eval(self):
entries = [Environment(
{"text_query": "ab test", "ground_truth": "y = x + 1"},
set(["ground_truth"])
)]
dataset = ListDataset(entries)
d = get_samples(dataset, MockParserWithoutVariadicArgs())
aencoder = ActionSequenceEncoder(d, 0)
action_sequence = GroundTruthToActionSequence(MockParserWithoutVariadicArgs())(
"y = x + 1"
)
transform = AddQueryForTreeGenDecoder(aencoder, 3,)
query = transform(
reference=[Token(None, "ab", "ab"), Token(None, "test", "test")],
action_sequence=action_sequence,
train=False
)
assert np.array_equal(
[
[-1, -1, -1], [2, -1, -1], [3, 2, -1], [4, 3, 2],
[3, 2, -1], [5, 3, 2], [5, 3, 2], [4, 5, 3],
[5, 3, 2], [6, 5, 3]
],
query.numpy()
)
def test_simple_case(self):
entries = [Environment(
{"ground_truth": "y = x + 1"},
set(["ground_truth"])
)]
dataset = ListDataset(entries)
d = get_samples(dataset, MockParser())
aencoder = ActionSequenceEncoder(d, 0)
action_sequence = GroundTruthToActionSequence(MockParser())(
ground_truth="y = x + 1"
)
transform = EncodeActionSequence(aencoder)
ground_truth = transform(
action_sequence=action_sequence,
reference=[Token(None, "foo", "foo"), Token(None, "bar", "bar")],
)
assert np.array_equal(
[
[3, -1, -1], [4, -1, -1], [-1, 1, -1], [1, -1, -1],
[5, -1, -1], [-1, 2, -1], [1, -1, -1], [4, -1, -1],
[-1, 3, -1], [1, -1, -1], [6, -1, -1], [-1, 4, -1],
[1, -1, -1]
],
ground_truth.numpy()
)
def test_eval(self):
entries = [Environment(
{"text_query": "ab test", "ground_truth": "y = x + 1"},
set(["ground_truth"])
)]
dataset = ListDataset(entries)
d = get_samples(dataset, MockParserWithoutVariadicArgs())
aencoder = ActionSequenceEncoder(d, 0)
action_sequence = GroundTruthToActionSequence(MockParserWithoutVariadicArgs())(
"y = x + 1"
)
transform = AddActionSequenceAsTree(aencoder,)
matrix, depth = transform(
reference=[Token(None, "ab", "ab"), Token(None, "test", "test")],
action_sequence=action_sequence,
train=False
)
assert np.array_equal(
[0, 1, 2, 3, 2, 3, 3, 4, 3, 4],
depth.numpy()
)
assert np.array_equal(
[[0, 1, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 1, 0, 0, 0, 0, 0],
[0, 0, 0, 1, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 1, 1, 0, 1, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 1, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 1],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],
matrix.numpy()
)
def test_n_dependent(self):
entries = [Environment(
{"text_query": "ab test", "ground_truth": "y = x + 1"},
set(["ground_truth"])
)]
dataset = ListDataset(entries)
d = get_samples(dataset, MockParserWithoutVariadicArgs())
aencoder = ActionSequenceEncoder(d, 0)
action_sequence = GroundTruthToActionSequence(MockParserWithoutVariadicArgs())(
"y = x + 1"
)
transform = AddPreviousActionRules(aencoder, 2, n_dependent=3)
prev_rule_action = transform(
reference=[Token(None, "ab", "ab"), Token(None, "test", "test")],
action_sequence=action_sequence,
train=False,
)
assert np.array_equal(
[
# str -> "y"
[[-1, -1, -1], [-1, 3, -1], [-1, -1, -1]],
# Number -> number
[[8, -1, -1], [9, -1, -1], [-1, -1, -1]],
[[-1, -1, -1], [-1, 4, -1], [-1, -1, -1]],
],
prev_rule_action.numpy()
)
def test_encode_tree(self):
funcdef = ExpandTreeRule(
NodeType("def", NodeConstraint.Node, False),
[("name", NodeType("value", NodeConstraint.Token, True)),
("body", NodeType("expr", NodeConstraint.Node, True))])
expr = ExpandTreeRule(
NodeType("expr", NodeConstraint.Node, False),
[("op", NodeType("value", NodeConstraint.Token, True)),
("arg0", NodeType("value", NodeConstraint.Token, True)),
("arg1", NodeType("value", NodeConstraint.Token, True))])
encoder = ActionSequenceEncoder(
Samples([funcdef, expr], [
NodeType("def", NodeConstraint.Node, False),
NodeType("value", NodeConstraint.Token, True),
NodeType("expr", NodeConstraint.Node, False)
], [("", "f"), ("", "2")]), 0)
action_sequence = ActionSequence()
action_sequence.eval(ApplyRule(funcdef))
action_sequence.eval(GenerateToken("", "f"))
action_sequence.eval(GenerateToken("", "1"))
d, m = encoder.encode_tree(action_sequence)
assert np.array_equal([0, 1, 1], d.numpy())
assert np.array_equal([[0, 1, 1], [0, 0, 0], [0, 0, 0]], m.numpy())
def test_decode(self):
funcdef = ExpandTreeRule(
NodeType("def", NodeConstraint.Node, False),
[("name", NodeType("value", NodeConstraint.Token, True)),
("body", NodeType("expr", NodeConstraint.Node, True))])
expr = ExpandTreeRule(
NodeType("expr", NodeConstraint.Node, False),
[("op", NodeType("value", NodeConstraint.Token, True)),
("arg0", NodeType("value", NodeConstraint.Token, True)),
("arg1", NodeType("value", NodeConstraint.Token, True))])
encoder = ActionSequenceEncoder(
Samples([funcdef, expr], [
NodeType("def", NodeConstraint.Node, False),
NodeType("value", NodeConstraint.Token, True),
NodeType("expr", NodeConstraint.Node, False)
], [("", "f")]), 0)
action_sequence = ActionSequence()
action_sequence.eval(ApplyRule(funcdef))
action_sequence.eval(GenerateToken("", "f"))
action_sequence.eval(GenerateToken("", "1"))
action_sequence.eval(ApplyRule(CloseVariadicFieldRule()))
expected_action_sequence = ActionSequence()
expected_action_sequence.eval(ApplyRule(funcdef))
expected_action_sequence.eval(GenerateToken("", "f"))
expected_action_sequence.eval(GenerateToken("", "1"))
expected_action_sequence.eval(ApplyRule(CloseVariadicFieldRule()))
result = encoder.decode(
encoder.encode_action(action_sequence,
[Token(None, "1", "1")])[:-1, 1:],
[Token(None, "1", "1")])
assert \
expected_action_sequence.action_sequence == result.action_sequence
def test_eval(self):
entries = [Environment(
{"ground_truth": "y = x + 1"},
set(["ground_truth"])
)]
dataset = ListDataset(entries)
d = get_samples(dataset, MockParser())
aencoder = ActionSequenceEncoder(d, 0)
action_sequence = GroundTruthToActionSequence(MockParser())(
"y = x + 1"
)
transform = AddPreviousActions(aencoder)
prev_action_tensor = transform(
reference=[Token(None, "foo", "foo"), Token(None, "bar", "bar")],
action_sequence=action_sequence,
train=False
)
assert np.array_equal(
[
[2, -1, -1], [3, -1, -1], [4, -1, -1], [-1, 1, -1],
[1, -1, -1], [5, -1, -1], [-1, 2, -1], [1, -1, -1],
[4, -1, -1], [-1, 3, -1], [1, -1, -1], [6, -1, -1],
[-1, 4, -1], [1, -1, -1]
],
prev_action_tensor.numpy()
)
def test_encode_path(self):
funcdef = ExpandTreeRule(
NodeType("def", NodeConstraint.Node, False),
[("name", NodeType("value", NodeConstraint.Token, True)),
("body", NodeType("expr", NodeConstraint.Node, True))])
expr = ExpandTreeRule(
NodeType("expr", NodeConstraint.Node, False),
[("constant", NodeType("value", NodeConstraint.Token, True))])
encoder = ActionSequenceEncoder(
Samples([funcdef, expr], [
NodeType("def", NodeConstraint.Node, False),
NodeType("value", NodeConstraint.Token, True),
NodeType("expr", NodeConstraint.Node, True)
], [("", "f"), ("", "2")]), 0)
action_sequence = ActionSequence()
action_sequence.eval(ApplyRule(funcdef))
action_sequence.eval(GenerateToken("", "f"))
action_sequence.eval(GenerateToken("", "1"))
action_sequence.eval(GenerateToken("", "2"))
action_sequence.eval(ApplyRule(CloseVariadicFieldRule()))
action_sequence.eval(ApplyRule(expr))
action_sequence.eval(GenerateToken("", "f"))
action_sequence.eval(ApplyRule(CloseVariadicFieldRule()))
action_sequence.eval(ApplyRule(CloseVariadicFieldRule()))
path = encoder.encode_path(action_sequence, 2)
assert np.array_equal(
np.array(
[
[-1, -1], # funcdef
[2, -1], # f
[2, -1], # 1
[2, -1], # 2
[2, -1], # CloseVariadicField
[2, -1], # expr
[3, 2], # f
[3, 2], # CloseVariadicField
[2, -1], # CloseVariadicField
],
dtype=np.long),
path.numpy())
path = encoder.encode_path(action_sequence, 1)
assert np.array_equal(
np.array(
[
[-1], # funcdef
[2], # f
[2], # 1
[2], # 2
[2], # CloseVariadicField
[2], # expr
[3], # f
[3], # CloseVariadicField
[2], # CloseVariadicField
],
dtype=np.long),
path.numpy())
def prepare_encoder(self, dataset, parser):
words = get_words(dataset, tokenize)
chars = get_characters(dataset, tokenize)
samples = get_samples(dataset, parser)
qencoder = LabelEncoder(words, 2)
cencoder = LabelEncoder(chars, 0)
aencoder = ActionSequenceEncoder(samples, 2)
return qencoder, cencoder, aencoder
def test_encode_completed_sequence(self):
none = ExpandTreeRule(NodeType("value", NodeConstraint.Node, False),
[])
encoder = ActionSequenceEncoder(
Samples([none], [NodeType("value", NodeConstraint.Node, False)],
[("", "f")]), 0)
action_sequence = ActionSequence()
action_sequence.eval(ApplyRule(none))
action = encoder.encode_action(action_sequence, [Token("", "1", "1")])
parent = encoder.encode_parent(action_sequence)
assert np.array_equal([[-1, 2, -1, -1], [-1, -1, -1, -1]],
action.numpy())
assert np.array_equal([[-1, -1, -1, -1], [-1, -1, -1, -1]],
parent.numpy())
def test_impossible_case(self):
entries = [Environment(
{"ground_truth": "y = x + 1"},
set(["ground_truth"])
)]
dataset = ListDataset(entries)
d = get_samples(dataset, MockParser())
d.tokens = [("", "y"), ("", "1")]
aencoder = ActionSequenceEncoder(d, 0)
action_sequence = GroundTruthToActionSequence(MockParser())(
ground_truth="y = x + 1"
)
transform = EncodeActionSequence(aencoder)
with pytest.raises(RuntimeError):
transform(
reference=[Token(None, "foo", "foo"), Token(None, "bar", "bar")],
action_sequence=action_sequence,
)
def test_decode_invalid_tensor(self):
funcdef = ExpandTreeRule(
NodeType("def", NodeConstraint.Node, False),
[("name", NodeType("value", NodeConstraint.Token, False)),
("body", NodeType("expr", NodeConstraint.Node, True))])
expr = ExpandTreeRule(
NodeType("expr", NodeConstraint.Node, False),
[("op", NodeType("value", NodeConstraint.Token, False)),
("arg0", NodeType("value", NodeConstraint.Token, False)),
("arg1", NodeType("value", NodeConstraint.Token, False))])
encoder = ActionSequenceEncoder(
Samples([funcdef, expr], [
NodeType("def", NodeConstraint.Node, False),
NodeType("value", NodeConstraint.Token, False),
NodeType("expr", NodeConstraint.Node, False)
], [("", "f")]), 0)
assert encoder.decode(torch.LongTensor([[-1, -1, -1]]), []) is None
assert encoder.decode(torch.LongTensor([[-1, -1, 1]]), []) is None
def test_eval(self):
entries = [Environment(
{"text_query": "ab test", "ground_truth": "y = x + 1"},
set(["ground_truth"])
)]
dataset = ListDataset(entries)
d = get_samples(dataset, MockParserWithoutVariadicArgs())
aencoder = ActionSequenceEncoder(d, 0)
action_sequence = GroundTruthToActionSequence(MockParserWithoutVariadicArgs())(
"y = x + 1"
)
transform = AddPreviousActionRules(aencoder, 2)
prev_rule_action = transform(
reference=[Token(None, "ab", "ab"), Token(None, "test", "test")],
action_sequence=action_sequence,
train=False
)
assert np.array_equal(
[
# None -> Root
[[1, -1, -1], [2, -1, -1], [-1, -1, -1]],
# Assign -> Name, expr
[[3, -1, -1], [4, -1, -1], [5, -1, -1]],
# Name -> str
[[4, -1, -1], [6, -1, -1], [-1, -1, -1]],
# str -> "x"
[[-1, -1, -1], [-1, 1, -1], [-1, -1, -1]],
# Op -> str, expr, expr
[[7, -1, -1], [6, -1, -1], [5, -1, -1]],
# str -> "+"
[[-1, -1, -1], [-1, 2, -1], [-1, -1, -1]],
# Name -> str
[[4, -1, -1], [6, -1, -1], [-1, -1, -1]],
# str -> "y"
[[-1, -1, -1], [-1, 3, -1], [-1, -1, -1]],
# Number -> number
[[8, -1, -1], [9, -1, -1], [-1, -1, -1]],
[[-1, -1, -1], [-1, 4, -1], [-1, -1, -1]],
],
prev_rule_action.numpy()
)
def prepare_encoder(self, dataset, parser):
return ActionSequenceEncoder(get_samples(dataset, parser, reference=True),
0)
def create_encoder():
return ActionSequenceEncoder(Samples(
[Root2X, Root2Y, X2Y_list, Ysub2Str],
[R, X, Y, Ysub, Y_list, Str],
[("Str", "x"),
("Int", "1")]), 0)
def test_reserved_labels(self):
encoder = ActionSequenceEncoder(Samples([], [], []), 0)
assert 2 == len(encoder._rule_encoder.vocab)
assert 1 == len(encoder._token_encoder.vocab)
def test_encode_raw_value(self):
encoder = ActionSequenceEncoder(
Samples([], [], [("", "foo"), ("x", "foo")]), 0)
assert [1, 2] == encoder.encode_raw_value("foo")
assert [0] == encoder.encode_raw_value("bar")