def test_instance_implements_mutable_mapping(self):
words_field = TextField([Token("hello")], {})
label_field = LabelField(1, skip_indexing=True)
instance = Instance({"words": words_field, "labels": label_field})
assert instance["words"] == words_field
assert instance["labels"] == label_field
assert len(instance) == 2
keys = {k for k, v in instance.items()}
assert keys == {"words", "labels"}
values = [v for k, v in instance.items()]
assert words_field in values
assert label_field in values
def text_to_instance(self,
comment_text: str,
sentiment: int = None) -> Instance:
tokens = self._tokenizer.tokenize(comment_text)
if self._max_seq_len is not None:
tokens = tokens[:self._max_seq_len]
sentence_field = TextField(tokens, self._token_indexers)
fields = {"tokens": sentence_field}
label_field = LabelField(str(sentiment))
fields["label"] = label_field
return Instance(fields)
def text_to_instance(self,
tokens: List[Token],
id: str = None,
labels: np.ndarray = None) -> Instance:
sentence_field = TextField(tokens, self.token_indexers)
fields = {"tokens": sentence_field}
id_field = MetadataField(id)
fields["id"] = id_field
if labels is None:
labels = np.zeros(len(label_cols))
label_field = ArrayField(array=labels)
fields["label"] = label_field
return Instance(fields)
def text_to_instance(self,
text: str,
hypothesis: str,
label: str = None) -> Instance:
fields = {}
# to make it look like two instances [cls] sent1 [sep] & [cls] sent2 [sep]
tokens1 = self._tokenizer.tokenize(text)
tokens2 = self._tokenizer.tokenize(hypothesis)
fields["tokens1"] = TextField(tokens1, self._token_indexers)
fields["tokens2"] = TextField(tokens2, self._token_indexers)
if label is not None:
fields["labels"] = LabelField(label)
return Instance(fields)
def _sentences_to_ids(self, sentences):
indexer = ELMoTokenCharactersIndexer()
# For each sentence, first create a TextField, then create an instance
instances = []
for sentence in sentences:
tokens = [Token(token) for token in sentence]
field = TextField(tokens, {'character_ids': indexer})
instance = Instance({'elmo': field})
instances.append(instance)
dataset = Batch(instances)
vocab = Vocabulary()
dataset.index_instances(vocab)
return dataset.as_tensor_dict()['elmo']['character_ids']
def text_to_instance(self, source_tokens: List[Token],
target_tokens: List[Token]) -> Optional[Instance]:
source_tokens.insert(0, Token(START_SYMBOL))
source_tokens.append(Token(END_SYMBOL))
target_tokens.insert(0, Token(START_SYMBOL))
target_tokens.append(Token(END_SYMBOL))
fields = {
'source_tokens': TextField(source_tokens, self.token_indexers),
'target_tokens': TextField(target_tokens,
self.target_token_indexers),
}
return Instance(fields)
def test_registrability(self):
@Vocabulary.register('my-vocabulary')
class MyVocabulary:
@classmethod
def from_params(cls, params, instances=None):
# pylint: disable=unused-argument
return MyVocabulary()
params = Params({'type': 'my-vocabulary'})
instance = Instance(fields={})
vocab = Vocabulary.from_params(params=params, instances=[instance])
assert isinstance(vocab, MyVocabulary)
def predict_batch(self, texts):
instances = []
for text in texts:
tokens = self._tokenizer.tokenize(text)
instance = Instance(
{'tokens': TextField(tokens, self._token_indexers)})
instances.append(instance)
result = self.model.forward_on_instances(instances)
results = []
for instance_result in result:
results.append(self._format_instance_result(instance_result))
return results
def toInstance(self,
names: List[str],
categories: List[str] = None) -> Instance:
token_field = TextField([Token(nm) for nm in names],
self.token_indexers)
fields = {"tokens": token_field}
fields["token_characters"] = TextField([Token(nm) for nm in names],
self.char_indexers)
if categories:
fields["labels"] = SequenceLabelField(labels=categories,
sequence_field=token_field)
return Instance(fields)
def test_registrability(self):
@Vocabulary.register("my-vocabulary")
class MyVocabulary:
@classmethod
def from_params(cls, params, instances=None):
return MyVocabulary()
params = Params({"type": "my-vocabulary"})
instance = Instance(fields={})
vocab = Vocabulary.from_params(params=params, instances=[instance])
assert isinstance(vocab, MyVocabulary)
def get_dataset(self):
field1 = TextField(
[Token(t) for t in ["this", "is", "a", "sentence", "."]],
self.token_indexer)
field2 = TextField([
Token(t)
for t in ["this", "is", "a", "different", "sentence", "."]
], self.token_indexer)
field3 = TextField(
[Token(t) for t in ["here", "is", "a", "sentence", "."]],
self.token_indexer)
field4 = TextField([Token(t) for t in ["this", "is", "short"]],
self.token_indexer)
instances = [
Instance({
"text1": field1,
"text2": field2
}),
Instance({
"text1": field3,
"text2": field4
})
]
return Dataset(instances)
def text_to_instance(self, src_seq: Iterable[Token], tgt_seq: str, salience_seq: Iterable[float]) -> Instance:
indexer = SingleIdTokenIndexer(lowercase_tokens=True)
tokenized_src = src_seq[:self._source_max_tokens]
tokenized_tgt = self._tokenizer.tokenize(tgt_seq)[:self._target_max_tokens]
source_field = TextField(tokenized_src, {'tokens': indexer})
target_field = TextField(tokenized_tgt, {'tokens': indexer})
if self._interpolation:
saliency_field = ArrayField(np.array(self.smooth_and_norm(salience_seq)[:self._source_max_tokens]))
else:
saliency_field = ArrayField(np.array(salience_seq[:self._source_max_tokens]))
return Instance({
'source_tokens': source_field,
'target_tokens': target_field,
'salience_values': saliency_field
})
def text_to_instance(self, string: str,
lang2_lang: str) -> Instance: # type: ignore
# pylint: disable=arguments-differ
"""
Used in predicition time
"""
lang_pair = self._undefined_lang_id + '-' + lang2_lang
tokenized_string = self._tokenizer.tokenize(string)
string_field = TextField(tokenized_string, self._token_indexers)
return Instance({
self._mingler.dataset_name_field:
MetadataField(lang_pair),
'lang1_tokens':
string_field
})
def test_elmo_bilm(self):
# get the raw data
sentences, expected_lm_embeddings = self._load_sentences_embeddings()
# load the test model
elmo_bilm = _ElmoBiLm(self.options_file, self.weight_file)
# Deal with the data.
indexer = ELMoTokenCharactersIndexer()
# For each sentence, first create a TextField, then create an instance
instances = []
for batch in zip(*sentences):
for sentence in batch:
tokens = [Token(token) for token in sentence.split()]
field = TextField(tokens, {"character_ids": indexer})
instance = Instance({"elmo": field})
instances.append(instance)
vocab = Vocabulary()
dataset = AllennlpDataset(instances, vocab)
# Now finally we can iterate through batches.
loader = DataLoader(dataset, 3)
for i, batch in enumerate(loader):
lm_embeddings = elmo_bilm(
batch["elmo"]["character_ids"]["elmo_tokens"])
top_layer_embeddings, mask = remove_sentence_boundaries(
lm_embeddings["activations"][2], lm_embeddings["mask"])
# check the mask lengths
lengths = mask.data.numpy().sum(axis=1)
batch_sentences = [sentences[k][i] for k in range(3)]
expected_lengths = [
len(sentence.split()) for sentence in batch_sentences
]
self.assertEqual(lengths.tolist(), expected_lengths)
# get the expected embeddings and compare!
expected_top_layer = [
expected_lm_embeddings[k][i] for k in range(3)
]
for k in range(3):
self.assertTrue(
numpy.allclose(
top_layer_embeddings[k, :lengths[k], :].data.numpy(),
expected_top_layer[k],
atol=1.0e-6,
))
def text_to_instance(
self, # type: ignore
qid: str,
start: str,
alternatives: List[str],
label: Optional[int] = None,
) -> Instance:
# tokenize
start = self._tokenizer.tokenize(start)
sequences = []
for alternative in alternatives:
alternative = self._tokenizer.tokenize(alternative)
length_for_start = (self.length_limit - len(alternative) -
self._tokenizer.num_special_tokens_for_pair())
if length_for_start < 0:
# If the alternative is too long by itself, we take the beginning and add no tokens from the start.
alternative = alternative[:length_for_start]
length_for_start = 0
sequences.append(
self._tokenizer.add_special_tokens(start[:length_for_start],
alternative))
# make fields
from allennlp.data.fields import TextField
sequences = [
TextField(sequence, self._token_indexers) for sequence in sequences
]
from allennlp.data.fields import ListField
sequences = ListField(sequences)
from allennlp.data.fields import MetadataField
fields = {
"alternatives": sequences,
"qid": MetadataField(qid),
}
if label is not None:
if label < 0 or label >= len(sequences):
raise ValueError("Alternative %d does not exist", label)
from allennlp.data.fields import IndexField
fields["correct_alternative"] = IndexField(label, sequences)
return Instance(fields)
def text_to_instance(self,
sup_sents: List[str],
sup_labels: List[str] = None,
ori_unsup_sents = None,
aug_unsup_sents = None) -> Instance:
fields: Dict[str, Field] = {}
tokenized_sents = [self._tokenizer.tokenize(sent) for sent in sup_sents]
sentence_sequence = ListField([TextField(tk, self._token_indexers) for tk in tokenized_sents])
fields['sentences'] = sentence_sequence
if sup_labels is not None:
fields['labels'] = SequenceLabelField(sup_labels, sentence_sequence)
ori_name = 'ori_unsup_sentences_'
aug_name = 'aug_unsup_sentences_'
if ori_unsup_sents is not None and aug_unsup_sents is not None:
# Create TextField for each in ori_unsup_sentences
# ori_unsup_tokenized_sents = [self._tokenizer.tokenize(sent) for sent in ori_unsup_sents]
# ori_unsup_sentence_sequence = ListField([TextField(tk, self._token_indexers) for tk in ori_unsup_tokenized_sents])
ori_unsup_sentence_sequences = []
for ori_unsup_sent in ori_unsup_sents:
ori_unsup_tokenized_sent = [self._tokenizer.tokenize(sent) for sent in ori_unsup_sent]
ori_unsup_sentence_sequence = ListField([TextField(tk, self._token_indexers) for tk in ori_unsup_tokenized_sent])
ori_unsup_sentence_sequences.append(ori_unsup_sentence_sequence)
# Create TextField for aug_unsup_sentences
# aug_unsup_tokenized_sents = [self._tokenizer.tokenize(sent) for sent in aug_unsup_sents]
# aug_unsup_sentence_sequence = ListField([TextField(tk, self._token_indexers) for tk in aug_unsup_tokenized_sents])
aug_unsup_sentence_sequences = []
for aug_unsup_sent in aug_unsup_sents:
aug_unsup_tokenized_sent = [self._tokenizer.tokenize(sent) for sent in aug_unsup_sent]
aug_unsup_sentence_sequence = ListField([TextField(tk, self._token_indexers) for tk in aug_unsup_tokenized_sent])
aug_unsup_sentence_sequences.append(aug_unsup_sentence_sequence)
for i in range(len(ori_unsup_sentence_sequences)):
fields[ori_name + str(i)] = ori_unsup_sentence_sequences[i]
fields[aug_name + str(i)] = aug_unsup_sentence_sequences[i]
# This wont work because iterator in AllenNLP requires fields to be ListField to use BERT vocab on
# fields['ori_unsup_sentences'] = ori_unsup_sentence_sequences # List
# fields['aug_unsup_sentences'] = aug_unsup_sentence_sequences # List
# Fake data
# fields['ori_unsup_sentences'] = sentence_sequence
# fields['aug_unsup_sentences'] = sentence_sequence
return Instance(fields)
def test_elmo_bilm(self):
# get the raw data
sentences, expected_lm_embeddings = self._load_sentences_embeddings()
# load the test model
elmo_bilm = _ElmoBiLm(self.options_file, self.weight_file)
# Deal with the data.
indexer = ELMoTokenCharactersIndexer()
# For each sentence, first create a TextField, then create an instance
instances = []
for batch in izip(*sentences):
for sentence in batch:
tokens = [Token(token) for token in sentence.split()]
field = TextField(tokens, {u'character_ids': indexer})
instance = Instance({u"elmo": field})
instances.append(instance)
vocab = Vocabulary()
# Now finally we can iterate through batches.
iterator = BasicIterator(3)
iterator.index_with(vocab)
for i, batch in enumerate(
iterator(instances, num_epochs=1, shuffle=False)):
lm_embeddings = elmo_bilm(batch[u'elmo'][u'character_ids'])
top_layer_embeddings, mask = remove_sentence_boundaries(
lm_embeddings[u'activations'][2], lm_embeddings[u'mask'])
# check the mask lengths
lengths = mask.data.numpy().sum(axis=1)
batch_sentences = [sentences[k][i] for k in range(3)]
expected_lengths = [
len(sentence.split()) for sentence in batch_sentences
]
self.assertEqual(lengths.tolist(), expected_lengths)
# get the expected embeddings and compare!
expected_top_layer = [
expected_lm_embeddings[k][i] for k in range(3)
]
for k in range(3):
self.assertTrue(
numpy.allclose(
top_layer_embeddings[k, :lengths[k], :].data.numpy(),
expected_top_layer[k],
atol=1.0e-6))
def text_to_instance(self, mention_uniq_id, data=None) -> Instance:
l_tokenized = [Token('[CLS]')]
l_tokenized += [
Token(split_token)
for split_token in self.custom_tokenizer_class.tokenize(
txt=data['l'])
][:self.config.max_mention_length]
l_tokenized.append(Token('[SEP]'))
r_tokenized = [Token('[CLS]')]
r_tokenized += [
Token(split_token)
for split_token in self.custom_tokenizer_class.tokenize(
txt=data['r'])
][:self.config.max_mention_length]
r_tokenized.append(Token('[SEP]'))
l_plus_r = [Token('[CLS]')]
l_plus_r += [
Token(split_token)
for split_token in self.custom_tokenizer_class.tokenize(
txt=data['l'])
][:self.config.max_mention_length]
l_plus_r += [Token(BOND_TOKEN)]
l_plus_r += [
Token(split_token)
for split_token in self.custom_tokenizer_class.tokenize(
txt=data['r'])
][:self.config.max_mention_length]
l_plus_r += [Token('[SEP]')]
context_field = TextField(l_tokenized, self.token_indexers)
fields = {"l": context_field}
fields['lev'] = ArrayField(
np.array(Levenshtein.distance(data['l'], data['r'])))
# / (max(len(data['l']), len(data['r'])))))
fields['r'] = TextField(r_tokenized, self.token_indexers)
fields['l_plus_r'] = TextField(l_plus_r, self.token_indexers)
fields['label'] = ArrayField(np.array(data['label']))
fields['mention_uniq_id'] = ArrayField(np.array(mention_uniq_id))
fields['subword_match_num'] = ArrayField(
np.array(
len(
set(self.custom_tokenizer_class.tokenize(txt=data['l']))
& set(self.custom_tokenizer_class.tokenize(
txt=data['r'])))))
return Instance(fields)
def test_invalid_vocab_extension(self):
vocab_dir = self.TEST_DIR / "vocab_save"
original_vocab = Vocabulary(non_padded_namespaces=["tokens1"])
original_vocab.add_tokens_to_namespace(["a", "b"], namespace="tokens1")
original_vocab.add_token_to_namespace("p", namespace="tokens2")
original_vocab.save_to_files(vocab_dir)
text_field1 = TextField(
[Token(t) for t in ["a", "c"]], {"tokens1": SingleIdTokenIndexer("tokens1")}
)
text_field2 = TextField(
[Token(t) for t in ["p", "q", "r"]], {"tokens2": SingleIdTokenIndexer("tokens2")}
)
instances = Batch([Instance({"text1": text_field1, "text2": text_field2})])
# Following 2 should give error: tokens1 is non-padded in original_vocab but not in instances
params = Params(
{
"type": "extend",
"directory": vocab_dir,
"non_padded_namespaces": [],
"tokens_to_add": {"tokens1": ["a"], "tokens2": ["p"]},
}
)
with pytest.raises(ConfigurationError):
_ = Vocabulary.from_params(params, instances=instances)
# Following 2 should not give error: overlapping namespaces have same padding setting
params = Params(
{
"type": "extend",
"directory": vocab_dir,
"non_padded_namespaces": ["tokens1"],
"tokens_to_add": {"tokens1": ["a"], "tokens2": ["p"]},
}
)
Vocabulary.from_params(params, instances=instances)
# Following 2 should give error: tokens2 is padded in instances but not in original_vocab
params = Params(
{
"type": "extend",
"directory": vocab_dir,
"non_padded_namespaces": ["tokens1", "tokens2"],
"tokens_to_add": {"tokens1": ["a"], "tokens2": ["p"]},
}
)
with pytest.raises(ConfigurationError):
_ = Vocabulary.from_params(params, instances=instances)
def text_to_instance(self, prompt: List[List[str]], evidence: List[str],
non_evidence: List[str]):
fields = {
'comb_evidence':
TextField([
Token(x) for x in (['[CLS]'] + prompt[0] + prompt[1] +
prompt[2] + ['[SEP]'] + evidence)
], self.token_indexers),
'comb_non_evidence':
TextField([
Token(x) for x in (['[CLS]'] + prompt[0] + prompt[1] +
prompt[2] + ['[SEP]'] + non_evidence)
], self.token_indexers)
}
return Instance(fields)
def text_to_instance(x: Tuple[str]) -> Instance:
source_string = x[0]
target_string = x[1]
tokenized_source = source_tokenizer.tokenize(source_string)[:source_length_limit]
tokenized_source.insert(0, Token(START_SYMBOL))
tokenized_source.append(Token(END_SYMBOL))
source_field = TextField(tokenized_source, source_token_indexers)
tokenized_target = target_tokenizer.tokenize(target_string)[:target_length_limit]
tokenized_target.insert(0, Token(START_SYMBOL))
tokenized_target.append(Token(END_SYMBOL))
target_field = TextField(tokenized_target, target_token_indexers)
return Instance({source_field_name: source_field,
target_field_name: target_field})
def get_vocab_and_both_elmo_indexed_ids(batch: List[List[str]]):
instances = []
indexer = ELMoTokenCharactersIndexer()
indexer2 = SingleIdTokenIndexer()
for sentence in batch:
tokens = [Token(token) for token in sentence]
field = TextField(tokens,
{'character_ids': indexer,
'tokens': indexer2})
instance = Instance({"elmo": field})
instances.append(instance)
dataset = Batch(instances)
vocab = Vocabulary.from_instances(instances)
dataset.index_instances(vocab)
return vocab, dataset.as_tensor_dict()["elmo"]
def batch_to_ids(batch):
"""
Given a batch (as list of tokenized sentences), return a batch
of padded character ids.
"""
instances = []
for sentence in batch:
tokens = [Token(token) for token in sentence]
field = TextField(tokens, {'character_ids': indexer})
instance = Instance({"elmo": field})
instances.append(instance)
dataset = Batch(instances)
vocab = Vocabulary()
dataset.index_instances(vocab)
return dataset.as_tensor_dict()['elmo']['character_ids']
def text_to_instance(self,
tokens: List[Token],
tags: List[str] = None,
title: str = None) -> Instance:
tokens_field = TextField(tokens, self.token_indexers)
fields = {"tokens": tokens_field}
if tags:
tags_field = SequenceLabelField(labels=tags,
sequence_field=tokens_field)
fields["tags"] = tags_field
if title:
fields["title"] = MetadataField(title)
return Instance(fields)
def _make_instance(sent_):
""" Forward targs adds <s> as a target for input </s>
and bwd targs adds </s> as a target for input <s>
to avoid issues with needing to strip extra tokens
in the input for each direction """
d = {
"input":
sentence_to_text_field(sent_, indexers),
"targs":
sentence_to_text_field(sent_[1:] + [sent_[0]],
self.target_indexer),
"targs_b":
sentence_to_text_field([sent_[-1]] + sent_[:-1],
self.target_indexer),
}
return Instance(d)
def predictions_to_labeled_instances(self, instance: Instance,
outputs: Dict[str, numpy.ndarray]):
new_instance = instance.duplicate()
token_field: TextField = instance["tokens"] # type: ignore
mask_targets = [
Token(target_top_k_text[0], text_id=target_top_id_id)
for (target_top_k_text, target_top_id_id
) in zip(outputs["words"], outputs["token_ids"])
]
new_instance.add_field(
"target_ids",
TextField(mask_targets, token_field._token_indexers),
vocab=self._model.vocab,
)
return [new_instance]
def text_to_instance(self,
tokens: List[Token],
instance_id: int = -1,
labels: int = 0) -> Instance:
sentence_field = TextField(tokens, self.token_indexers)
label_field = LabelField(labels)
id_field = MetadataField(instance_id)
fields = {
"tokens": sentence_field,
'label': label_field,
"instance_id": id_field
}
return Instance(fields)
def text_to_instance(self,
tokens: List[Token],
label: int = None,
id: str = None) -> Instance:
fields = {}
sentence_field = TextField(tokens=tokens,
token_indexers=self.token_indexers)
fields['sentence'] = sentence_field
id_field = MetadataField(id)
fields['id'] = id_field
label_field = LabelField(label=label, skip_indexing=True)
fields['labels'] = label_field
return Instance(fields=fields)
def text_to_instance(self, tokens, label):
"""Build text and label field and convert tokens
to an ``Instance``.
Arguments:
tokens {List[str]} -- tokens
label {str} -- a label
Returns:
{Instance} -- a data instance
"""
sentence_field = TextField(tokens, self.token_indexers)
label_field = LabelField(label=label)
fields = {"text": sentence_field, "labels": label_field}
return Instance(fields)
def text_to_instance(self, tokenized_sentence: List[str],
spans: List[List[int]]) -> Instance:
allennlp_sentence_tokens = [Token(text=t) for t in tokenized_sentence]
sentence_token_indexes = TextField(allennlp_sentence_tokens,
self._token_indexers)
span_fields = []
for span_start, span_end_exclusive in spans:
span_field = SpanField(span_start, span_end_exclusive - 1,
sentence_token_indexes)
span_fields.append(span_field)
fields: Dict[str, Field] = {}
fields["tokens"] = sentence_token_indexes
fields["spans"] = ListField(span_fields)
return Instance(fields)
def text_to_instance(self, tokens: List[Token], tags: List[str]=None) -> Instance:
if len(tokens) > self._max_token_len:
tokens = tokens[:self._max_token_len]
print(f'Length of tokens exceeded the limit {self._max_token_len}. Truncating...')
if tags:
tags = tags[:self._max_token_len]
fields = {}
text_field = TextField(tokens, self._token_indexers)
fields['tokens'] = text_field
if tags:
fields['tags'] = SequenceLabelField(tags, text_field)
return Instance(fields)
def create_instance(self, str_tokens: List[str]):
tokens = [Token(t) for t in str_tokens]
instance = Instance({'text': TextField(tokens, self.token_indexers)})
instance.index_fields(self.vocab)
return instance
