def default_features(cls) -> feature.Features:
return feature.Features(
{
"text": feature.Value("string"),
"edits": feature.Dict(
feature={
"start_idx": feature.Sequence(feature=feature.Value("int32")),
"end_idx": feature.Sequence(feature=feature.Value("int32")),
"corrections": feature.Sequence(
feature=feature.Sequence(feature=feature.Value("string"))
),
}
),
"text_length": feature.Value(
dtype="float",
description="length of the text",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
}
)
def default_features(cls) -> feature.Features:
return feature.Features({
"true_head":
feature.Value("string"),
"true_head_decipher":
feature.Value("string"),
"true_link":
feature.Value("string"),
"true_tail":
feature.Value("string"),
"true_tail_decipher":
feature.Value("string"),
"predict":
feature.Value("string"),
"true_label":
feature.Value("string"),
"predictions":
feature.Sequence(feature=feature.Value("string")),
"tail_entity_length":
feature.Value(
dtype="float",
description="number of words in the tail entity",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"head_entity_length":
feature.Value(
dtype="float",
description="number of words in the head entity",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"tail_fre":
feature.Value(
dtype="float",
description="the frequency of tail entity in the training set",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
"link_fre":
feature.Value(
dtype="float",
description=
"the frequency of link relation in the training set",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
"head_fre":
feature.Value(
dtype="float",
description=
"the frequency of head relation in the training set",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
"symmetry":
feature.Value(
dtype="string",
description=(
"boolean feature: 'symmetric' or 'asymmetric'; more "
"granularity to be added"),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_discrete_value",
number=2,
setting=1),
),
"entity_type_level":
feature.Value(
dtype="string",
description=
("most specific (highest) entity type level of true tail entity"
),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_discrete_value",
number=8,
setting=1),
),
})
def test_get_bucket_features(self):
ner_task_features = feature.Features({
"tokens":
feature.Sequence(feature=feature.Value("string")),
"true_tags":
feature.Sequence(feature=feature.Value("string")),
"pred_tags":
feature.Sequence(feature=feature.Value("string")),
# --- the following are features of the sentences ---
"sentence_length":
feature.Value(
dtype="float",
description="sentence length",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"entity_density":
feature.Value(
dtype="float",
description="the ration between all entity "
"tokens and sentence tokens ",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"num_oov":
feature.Value(
dtype="float",
description="the number of out-of-vocabulary words",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
"fre_rank":
feature.Value(
dtype="float",
description=(
"the average rank of each word based on its frequency in "
"training set"),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
# --- the following are features of each entity ---
"true_entity_info":
feature.Sequence(feature=feature.Dict(
feature={
"span_text":
feature.Value("string"),
"span_tokens":
feature.Value(
dtype="float",
description="entity length",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"span_pos":
feature.Position(positions=[0, 0]),
"span_tag":
feature.Value(
dtype="string",
description="entity tag",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_discrete_value",
number=4,
setting=1,
),
),
"span_capitalness":
feature.Value(
dtype="string",
description=(
"The capitalness of an entity. For example, "
"first_caps represents only the first character of "
"the entity is capital. full_caps denotes all "
"characters of the entity are capital"),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_discrete_value",
number=4,
setting=1,
),
),
"span_rel_pos":
feature.Value(
dtype="float",
description=(
"The relative position of an entity in a sentence"
),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"span_chars":
feature.Value(
dtype="float",
description="The number of characters of an entity",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"span_econ":
feature.Value(
dtype="float",
description="entity label consistency",
is_bucket=True,
require_training_set=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"span_efre":
feature.Value(
dtype="float",
description="entity frequency",
is_bucket=True,
require_training_set=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
})),
})
bucket_features = ner_task_features.get_bucket_features()
self.assertEqual(
set(bucket_features),
set([
'sentence_length',
'entity_density',
'num_oov',
'fre_rank',
'span_tokens',
'span_tag',
'span_capitalness',
'span_rel_pos',
'span_chars',
'span_econ',
'span_efre',
]),
)
def default_features(cls) -> feature.Features:
return feature.Features(
{
"context": feature.Value("string"),
"question": feature.Value("string"),
"options": feature.Sequence(feature=feature.Value("string")),
"answers": feature.Sequence(
feature=feature.Dict(
feature={
"text": feature.Value("string"),
"option_index": feature.Value("int32"),
}
)
),
"context_length": feature.Value(
dtype="float",
description="the length of context",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"question_length": feature.Value(
dtype="float",
description="the length of question",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"answer_length": feature.Value(
dtype="float",
description="the length of answer",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"num_oov": feature.Value(
dtype="float",
description="the number of out-of-vocabulary words",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
"fre_rank": feature.Value(
dtype="float",
description=(
"the average rank of each word based on its frequency in "
"training set"
),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
}
)
def default_features(cls) -> feature.Features:
return feature.Features({
"source":
feature.Value("string"),
"reference":
feature.Value("string"),
"hypothesis":
feature.Value("string"),
"source_length":
feature.Value(
dtype="float",
description="length of the source",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"reference_length":
feature.Value(
dtype="float",
description="length of the reference",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"hypothesis_length":
feature.Value(
dtype="float",
description="length of the hypothesis",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"src_num_oov":
feature.Value(
dtype="float",
description="OOV words in the source",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
"src_fre_rank":
feature.Value(
dtype="float",
description=(
"average training-set frequency rank of words in sentence"
),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
"ref_num_oov":
feature.Value(
dtype="float",
description="number of OOV words in reference",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
"ref_fre_rank":
feature.Value(
dtype="float",
description=(
"average training-set frequency rank of words in sentence"
),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
# --- the following are features of each token ---
"ref_tok_info":
feature.Sequence(feature=feature.Dict(
feature={
"tok_text":
feature.Value("string"),
"tok_pos":
feature.Position(positions=[0, 0]),
"tok_matched":
feature.Value(
# this is actually "int" but int is not supported
dtype="float",
description=(
"which token the ref/hyp token matches in the "
"hyp/ref sentence, or -1 if none"),
is_bucket=False,
),
"tok_capitalness":
feature.Value(
dtype="string",
description=("capitalness of token"),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_discrete_value",
number=4,
setting=1,
),
),
"tok_position":
feature.Value(
dtype="float",
description=("relative position of token in sentence"),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"tok_chars":
feature.Value(
dtype="float",
description="number of characters in the token",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"tok_test_freq":
feature.Value(
dtype="float",
description="tok frequency in the test set",
is_bucket=True,
require_training_set=False,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"tok_train_freq":
feature.Value(
dtype="float",
description="tok frequency in the training set",
is_bucket=True,
require_training_set=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
})),
})
def default_features(cls) -> feature.Features:
return feature.Features({
"text":
feature.Value("string"),
"log_probs":
feature.Value("string"),
"text_length":
feature.Value(
dtype="float",
description="text length in tokens",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"text_chars":
feature.Value(
dtype="float",
description="text length in characters",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"num_oov":
feature.Value(
dtype="float",
description="the number of out-of-vocabulary words",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
"fre_rank":
feature.Value(
dtype="float",
description=(
"the average rank of each work based on its frequency in "
"training set"),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
"length_fre":
feature.Value(
dtype="float",
description="the frequency of text length in training set",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
require_training_set=True,
),
# --- the following are features of each token ---
"tok_info":
feature.Sequence(feature=feature.Dict(
feature={
"tok_text":
feature.Value("string"),
"tok_pos":
feature.Position(positions=[0, 0]),
"tok_log_prob":
feature.Value(
dtype="float",
description=(
"log probability of the token according to the LM"
),
is_bucket=False,
),
"tok_capitalness":
feature.Value(
dtype="string",
description=(
"The capitalness of an token. For example, "
"first_caps represents only the first character of "
"the token is capital. full_caps denotes all "
"characters of the token are capital"),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_discrete_value",
number=4,
setting=1,
),
),
"tok_position":
feature.Value(
dtype="float",
description=(
"The relative position of a token in a sentence"),
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"tok_chars":
feature.Value(
dtype="float",
description="The number of characters in a token",
is_bucket=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"tok_test_freq":
feature.Value(
dtype="float",
description="tok frequency in the test set",
is_bucket=True,
require_training_set=False,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
"tok_train_freq":
feature.Value(
dtype="float",
description="tok frequency in the training set",
is_bucket=True,
require_training_set=True,
bucket_info=feature.BucketInfo(
method="bucket_attribute_specified_bucket_value",
number=4,
setting=(),
),
),
})),
})