utils

download_file(save_dir, filename, url, md5=None)[源代码]

Download the file from the url to specified directory. Check md5 value when the file is exists, if the md5 value is the same as the existed file, just use the older file, if not, will download the file from the url.

参数
  • save_dir (string) -- The specified directory saving the file.

  • filename (string) -- The specified filename saving the file.

  • url (string) -- The url downling the file.

  • md5 (string, optional) -- The md5 value that checking the version downloaded.

download_check(task)[源代码]

Check the resource status in the specified task.

参数

task (string) -- The name of specified task.

add_docstrings(*docstr)[源代码]

The function that add the doc string to doc of class.

cut_chinese_sent(para)[源代码]

Cut the Chinese sentences more precisely, reference to "https://blog.csdn.net/blmoistawinde/article/details/82379256".

class TermTreeNode(sid: str, term: str, base: str, node_type: str = 'term', term_type: Optional[str] = None, hyper: Optional[str] = None, level: Optional[int] = None, alias: Optional[List[str]] = None, alias_ext: Optional[List[str]] = None, sub_type: Optional[List[str]] = None, sub_term: Optional[List[str]] = None, data: Optional[Dict[str, Any]] = None)[源代码]

基类:object

Defination of term node. All members are protected, to keep rigorism of data struct.

参数
  • sid (str) -- term id of node.

  • term (str) -- term, common name of this term.

  • base (str) -- cb indicates concept base, eb indicates entity base.

  • term_type (Optional[str], optional) -- type of this term, constructs hirechical of term node. Defaults to None.

  • hyper (Optional[str], optional) -- parent type of a type node. Defaults to None.

  • node_type (str, optional) -- type statement of node, type or term. Defaults to "term".

  • alias (Optional[List[str]], optional) -- alias of this term. Defaults to None.

  • alias_ext (Optional[List[str]], optional) -- extended alias of this term, CANNOT be used in matching. Defaults to None.

  • sub_type (Optional[List[str]], optional) -- grouped by some term. Defaults to None.

  • sub_term (Optional[List[str]], optional) -- some lower term. Defaults to None.

  • data (Optional[Dict[str, Any]], optional) -- to sore full imformation of a term. Defaults to None.

classmethod from_dict(data: Dict[str, Any])[源代码]

Build a node from dictionary data.

参数

data (Dict[str, Any]) -- Dictionary data contain all k-v data.

返回

TermTree node object.

返回类型

[type]

classmethod from_json(json_str: str)[源代码]

Build a node from JSON string.

参数

json_str (str) -- JSON string formatted by TermTree data.

返回

TermTree node object.

返回类型

[type]

class TermTree[源代码]

基类:object

TermTree class.

add_term(term: Optional[str] = None, base: Optional[str] = None, term_type: Optional[str] = None, sub_type: Optional[List[str]] = None, sub_term: Optional[List[str]] = None, alias: Optional[List[str]] = None, alias_ext: Optional[List[str]] = None, data: Optional[Dict[str, Any]] = None)[源代码]

Add a term into TermTree.

参数
  • term (str) -- common name of name.

  • base (str) -- term is concept or entity.

  • term_type (str) -- term type of this term

  • sub_type (Optional[List[str]], optional) -- sub type of this term, must exists in TermTree. Defaults to None.

  • sub_terms (Optional[List[str]], optional) -- sub terms of this term. Defaults to None.

  • alias (Optional[List[str]], optional) -- alias of this term. Defaults to None.

  • alias_ext (Optional[List[str]], optional) -- . Defaults to None.

  • data (Optional[Dict[str, Any]], optional) -- [description]. Defaults to None.

find_term(term: str, term_type: Optional[str] = None)Tuple[bool, Optional[List[str]]][源代码]

Find a term in Term Tree. If term not exists, return None. If term_type is not None, will find term with this type.

参数
  • term (str) -- term to look up.

  • term_type (Optional[str], optional) -- find term in this term_type. Defaults to None.

返回

[description]

返回类型

Union[None, List[str]]

build_from_dir(term_schema_path, term_data_path, linking=True)[源代码]

Build TermTree from a directory which should contain type schema and term data.

参数

dir ([type]) -- [description]

classmethod from_dir(term_schema_path, term_data_path, linking)paddlenlp.taskflow.utils.TermTree[源代码]

Build TermTree from a directory which should contain type schema and term data.

参数

source_dir ([type]) -- [description]

返回

[description]

返回类型

TermTree

save(save_dir)[源代码]

Save term tree to directory save_dir

参数

save_dir ([type]) -- Directory.

levenstein_distance(s1: str, s2: str)int[源代码]

Calculate minimal Levenstein distance between s1 and s2.

参数
  • s1 (str) -- string

  • s2 (str) -- string

返回

the minimal distance.

返回类型

int

class BurkhardKellerNode(word: str)[源代码]

基类:object

Node implementatation for BK-Tree. A BK-Tree node stores the information of current word, and its approximate words calculated by levenstein distance.

参数

word (str) -- word of current node.

class BurkhardKellerTree[源代码]

基类:object

Implementataion of BK-Tree

add(word: str)[源代码]

Insert a word into current tree. If tree is empty, set this word to root.

参数

word (str) -- word to be inserted.

search_similar_word(word: str)List[str][源代码]

Search the most similar (minimal levenstain distance) word between s.

参数

s (str) -- target word

返回

similar words.

返回类型

List[str]

class TriedTree[源代码]

基类:object

Implementataion of TriedTree

add_word(word)[源代码]

add single word into TriedTree

search(content)[源代码]

Backward maximum matching

参数

content (str) -- string to be searched

返回

list of maximum matching words, each element represents

the starting and ending position of the matching string.

返回类型

List[Tuple]

class Customization[源代码]

基类:object

User intervention based on Aho-Corasick automaton

load_customization(filename, sep=None)[源代码]

Load the custom vocab

parse_customization(query, lac_tags, prefix=False)[源代码]

Use custom vocab to modify the lac results

class SchemaTree(name='root', children=None)[源代码]

基类:object

Implementataion of SchemaTree

get_bool_ids_greater_than(probs, limit=0.5, return_prob=False)[源代码]

get idx of the last dim in prob arraies, which is greater than a limitation input: [[0.1, 0.1, 0.2, 0.5, 0.1, 0.3], [0.7, 0.6, 0.1, 0.1, 0.1, 0.1]]

0.4

output: [[3], [0, 1]]

get_span(start_ids, end_ids, with_prob=False)[源代码]

every id can only be used once get span set from position start and end list input: [1, 2, 10] [4, 12] output: set((2, 4), (10, 12))

get_id_and_prob(span_set, offset_mapping)[源代码]

Return text id and probability of predicted spans

参数
  • span_set (set) -- set of predicted spans.

  • offset_mapping (list[int]) -- list of pair preserving the index of start and end char in original text pair (prompt + text) for each token.

返回

index of start and end char in original text. prob (list[float]): probabilities of predicted spans.

返回类型

sentence_id (list[tuple])

class WordTagRelationExtractor(schema)[源代码]

基类:object

Implement of information extractor.

classmethod from_dict(config_dict)[源代码]

Make an instance from a configuration dictionary.

参数

config_dict (Dict[str, Any]) -- configuration dict.

classmethod from_json(json_str)[源代码]

Implement an instance from JSON str.

classmethod from_pkl(pkl_path)[源代码]

Implement an instance from a serialized pickle package.

classmethod from_config(config_path)[源代码]

Implement an instance from a configuration file.

add_schema_from_dict(config_dict)[源代码]

Add the schema from the dict.

extract_spo(all_items)[源代码]

Pipeline of mining procedure.

参数

all_items ([type]) -- [description]

class DataCollatorGP(tokenizer: paddlenlp.transformers.tokenizer_utils_base.PretrainedTokenizerBase, padding: Union[bool, str, paddlenlp.transformers.tokenizer_utils_base.PaddingStrategy] = True, max_length: Union[int, NoneType] = None, label_maps: Union[dict, NoneType] = None, task_type: Union[str, NoneType] = None)[源代码]

基类:object

class DocSpan(start, length)

基类:tuple

length

Alias for field number 1

start

Alias for field number 0

class Example(keys, key_labels, doc_tokens, text, qas_id, model_type, seq_labels, ori_boxes, boxes, segment_ids, symbol_ids, im_base64, image_rois)

基类:tuple

boxes

Alias for field number 8

doc_tokens

Alias for field number 2

im_base64

Alias for field number 11

image_rois

Alias for field number 12

key_labels

Alias for field number 1

keys

Alias for field number 0

model_type

Alias for field number 5

ori_boxes

Alias for field number 7

qas_id

Alias for field number 4

segment_ids

Alias for field number 9

seq_labels

Alias for field number 6

symbol_ids

Alias for field number 10

text

Alias for field number 3

class Feature(unique_id, example_index, qas_id, doc_span_index, tokens, token_to_orig_map, token_is_max_context, token_ids, position_ids, text_type_ids, text_symbol_ids, overlaps, key_labels, seq_labels, se_seq_labels, bio_seq_labels, bioes_seq_labels, keys, model_type, doc_tokens, doc_labels, text, boxes, segment_ids, im_base64, image_rois)

基类:tuple

bio_seq_labels

Alias for field number 15

bioes_seq_labels

Alias for field number 16

boxes

Alias for field number 22

doc_labels

Alias for field number 20

doc_span_index

Alias for field number 3

doc_tokens

Alias for field number 19

example_index

Alias for field number 1

im_base64

Alias for field number 24

image_rois

Alias for field number 25

key_labels

Alias for field number 12

keys

Alias for field number 17

model_type

Alias for field number 18

overlaps

Alias for field number 11

position_ids

Alias for field number 8

qas_id

Alias for field number 2

se_seq_labels

Alias for field number 14

segment_ids

Alias for field number 23

seq_labels

Alias for field number 13

text

Alias for field number 21

text_symbol_ids

Alias for field number 10

text_type_ids

Alias for field number 9

token_ids

Alias for field number 7

token_is_max_context

Alias for field number 6

token_to_orig_map

Alias for field number 5

tokens

Alias for field number 4

unique_id

Alias for field number 0

class Compose(transforms, ctx=None)[源代码]

基类:object

compose

class ProcessReader(dataset=None, sample_transforms=None, batch_transforms=None, batch_size=None, shuffle=False, drop_last=False, drop_empty=True, mixup_epoch=- 1, cutmix_epoch=- 1, class_aware_sampling=False, use_process=False, use_fine_grained_loss=False, num_classes=80, bufsize=- 1, memsize='3G', inputs_def=None, devices_num=1, num_trainers=1)[源代码]

基类:object

参数
  • dataset (DataSet) -- DataSet object

  • sample_transforms (list of BaseOperator) -- a list of sample transforms operators.

  • batch_transforms (list of BaseOperator) -- a list of batch transforms operators.

  • batch_size (int) -- batch size.

  • shuffle (bool) -- whether shuffle dataset or not. Default False.

  • drop_last (bool) -- whether drop last batch or not. Default False.

  • drop_empty (bool) -- whether drop sample when it's gt is empty or not. Default True.

  • mixup_epoch (int) -- mixup epoc number. Default is -1, meaning not use mixup.

  • cutmix_epoch (int) -- cutmix epoc number. Default is -1, meaning not use cutmix.

  • class_aware_sampling (bool) -- whether use class-aware sampling or not. Default False.

  • worker_num (int) -- number of working threads/processes. Default -1, meaning not use multi-threads/multi-processes.

  • use_process (bool) -- whether use multi-processes or not. It only works when worker_num > 1. Default False.

  • bufsize (int) -- buffer size for multi-threads/multi-processes, please note, one instance in buffer is one batch data.

  • memsize (str) -- size of shared memory used in result queue when use_process is true. Default 3G.

  • inputs_def (dict) -- network input definition use to get input fields, which is used to determine the order of returned data.

  • devices_num (int) -- number of devices.

  • num_trainers (int) -- number of trainers. Default 1.

process(dataset)[源代码]
worker(drop_empty=True, batch_samples=None)[源代码]

sample transform and batch transform.

pad_batch_data(insts, pad_idx=0, max_seq_len=None, return_pos=False, return_input_mask=False, return_max_len=False, return_num_token=False, return_seq_lens=False, pad_2d_pos_ids=False, pad_segment_id=False, select=False, extract=False)[源代码]

Pad the instances to the max sequence length in batch, and generate the corresponding position data and attention bias.

get_final_text(pred_text, orig_text, tokenizer, do_lower_case)[源代码]

Project the tokenized prediction back to the original text.

find_bio_pos(label)[源代码]

find answer position from BIO label

calEuclidean(x_list, y_list)[源代码]

Calculate euclidean distance

longestCommonSequence(question_tokens, context_tokens)[源代码]

Longest common sequence