Source code for textattack.shared.utils.strings

import re
import string

import flair
import jieba

from .importing import LazyLoader


[docs]def has_letter(word): """Returns true if `word` contains at least one character in [A-Za-z].""" return re.search("[A-Za-z]+", word) is not None
[docs]def is_one_word(word): return len(words_from_text(word)) == 1
[docs]def add_indent(s_, numSpaces): s = s_.split("\n") # don't do anything for single-line stuff if len(s) == 1: return s_ first = s.pop(0) s = [(numSpaces * " ") + line for line in s] s = "\n".join(s) s = first + "\n" + s return s
[docs]def words_from_text(s, words_to_ignore=[]): """Lowercases a string, removes all non-alphanumeric characters, and splits into words.""" try: if re.search("[\u4e00-\u9FFF]", s): seg_list = jieba.cut(s, cut_all=False) s = " ".join(seg_list) else: s = " ".join(s.split()) except Exception: s = " ".join(s.split()) homos = """˗৭Ȣ𝟕бƼᏎƷᒿlO`ɑЬϲԁе𝚏ɡհіϳ𝒌ⅼmոорԛⲅѕ𝚝սѵԝ×уᴢ""" exceptions = """'-_*@""" filter_pattern = homos + """'\\-_\\*@""" # TODO: consider whether one should add "." to `exceptions` (and "\." to `filter_pattern`) # example "My email address is xxx@yyy.com" filter_pattern = f"[\\w{filter_pattern}]+" words = [] for word in s.split(): # Allow apostrophes, hyphens, underscores, asterisks and at signs as long as they don't begin the word. word = word.lstrip(exceptions) filt = [w.lstrip(exceptions) for w in re.findall(filter_pattern, word)] words.extend(filt) words = list(filter(lambda w: w not in words_to_ignore + [""], words)) return words
[docs]class TextAttackFlairTokenizer(flair.data.Tokenizer):
[docs] def tokenize(self, text: str): return words_from_text(text)
[docs]def default_class_repr(self): if hasattr(self, "extra_repr_keys"): extra_params = [] for key in self.extra_repr_keys(): extra_params.append(" (" + key + ")" + ": {" + key + "}") if len(extra_params): extra_str = "\n" + "\n".join(extra_params) + "\n" extra_str = f"({extra_str})" else: extra_str = "" extra_str = extra_str.format(**self.__dict__) else: extra_str = "" return f"{self.__class__.__name__}{extra_str}"
[docs]class ReprMixin(object): """Mixin for enhanced __repr__ and __str__.""" def __repr__(self): return default_class_repr(self) __str__ = __repr__
[docs] def extra_repr_keys(self): """Extra fields to be included in the representation of a class.""" return []
LABEL_COLORS = [ "red", "green", "blue", "purple", "yellow", "orange", "pink", "cyan", "gray", "brown", ]
[docs]def process_label_name(label_name): """Takes a label name from a dataset and makes it nice. Meant to correct different abbreviations and automatically capitalize. """ label_name = label_name.lower() if label_name == "neg": label_name = "negative" elif label_name == "pos": label_name = "positive" return label_name.capitalize()
[docs]def color_from_label(label_num): """Arbitrary colors for different labels.""" try: label_num %= len(LABEL_COLORS) return LABEL_COLORS[label_num] except TypeError: return "blue"
[docs]def color_from_output(label_name, label): """Returns the correct color for a label name, like 'positive', 'medicine', or 'entailment'.""" label_name = label_name.lower() if label_name in {"entailment", "positive"}: return "green" elif label_name in {"contradiction", "negative"}: return "red" elif label_name in {"neutral"}: return "gray" else: # if no color pre-stored for label name, return color corresponding to # the label number (so, even for unknown datasets, we can give each # class a distinct color) return color_from_label(label)
[docs]class ANSI_ESCAPE_CODES: """Escape codes for printing color to the terminal.""" HEADER = "\033[95m" OKBLUE = "\033[94m" OKGREEN = "\033[92m" GRAY = "\033[37m" PURPLE = "\033[35m" YELLOW = "\033[93m" ORANGE = "\033[38:5:208m" PINK = "\033[95m" CYAN = "\033[96m" GRAY = "\033[38:5:240m" BROWN = "\033[38:5:52m" WARNING = "\033[93m" FAIL = "\033[91m" BOLD = "\033[1m" UNDERLINE = "\033[4m" """This color stops the current color sequence.""" STOP = "\033[0m"
[docs]def color_text(text, color=None, method=None): if not (isinstance(color, str) or isinstance(color, tuple)): raise TypeError(f"Cannot color text with provided color of type {type(color)}") if isinstance(color, tuple): if len(color) > 1: text = color_text(text, color[1:], method) color = color[0] if method is None: return text if method == "html": return f"<font color = {color}>{text}</font>" elif method == "ansi": if color == "green": color = ANSI_ESCAPE_CODES.OKGREEN elif color == "red": color = ANSI_ESCAPE_CODES.FAIL elif color == "blue": color = ANSI_ESCAPE_CODES.OKBLUE elif color == "purple": color = ANSI_ESCAPE_CODES.PURPLE elif color == "yellow": color = ANSI_ESCAPE_CODES.YELLOW elif color == "orange": color = ANSI_ESCAPE_CODES.ORANGE elif color == "pink": color = ANSI_ESCAPE_CODES.PINK elif color == "cyan": color = ANSI_ESCAPE_CODES.CYAN elif color == "gray": color = ANSI_ESCAPE_CODES.GRAY elif color == "brown": color = ANSI_ESCAPE_CODES.BROWN elif color == "bold": color = ANSI_ESCAPE_CODES.BOLD elif color == "underline": color = ANSI_ESCAPE_CODES.UNDERLINE elif color == "warning": color = ANSI_ESCAPE_CODES.WARNING else: raise ValueError(f"unknown text color {color}") return color + text + ANSI_ESCAPE_CODES.STOP elif method == "file": return "[[" + text + "]]"
_flair_pos_tagger = None
[docs]def flair_tag(sentence, tag_type="upos-fast"): """Tags a `Sentence` object using `flair` part-of-speech tagger.""" global _flair_pos_tagger if not _flair_pos_tagger: from flair.models import SequenceTagger _flair_pos_tagger = SequenceTagger.load(tag_type) _flair_pos_tagger.predict(sentence, force_token_predictions=True)
[docs]def zip_flair_result(pred, tag_type="upos-fast"): """Takes a sentence tagging from `flair` and returns two lists, of words and their corresponding parts-of-speech.""" from flair.data import Sentence if not isinstance(pred, Sentence): raise TypeError("Result from Flair POS tagger must be a `Sentence` object.") tokens = pred.tokens word_list = [] pos_list = [] for token in tokens: word_list.append(token.text) if "pos" in tag_type: pos_list.append(token.annotation_layers["upos"][0]._value) elif tag_type == "ner": pos_list.append(token.get_label("ner")) return word_list, pos_list
stanza = LazyLoader("stanza", globals(), "stanza")
[docs]def zip_stanza_result(pred, tagset="universal"): """Takes the first sentence from a document from `stanza` and returns two lists, one of words and the other of their corresponding parts-of- speech.""" if not isinstance(pred, stanza.models.common.doc.Document): raise TypeError("Result from Stanza POS tagger must be a `Document` object.") word_list = [] pos_list = [] for sentence in pred.sentences: for word in sentence.words: word_list.append(word.text) if tagset == "universal": pos_list.append(word.upos) else: pos_list.append(word.xpos) return word_list, pos_list
[docs]def check_if_subword(token, model_type, starting=False): """Check if ``token`` is a subword token that is not a standalone word. Args: token (str): token to check. model_type (str): type of model (options: "bert", "roberta", "xlnet"). starting (bool): Should be set ``True`` if this token is the starting token of the overall text. This matters because models like RoBERTa does not add "Ġ" to beginning token. Returns: (bool): ``True`` if ``token`` is a subword token. """ avail_models = [ "bert", "gpt", "gpt2", "roberta", "bart", "electra", "longformer", "xlnet", ] if model_type not in avail_models: raise ValueError( f"Model type {model_type} is not available. Options are {avail_models}." ) if model_type in ["bert", "electra"]: return True if "##" in token else False elif model_type in ["gpt", "gpt2", "roberta", "bart", "longformer"]: if starting: return False else: return False if token[0] == "Ġ" else True elif model_type == "xlnet": return False if token[0] == "_" else True else: return False
[docs]def strip_BPE_artifacts(token, model_type): """Strip characters such as "Ġ" that are left over from BPE tokenization. Args: token (str) model_type (str): type of model (options: "bert", "roberta", "xlnet") """ avail_models = [ "bert", "gpt", "gpt2", "roberta", "bart", "electra", "longformer", "xlnet", ] if model_type not in avail_models: raise ValueError( f"Model type {model_type} is not available. Options are {avail_models}." ) if model_type in ["bert", "electra"]: return token.replace("##", "") elif model_type in ["gpt", "gpt2", "roberta", "bart", "longformer"]: return token.replace("Ġ", "") elif model_type == "xlnet": if len(token) > 1 and token[0] == "_": return token[1:] else: return token else: return token
[docs]def check_if_punctuations(word): """Returns ``True`` if ``word`` is just a sequence of punctuations.""" for c in word: if c not in string.punctuation: return False return True