"""
:type: OpenAttack.utils.BertClassifier
:Size: 1.23GB
:Package Requirements:
* transformers
* pytorch
Pretrained BERT model on MNLI dataset. See :py:data:`Dataset.MNLI` for detail.
"""
from OpenAttack.utils import make_zip_downloader, BertClassifier
NAME = "Victim.BERT.MNLI"
URL = "https://cdn.data.thunlp.org/TAADToolbox/victim/bert_mnli.zip"
DOWNLOAD = make_zip_downloader(URL)
def LOAD(path):
from OpenAttack import Classifier
return BertClassifier(path, 2)
"""
:type: function
:Size: 2.41MB
Model files for pos tagger in nltk.
`[code] <https://github.com/sloria/textblob-aptagger>`__
"""
from OpenAttack.utils import make_zip_downloader
import os
NAME = "TProcess.NLTKPerceptronPosTagger"
URL = "/TAADToolbox/averaged_perceptron_tagger.pickle.zip"
DOWNLOAD = make_zip_downloader(URL, "averaged_perceptron_tagger.pickle")
def LOAD(path):
ret = __import__("nltk").tag.PerceptronTagger(load=False)
ret.load("file:" + os.path.join(path, "averaged_perceptron_tagger.pickle"))
return ret.tag
"""
:type: OpenAttack.utils.WordVector
:Size: 3GB
"""
import numpy as np
import os
from OpenAttack.utils import make_zip_downloader
NAME = "AttackAssist.ChineseWord2Vec"
URL = "/TAADToolbox/chinese-merge-word-embedding.txt.zip"
DOWNLOAD = make_zip_downloader(URL, "chinese-merge-word-embedding.txt")
def LOAD(path):
from OpenAttack.attack_assist import WordEmbedding
with open(os.path.join(path, "chinese-merge-word-embedding.txt"), "r", encoding="utf-8") as f:
id2vec = []
word2id = {}
# f.readline()
for line in f.readlines():
tmp = line.strip().split(' ')
word = tmp[0]
embed = np.array([float(x) for x in tmp[1:]])
if len(embed) != 300:
continue
word2id[word] = len(word2id)
id2vec.append(embed)
id2vec = np.stack(id2vec)
return WordEmbedding(word2id, id2vec)
"""
:type: function
:Size: 158.351KB
Model files for nltk punkt sentence tokenizer.
"""
from OpenAttack.utils import make_zip_downloader
import os
NAME = "TProcess.NLTKSentTokenizer"
URL = "https://thunlp.oss-cn-qingdao.aliyuncs.com/TAADToolbox/punkt.english.pickle.zip"
DOWNLOAD = make_zip_downloader(URL, "english.pickle")
def LOAD(path):
return __import__("nltk").data.load("file:" + os.path.join(path, "english.pickle")).tokenize
"""
:type: OpenAttack.utils.WordVector
:Size: 61.998MB
Counter-fitting Word Vectors to Linguistic Constraints.
`[pdf] <https://www.aclweb.org/anthology/N16-1018.pdf>`__
"""
import numpy as np
import os
from OpenAttack.utils import make_zip_downloader
NAME = "AttackAssist.CounterFit"
URL = "/TAADToolbox/counter-fitted-vectors.txt.zip"
DOWNLOAD = make_zip_downloader(URL, "counter-fitted-vectors.txt")
def LOAD(path):
from OpenAttack.attack_assist import WordEmbedding
with open(os.path.join(path, "counter-fitted-vectors.txt"), "r", encoding='utf-8') as f:
id2vec = []
word2id = {}
for line in f.readlines():
tmp = line.strip().split(" ")
word = tmp[0]
embed = np.array([float(x) for x in tmp[1:]])
if len(embed) != 300:
continue
word2id[word] = len(word2id)
id2vec.append(embed)
id2vec = np.stack(id2vec)
