def _get_vocab(data_list):
vocab = Vocabulary(unknown=unk_str, padding=pad_str)
for l in data_list:
vocab.add_word_lst(l)
vocab.build_vocab()
print('vocab', len(vocab))
return vocab
def test_same_vector5(self):
# 检查通过使用min_freq后的word是否内容一致
word_lst = ["they", "the", "they", "the", 'he', 'he', "a", "A"]
no_create_word_lst = ['of', "of", "she", "she", 'With', 'with']
all_words = word_lst[:-2] + no_create_word_lst[:-2]
vocab = Vocabulary().add_word_lst(word_lst)
vocab.add_word_lst(no_create_word_lst, no_create_entry=True)
embed = StaticEmbedding(vocab,
model_dir_or_name='en-glove-6B-100d',
lower=False,
min_freq=2)
words = torch.LongTensor([[vocab.to_index(word)
for word in all_words]])
words = embed(words)
min_freq_vocab = Vocabulary(min_freq=2).add_word_lst(word_lst)
min_freq_vocab.add_word_lst(no_create_word_lst, no_create_entry=True)
min_freq_embed = StaticEmbedding(min_freq_vocab,
model_dir_or_name='en-glove-6B-100d',
lower=False)
min_freq_words = torch.LongTensor(
[[min_freq_vocab.to_index(word.lower()) for word in all_words]])
min_freq_words = min_freq_embed(min_freq_words)
for idx in range(len(all_words)):
word_i, word_j = words[0, idx], min_freq_words[0, idx]
with self.subTest(idx=idx, word=all_words[idx]):
assert torch.sum(word_i == word_j).eq(
min_freq_embed.embed_size)
def test_save_and_load(self):
fp = 'vocab_save_test.txt'
try:
# check word2idx没变，no_create_entry正常
words = list('abcdefaddfdkjfe')
no_create_entry = list('12342331')
unk = '[UNK]'
vocab = Vocabulary(unknown=unk, max_size=500)
vocab.add_word_lst(words)
vocab.add_word_lst(no_create_entry, no_create_entry=True)
vocab.save(fp)
new_vocab = Vocabulary.load(fp)
for word, index in vocab:
self.assertEqual(new_vocab.to_index(word), index)
for word in no_create_entry:
self.assertTrue(new_vocab._is_word_no_create_entry(word))
for word in words:
self.assertFalse(new_vocab._is_word_no_create_entry(word))
for idx in range(len(vocab)):
self.assertEqual(vocab.to_word(idx), new_vocab.to_word(idx))
self.assertEqual(vocab.unknown, new_vocab.unknown)
except:
import os
if os.path.exists(fp):
os.remove(fp)
def test_same_vector4(self):
# 验证在有min_freq下的lower
word_lst = ["The", "the", "the", "The", "a", "A"]
no_create_word_lst = ['of', 'Of', "Of", "of", 'With', 'with']
all_words = word_lst[:-2] + no_create_word_lst[:-2]
vocab = Vocabulary(min_freq=2).add_word_lst(word_lst)
vocab.add_word_lst(no_create_word_lst, no_create_entry=True)
embed = StaticEmbedding(vocab,
model_dir_or_name='en-glove-6B-100d',
lower=True)
words = torch.LongTensor([[vocab.to_index(word)
for word in all_words]])
words = embed(words)
lowered_word_lst = [word.lower() for word in word_lst]
lowered_no_create_word_lst = [
word.lower() for word in no_create_word_lst
]
lowered_vocab = Vocabulary().add_word_lst(lowered_word_lst)
lowered_vocab.add_word_lst(lowered_no_create_word_lst,
no_create_entry=True)
lowered_embed = StaticEmbedding(lowered_vocab,
model_dir_or_name='en-glove-6B-100d',
lower=False)
lowered_words = torch.LongTensor(
[[lowered_vocab.to_index(word.lower()) for word in all_words]])
lowered_words = lowered_embed(lowered_words)
for idx in range(len(all_words)):
word_i, word_j = words[0, idx], lowered_words[0, idx]
with self.subTest(idx=idx, word=all_words[idx]):
assert torch.sum(word_i == word_j).eq(lowered_embed.embed_size)
def test_case(self):
vocab = Vocabulary().add_word_lst("This is a test .".split())
vocab.add_word_lst("Another test !".split())
embed = StaticEmbedding(vocab,
model_dir_or_name=None,
embedding_dim=10)
encoder_output = torch.randn(2, 3, 10)
tgt_words_idx = torch.LongTensor([[1, 2, 3, 4], [2, 3, 0, 0]])
src_seq_len = torch.LongTensor([3, 2])
encoder_mask = seq_len_to_mask(src_seq_len)
for flag in [True, False]:
for attention in [True, False]:
with self.subTest(bind_decoder_input_output_embed=flag,
attention=attention):
decoder = LSTMSeq2SeqDecoder(
embed=embed,
num_layers=2,
hidden_size=10,
dropout=0.3,
bind_decoder_input_output_embed=flag,
attention=attention)
state = decoder.init_state(encoder_output, encoder_mask)
output = decoder(tgt_words_idx, state)
self.assertEqual(tuple(output.size()), (2, 4, len(vocab)))
def test_case(self):
vocab = Vocabulary().add_word_lst("This is a test .".split())
vocab.add_word_lst("Another test !".split())
embed = StaticEmbedding(vocab, embedding_dim=10)
encoder_output = torch.randn(2, 3, 10)
src_seq_len = torch.LongTensor([3, 2])
encoder_mask = seq_len_to_mask(src_seq_len)
for flag in [True, False]:
with self.subTest(bind_decoder_input_output_embed=flag):
decoder = TransformerSeq2SeqDecoder(
embed=embed,
pos_embed=None,
d_model=10,
num_layers=2,
n_head=5,
dim_ff=20,
dropout=0.1,
bind_decoder_input_output_embed=True)
state = decoder.init_state(encoder_output, encoder_mask)
output = decoder(tokens=torch.randint(0,
len(vocab),
size=(2, 4)),
state=state)
self.assertEqual(output.size(), (2, 4, len(vocab)))
def word_to_id(glove_data, glove_matrix, vocab_dict_path, file_path):
if os.path.exists(glove_data) == False or os.path.exists(
glove_matrix) == False:
data, feature_words, user_num, item_num, = feature_word(file_path)
vocab = Vocabulary(max_size=len(feature_words) + 1,
unknown='unk',
padding='PAD')
vocab.add_word_lst(feature_words)
vocab.build_vocab()
matrix = EmbedLoader.load_with_vocab(vocab_dict_path, vocab)
matrix = torch.tensor(matrix)
for d in range(len(data)):
review = []
for word in data[d]['reviewText']:
review.append(vocab.to_index(word))
data[d]['reviewText'] = review
with open(glove_data, 'wb') as f:
pickle.dump(data, f)
with open(glove_matrix, 'wb') as f:
pickle.dump(matrix, f)
with open(glove_data, 'rb') as f:
glove_data = pickle.load(f)
with open(glove_matrix, 'rb') as f:
matrix = pickle.load(f)
return glove_data, matrix, len(glove_data[0]['reviewText'])
def test_search(self):
"""语义搜索.TypeError: expected dimension <= 2 array or matrix
"""
print('{} test_search {}'.format('-' * 15, '-' * 15))
texts = ['温都尔站', '东乌广厦', '国电四郎', '阿尔善站', '朱日和基']
# 文本向量化
vocab = Vocabulary()
for text in texts:
vocab.add_word_lst(list(text))
print(len(vocab))
embed = StaticEmbedding(
vocab, model_dir_or_name='./data/cn_char_fastnlp_100d.txt')
texts_to_id = [[vocab.to_index(word) for word in list(text)]
for text in texts]
words = torch.LongTensor(texts_to_id) # 将文本转为index
features_vec = embed(words)
print(features_vec.shape)
# build the search index!
cp = ci.MultiClusterIndex(features_vec.detach().numpy(), texts)
search_texts = ['朱日和站', '温都尔站', '国电站']
for text in search_texts:
texts_to_id = [[vocab.to_index(word) for word in list(text)]]
words = torch.LongTensor(texts_to_id) # 将文本转为index
features_vec = embed(words)
search_features_vec = features_vec.detach().numpy()
search_result = cp.search(search_features_vec,
k=2,
k_clusters=2,
return_distance=True)
print('text:{}'.format(text))
print('search_result:{}'.format(search_result))
"""
def test_same_vector3(self):
# 验证lower
word_lst = ["The", "the"]
no_create_word_lst = ['of', 'Of', 'With', 'with']
vocab = Vocabulary().add_word_lst(word_lst)
vocab.add_word_lst(no_create_word_lst, no_create_entry=True)
embed = StaticEmbedding(vocab,
model_dir_or_name='en-glove-6B-100d',
lower=True)
words = torch.LongTensor(
[[vocab.to_index(word) for word in word_lst + no_create_word_lst]])
words = embed(words)
lowered_word_lst = [word.lower() for word in word_lst]
lowered_no_create_word_lst = [
word.lower() for word in no_create_word_lst
]
lowered_vocab = Vocabulary().add_word_lst(lowered_word_lst)
lowered_vocab.add_word_lst(lowered_no_create_word_lst,
no_create_entry=True)
lowered_embed = StaticEmbedding(lowered_vocab,
model_dir_or_name='en-glove-6B-100d',
lower=False)
lowered_words = torch.LongTensor([[
lowered_vocab.to_index(word)
for word in lowered_word_lst + lowered_no_create_word_lst
]])
lowered_words = lowered_embed(lowered_words)
all_words = word_lst + no_create_word_lst
for idx, (word_i, word_j) in enumerate(zip(words[0],
lowered_words[0])):
with self.subTest(idx=idx, word=all_words[idx]):
assert torch.sum(word_i == word_j).eq(lowered_embed.embed_size)
def get_vocab(dataset):
vocabulary = Vocabulary(unknown=unk_str, padding=pad_str)
for data, _ in dataset:
vocabulary.add_word_lst(data)
print('vocab', len(vocabulary))
print('pad', vocabulary.to_index(pad_str))
return vocabulary
def test_Index2WordProcessor(self):
vocab = Vocabulary()
vocab.add_word_lst(["a", "b", "c", "d", "e"])
proc = Index2WordProcessor(vocab, "tag_id", "tag")
data_set = DataSet(
[Instance(tag_id=[np.random.randint(0, 7) for _ in range(32)])])
data_set = proc(data_set)
self.assertTrue("tag" in data_set)
def get_vocabulary(data, min_freq):
# train data -> vocabulary
# alphabet = "abcdefghijklmnopqrstuvwxyz0123456789,;.!?:'\"/\\|_@#$%^&*~`+-=<>()[]{}"
alphabet = "0123456789,"
char_list = [c for c in alphabet]
vocabulary = Vocabulary(padding='<pad>', unknown='<unk>')
vocabulary.add_word_lst(char_list)
vocabulary.build_vocab()
print('vocab size', len(vocabulary), 'pad', vocabulary.padding_idx, 'unk', vocabulary.unknown_idx)
return vocabulary
def prepare_env():
vocab = Vocabulary().add_word_lst("This is a test .".split())
vocab.add_word_lst("Another test !".split())
embed = StaticEmbedding(vocab, model_dir_or_name=None, embedding_dim=5)
encoder_output = torch.randn(2, 3, 10)
src_seq_len = torch.LongTensor([3, 2])
encoder_mask = seq_len_to_mask(src_seq_len)
return embed, encoder_output, encoder_mask
def get_vocabulary(data, min_freq):
# train data -> vocabulary
vocabulary = Vocabulary(min_freq=min_freq, padding='<pad>', unknown='<unk>')
for filename in data:
for value in data[filename]:
for word_list in data[filename][value]['data']:
vocabulary.add_word_lst(word_list)
vocabulary.build_vocab()
print('vocab size', len(vocabulary), 'pad', vocabulary.padding_idx, 'unk', vocabulary.unknown_idx)
return vocabulary
def prepare_env():
vocab = Vocabulary().add_word_lst("This is a test .".split())
vocab.add_word_lst("Another test !".split())
embed = StaticEmbedding(vocab, model_dir_or_name=None, embedding_dim=5)
src_words_idx = torch.LongTensor([[3, 1, 2], [1, 2, 0]])
tgt_words_idx = torch.LongTensor([[1, 2, 3, 4], [2, 3, 0, 0]])
src_seq_len = torch.LongTensor([3, 2])
tgt_seq_len = torch.LongTensor([4, 2])
return embed, src_words_idx, tgt_words_idx, src_seq_len, tgt_seq_len
def load_dataset(
data_dir='/remote-home/ygxu/workspace/Product_all',
data_path='mr.task.train',
# bert_dir='/home/ygxu/BERT/BERT_English_uncased_L-12_H-768_A_12',
bert_dir='/remote-home/ygxu/workspace/BERT/BERT_English_uncased_L-24_H-1024_A_16',
):
path = os.path.join(data_dir, data_path)
ds = DataSet.read_csv(path, headers=('label', 'raw_sentence'), sep='\t')
ds.apply(lambda x: x['raw_sentence'].lower(),
new_field_name='raw_sentence')
ds.apply(lambda x: int(x['label']),
new_field_name='label_seq',
is_target=True)
def transfer_bert_to_fastnlp(ins):
result = "[CLS] "
bert_text = ins['bert_tokenize_list']
for text in bert_text:
result += text + " "
return result.strip()
with open(os.path.join(bert_dir, 'vocab.txt')) as f:
lines = f.readlines()
vocabs = []
for line in lines:
vocabs.append(line[:-1])
vocab_bert = Vocabulary(unknown=None, padding=None)
vocab_bert.add_word_lst(vocabs)
vocab_bert.build_vocab()
vocab_bert.unknown = '[UNK]'
vocab_bert.padding = '[PAD]'
from pytorch_pretrained_bert import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained(
os.path.join(bert_dir, 'vocab.txt'))
ds.apply(lambda x: tokenizer.tokenize(x['raw_sentence']),
new_field_name='bert_tokenize_list')
ds.apply(transfer_bert_to_fastnlp, new_field_name='bert_tokenize')
ds.apply(lambda x:
[vocab_bert.to_index(word) for word in x['bert_tokenize_list']],
new_field_name='index_words',
is_input=True)
ds.rename_field('index_words', 'tokens')
ds.apply(lambda x: [1.] * len(x['tokens']),
new_field_name='masks',
is_input=True)
return ds
def read_vocab(file_name):
# 读入vocab文件
with open(file_name) as f:
lines = f.readlines()
vocabs = []
for line in lines:
vocabs.append(line.strip())
# 实例化Vocabulary
vocab = Vocabulary(unknown='<unk>', padding='<pad>')
# 将vocabs列表加入Vocabulary
vocab.add_word_lst(vocabs)
# 构建词表
vocab.build_vocab()
return vocab
def test_rebuild(self):
# 测试build之后新加入词，原来的词顺序不变
vocab = Vocabulary()
text = [str(idx) for idx in range(10)]
vocab.update(text)
for i in text:
self.assertEqual(int(i) + 2, vocab.to_index(i))
indexes = []
for word, index in vocab:
indexes.append((word, index))
vocab.add_word_lst([str(idx) for idx in range(10, 13)])
for idx, pair in enumerate(indexes):
self.assertEqual(pair[1], vocab.to_index(pair[0]))
for i in range(13):
self.assertEqual(int(i) + 2, vocab.to_index(str(i)))
def test_fit(self):
"""文本编码.
"""
print('{} test_fit {}'.format('-' * 15, '-' * 15))
texts = ['温都尔站', '东乌广厦', '国电四郎', '阿尔善站', '朱日和基']
vocab = Vocabulary()
for text in texts:
vocab.add_word_lst(list(text))
print(len(vocab))
embed = StaticEmbedding(
vocab, model_dir_or_name='./data/cn_char_fastnlp_100d.txt')
texts_to_id = [[vocab.to_index(word) for word in list(text)]
for text in ['朱日和', '东台变']]
print(texts_to_id) # [[16, 17, 18], [6, 1, 1]]
words = torch.LongTensor(texts_to_id) # 将文本转为index
print(embed(words).size()) # torch.Size([2, 3, 100])
def get_vocab(trainset, testset):
# 构建vocab以及word2idx
#tok
tok_vocab = Vocabulary()
tok_vocab.from_dataset(trainset,
field_name="tok",
no_create_entry_dataset=testset)
tok_vocab.index_dataset(trainset,
testset,
field_name="tok",
new_field_name="chars")
tok_vocab.index_dataset(trainset,
testset,
field_name="asp",
new_field_name="aspect")
# deprel
dep_vocab = Vocabulary()
dep_vocab.from_dataset(trainset, field_name="deprel")
dep_vocab.index_dataset(trainset,
testset,
field_name="deprel",
new_field_name="depidx")
# pol(target)
pol_vocab = Vocabulary(padding=None, unknown=None)
pol_vocab.from_dataset(trainset, field_name="pol")
pol_vocab.index_dataset(trainset,
testset,
field_name="pol",
new_field_name="target")
# pos
pos_vocab = Vocabulary()
pos_vocab.from_dataset(trainset, field_name="pos")
pos_vocab.index_dataset(trainset,
testset,
field_name="pos",
new_field_name="posidx")
# post
max_len = max(max(trainset["seq_len"]), max(testset["seq_len"]))
post_vocab = Vocabulary()
post_vocab.add_word_lst(list(range(-max_len, max_len)))
post_vocab.index_dataset(trainset,
testset,
field_name="post",
new_field_name="postidx")
return tok_vocab, pos_vocab, post_vocab, trainset, testset
def handle_data(n_class):
train_data = get_text_classification_datasets(n_class)
dataset = DataSet()
vocab = Vocabulary(min_freq=0, unknown='<unk>', padding='<pad>')
for i in range(len(train_data.data)):
ans = remove_punc(train_data.data[i])
dataset.append((Instance(content=ans,
target=int(train_data.target[i]))))
dataset.apply(lambda x: x['content'].lower().split(),
new_field_name='words',
is_input=True)
for txt in dataset:
vocab.add_word_lst(txt['words'])
vocab.build_vocab()
# index句子, Vocabulary.to_index(word)
dataset.apply(lambda x: [vocab.to_index(word) for word in x['words']],
new_field_name='index')
dataset.set_input("index")
dataset.set_target("target")
tra, dev = dataset.split(0.2)
return tra, dev, len(vocab)
def prepare_env():
vocab = Vocabulary().add_word_lst("This is a test .".split())
vocab.add_word_lst("Another test !".split())
embed = StaticEmbedding(vocab, model_dir_or_name=None, embedding_dim=5)
src_words_idx = [[3, 1, 2], [1, 2]]
# tgt_words_idx = [[1, 2, 3, 4], [2, 3]]
src_seq_len = [3, 2]
# tgt_seq_len = [4, 2]
ds = DataSet({
'src_tokens': src_words_idx,
'src_seq_len': src_seq_len,
'tgt_tokens': src_words_idx,
'tgt_seq_len': src_seq_len
})
ds.set_input('src_tokens', 'tgt_tokens', 'src_seq_len')
ds.set_target('tgt_seq_len', 'tgt_tokens')
return embed, ds
def get_data(filepath):
data = np.load(filepath, allow_pickle=True)
data, _, ix2word = data['data'], data['word2ix'].item(
), data['ix2word'].item()
wordlist = []
for d in data:
for ix in d:
wordlist.append(ix2word[ix])
vocab = Vocabulary(min_freq=10, padding="</s>")
vocab.add_word_lst(wordlist)
vocab.build_vocab()
# vocab = Vocabulary(min_freq=10, padding="</s>").add_word_lst(wordlist).build_vocab()
vocab_size = len(vocab.word2idx)
for d in data:
for i in range(len(d)):
# d[i] = vocab[vocab.to_word(d[i])]
if d[i] >= vocab_size:
d[i] = vocab["<unk>"]
print(vocab_size)
return data, vocab
def test1(self):
# 测试能否正常打印
from fastNLP import Vocabulary
from fastNLP.core.utils import ConfusionMatrix
import numpy as np
vocab = Vocabulary(unknown=None, padding=None)
vocab.add_word_lst(list('abcdef'))
confusion_matrix = ConfusionMatrix(vocab)
for _ in range(3):
length = np.random.randint(1, 5)
pred = np.random.randint(0, 3, size=(length,))
target = np.random.randint(0, 3, size=(length,))
confusion_matrix.add_pred_target(pred, target)
print(confusion_matrix)
# 测试print_ratio
confusion_matrix = ConfusionMatrix(vocab, print_ratio=True)
for _ in range(3):
length = np.random.randint(1, 5)
pred = np.random.randint(0, 3, size=(length,))
target = np.random.randint(0, 3, size=(length,))
confusion_matrix.add_pred_target(pred, target)
print(confusion_matrix)
def test_no_entry(self):
# 先建立vocabulary，然后变化no_create_entry, 测试能否正确识别
text = [
"FastNLP", "works", "well", "in", "most", "cases", "and", "scales",
"well", "in", "works", "well", "in", "most", "cases", "scales",
"well"
]
vocab = Vocabulary()
vocab.add_word_lst(text)
self.assertFalse(vocab._is_word_no_create_entry('FastNLP'))
vocab.add_word('FastNLP', no_create_entry=True)
self.assertFalse(vocab._is_word_no_create_entry('FastNLP'))
vocab.add_word('fastnlp', no_create_entry=True)
self.assertTrue(vocab._is_word_no_create_entry('fastnlp'))
vocab.add_word('fastnlp', no_create_entry=False)
self.assertFalse(vocab._is_word_no_create_entry('fastnlp'))
vocab.add_word_lst(['1'] * 10, no_create_entry=True)
self.assertTrue(vocab._is_word_no_create_entry('1'))
vocab.add_word('1')
self.assertFalse(vocab._is_word_no_create_entry('1'))
def equip_chinese_ner_with_skip(datasets, vocabs, embeddings, w_list, word_embedding_path=None,
word_min_freq=1, only_train_min_freq=0):
from utils_ import Trie, get_skip_path
from functools import partial
w_trie = Trie()
for w in w_list:
w_trie.insert(w)
# for k,v in datasets.items():
# v.apply_field(partial(get_skip_path,w_trie=w_trie),'chars','skips')
def skips2skips_l2r(chars, w_trie):
'''
:param lexicons: list[[int,int,str]]
:return: skips_l2r
'''
# print(lexicons)
# print('******')
lexicons = get_skip_path(chars, w_trie=w_trie)
# max_len = max(list(map(lambda x:max(x[:2]),lexicons)))+1 if len(lexicons) != 0 else 0
result = [[] for _ in range(len(chars))]
for lex in lexicons:
s = lex[0]
e = lex[1]
w = lex[2]
result[e].append([s, w])
return result
def skips2skips_r2l(chars, w_trie):
'''
:param lexicons: list[[int,int,str]]
:return: skips_l2r
'''
# print(lexicons)
# print('******')
lexicons = get_skip_path(chars, w_trie=w_trie)
# max_len = max(list(map(lambda x:max(x[:2]),lexicons)))+1 if len(lexicons) != 0 else 0
result = [[] for _ in range(len(chars))]
for lex in lexicons:
s = lex[0]
e = lex[1]
w = lex[2]
result[s].append([e, w])
return result
for k, v in datasets.items():
v.apply_field(partial(skips2skips_l2r, w_trie=w_trie), 'chars', 'skips_l2r')
for k, v in datasets.items():
v.apply_field(partial(skips2skips_r2l, w_trie=w_trie), 'chars', 'skips_r2l')
# print(v['skips_l2r'][0])
word_vocab = Vocabulary()
word_vocab.add_word_lst(w_list)
vocabs['word'] = word_vocab
for k, v in datasets.items():
v.apply_field(lambda x: [list(map(lambda x: x[0], p)) for p in x], 'skips_l2r', 'skips_l2r_source')
v.apply_field(lambda x: [list(map(lambda x: x[1], p)) for p in x], 'skips_l2r', 'skips_l2r_word')
for k, v in datasets.items():
v.apply_field(lambda x: [list(map(lambda x: x[0], p)) for p in x], 'skips_r2l', 'skips_r2l_source')
v.apply_field(lambda x: [list(map(lambda x: x[1], p)) for p in x], 'skips_r2l', 'skips_r2l_word')
for k, v in datasets.items():
v.apply_field(lambda x: list(map(len, x)), 'skips_l2r_word', 'lexicon_count')
v.apply_field(lambda x:
list(map(lambda y:
list(map(lambda z: word_vocab.to_index(z), y)), x)),
'skips_l2r_word', new_field_name='skips_l2r_word')
v.apply_field(lambda x: list(map(len, x)), 'skips_r2l_word', 'lexicon_count_back')
v.apply_field(lambda x:
list(map(lambda y:
list(map(lambda z: word_vocab.to_index(z), y)), x)),
'skips_r2l_word', new_field_name='skips_r2l_word')
if word_embedding_path is not None:
word_embedding = StaticEmbedding(word_vocab, word_embedding_path, word_dropout=0)
embeddings['word'] = word_embedding
vocabs['char'].index_dataset(datasets['train'], datasets['dev'], datasets['test'],
field_name='chars', new_field_name='chars')
vocabs['bigram'].index_dataset(datasets['train'], datasets['dev'], datasets['test'],
field_name='bigrams', new_field_name='bigrams')
vocabs['label'].index_dataset(datasets['train'], datasets['dev'], datasets['test'],
field_name='target', new_field_name='target')
return datasets, vocabs, embeddings
def process(self, paths, config, load_vocab_file=True):
"""
:param paths: dict path for each dataset
:param load_vocab_file: bool build vocab (False) or load vocab (True)
:return: DataBundle
datasets: dict keys correspond to the paths dict
vocabs: dict key: vocab(if "train" in paths), domain(if domain=True), tag(if tag=True)
embeddings: optional
"""
vocab_size = config.vocab_size
def _merge_abstracts(abstracts):
merged = []
for abstract in abstracts:
merged.extend(abstract[:self.max_concat_len] + [SEP])
if len(abstracts) == 0:
assert merged == []
return merged[:-1]
def _pad_graph_inputs(graph_inputs):
pad_text_wd = []
max_len = config.max_graph_enc_steps
for graph_input in graph_inputs:
if len(graph_input) < max_len:
pad_num = max_len - len(graph_input)
graph_input.extend([PAD_TOKEN] * pad_num)
else:
graph_input = graph_input[:max_len]
pad_text_wd.append(graph_input)
if len(pad_text_wd) == 0:
pad_text_wd.append([PAD_TOKEN] * max_len)
return pad_text_wd
def _get_nbr_input_len(input_wd):
enc_len = [
min(len(text), config.max_graph_enc_steps) for text in input_wd
]
if len(enc_len) == 0:
enc_len = [0]
return enc_len
def _pad_article(text_wd):
token_num = len(text_wd)
max_len = config.max_enc_steps
if config.neighbor_process == "sep":
max_len += self.max_concat_len * self.max_concat_num
if token_num < max_len:
padding = [PAD_TOKEN] * (max_len - token_num)
article = text_wd + padding
else:
article = text_wd[:max_len]
return article
def _split_list(input_list):
return [text.split() for text in input_list]
def sent_tokenize(abstract):
abs_list = abstract.split(".")
return [(abst + ".") for abst in abs_list[:-1]]
def _article_token_mask(text_wd):
max_enc_len = config.max_enc_steps
if config.neighbor_process == "sep":
max_enc_len += self.max_concat_len * self.max_concat_num
token_num = len(text_wd)
if token_num < max_enc_len:
mask = [1] * token_num + [0] * (max_enc_len - token_num)
else:
mask = [1] * max_enc_len
return mask
def generate_article_input(text, abstracts):
if config.neighbor_process == "sep":
text_wd = text.split()[:config.max_enc_steps]
text_wd.append(SEP)
abstracts_wd = _merge_abstracts(abstracts)
return text_wd + abstracts_wd
else:
return text.split()
def generate_graph_inputs(graph_struct):
graph_inputs_ = [
graph_strut_dict[pid][config.graph_input_type]
for pid in graph_struct
]
return _split_list(graph_inputs_[1:])
def generate_graph_structs(paper_id):
sub_graph_dict = {}
sub_graph_set = []
n_hop = config.n_hop
max_neighbor_num = config.max_neighbor_num
k_nbrs = _k_hop_neighbor(paper_id, n_hop, max_neighbor_num)
for sub_g in k_nbrs:
sub_graph_set += sub_g
for node in sub_graph_set:
sub_graph_dict[node] = []
for sub_g in k_nbrs:
for centre_node in sub_g:
nbrs = graph_strut_dict[centre_node]['references']
c_nbrs = list(set(nbrs).intersection(sub_graph_set))
sub_graph_dict[centre_node].extend(c_nbrs)
for c_nbr in c_nbrs:
sub_graph_dict[c_nbr].append(centre_node)
# in python 3.6, the first in subgraph dict is source paper
return sub_graph_dict
def _k_hop_neighbor(paper_id, n_hop, max_neighbor):
sub_graph = [[] for _ in range(n_hop + 1)]
level = 0
visited = set()
q = deque()
q.append([paper_id, level])
curr_node_num = 0
while len(q) != 0:
paper_first = q.popleft()
paper_id_first, level_first = paper_first
if level_first > n_hop:
return sub_graph
sub_graph[level_first].append(paper_id_first)
curr_node_num += 1
if curr_node_num > max_neighbor:
return sub_graph
visited.add(paper_id_first)
for pid in graph_strut_dict[paper_id_first]["references"]:
if pid not in visited and pid in graph_strut_dict:
q.append([pid, level_first + 1])
visited.add(pid)
return sub_graph
def generate_dgl_graph(paper_id, graph_struct, nodes_num):
g = dgl.DGLGraph()
assert len(graph_struct) == nodes_num
g.add_nodes(len(graph_struct))
pid2idx = {}
for index, key_node in enumerate(graph_struct):
pid2idx[key_node] = index
assert pid2idx[paper_id] == 0
for index, key_node in enumerate(graph_struct):
neighbor = [pid2idx[node] for node in graph_struct[key_node]]
# add self loop
neighbor.append(index)
key_nodes = [index] * len(neighbor)
g.add_edges(key_nodes, neighbor)
return g
train_ds = None
dataInfo = self.load(paths)
# pop nodes in train graph in inductive setting
if config.mode == "test" and self.setting == "inductive":
dataInfo.datasets.pop("train")
graph_strut_dict = {}
for key, ds in dataInfo.datasets.items():
for ins in ds:
graph_strut_dict[ins["paper_id"]] = ins
logger.info(f"the input graph G_v has {len(graph_strut_dict)} nodes")
for key, ds in dataInfo.datasets.items():
# process summary
ds.apply(lambda x: x['abstract'].split(),
new_field_name='summary_wd')
ds.apply(lambda x: sent_tokenize(x['abstract']),
new_field_name='abstract_sentences')
# generate graph
ds.apply(lambda x: generate_graph_structs(x["paper_id"]),
new_field_name="graph_struct")
ds.apply(lambda x: generate_graph_inputs(x["graph_struct"]),
new_field_name='graph_inputs_wd')
ds.apply(lambda x: len(x["graph_inputs_wd"]) + 1,
new_field_name="nodes_num")
# pad input
ds.apply(lambda x: generate_article_input(x['introduction'], x[
"graph_inputs_wd"]),
new_field_name='input_wd')
ds.apply(lambda x: _article_token_mask(x["input_wd"]),
new_field_name="enc_len_mask")
ds.apply(lambda x: sum(x["enc_len_mask"]),
new_field_name="enc_len")
ds.apply(lambda x: _pad_article(x["input_wd"]),
new_field_name="pad_input_wd")
ds.apply(lambda x: _get_nbr_input_len(x["graph_inputs_wd"]),
new_field_name="nbr_inputs_len")
ds.apply(lambda x: _pad_graph_inputs(x["graph_inputs_wd"]),
new_field_name="pad_graph_inputs_wd")
if key == "train":
train_ds = ds
vocab_dict = {}
if not load_vocab_file:
logger.info("[INFO] Build new vocab from training dataset!")
if train_ds is None:
raise ValueError("Lack train file to build vocabulary!")
vocabs = Vocabulary(max_size=config.vocab_size - 2,
padding=PAD_TOKEN,
unknown=UNKNOWN_TOKEN)
vocabs.from_dataset(train_ds,
field_name=["input_wd", "summary_wd"])
vocabs.add_word(START_DECODING)
vocabs.add_word(STOP_DECODING)
vocab_dict["vocab"] = vocabs
# save vocab
with open(os.path.join(config.train_path, "vocab"),
"w",
encoding="utf8") as f:
for w, idx in vocabs:
f.write(str(w) + "\t" + str(idx) + "\n")
logger.info(
"build new vocab ends.. please reRun the code with load_vocab = True"
)
exit(0)
else:
logger.info("[INFO] Load existing vocab from %s!" %
config.vocab_path)
word_list = []
cnt = 3 # pad and unk
if config.neighbor_process == "sep":
cnt += 1
with open(config.vocab_path, 'r', encoding='utf8') as vocab_f:
for line in vocab_f:
pieces = line.split("\t")
word_list.append(pieces[0])
cnt += 1
if cnt > vocab_size:
break
vocabs = Vocabulary(max_size=vocab_size,
padding=PAD_TOKEN,
unknown=UNKNOWN_TOKEN)
vocabs.add_word_lst(word_list)
vocabs.add(START_DECODING)
vocabs.add(STOP_DECODING)
if config.neighbor_process == "sep":
vocabs.add(SEP)
vocabs.build_vocab()
vocab_dict["vocab"] = vocabs
logger.info(f"vocab size = {len(vocabs)}")
assert len(vocabs) == config.vocab_size
dataInfo.set_vocab(vocabs, "vocab")
for key, dataset in dataInfo.datasets.items():
# do not process the training set in test mode
if config.mode == "test" and key == "train":
continue
data_dict = {
"enc_input": [],
"nbr_inputs": [],
"graph": [],
"dec_input": [],
"target": [],
"dec_len": [],
"article_oovs": [],
"enc_input_extend_vocab": [],
}
logger.info(
f"start construct the input of the model for {key} set, please wait..."
)
for instance in dataset:
graph_inputs = instance["pad_graph_inputs_wd"]
abstract_sentences = instance["summary_wd"]
enc_input = instance["pad_input_wd"]
enc_input, nbr_inputs, dec_input, target, dec_len, article_oovs, enc_input_extend_vocab = \
getting_full_info(enc_input, graph_inputs, abstract_sentences, dataInfo.vocabs['vocab'], config)
graph = generate_dgl_graph(instance["paper_id"],
instance["graph_struct"],
instance["nodes_num"])
data_dict["graph"].append(graph)
data_dict["enc_input"].append(enc_input)
data_dict["nbr_inputs"].append(nbr_inputs)
data_dict["dec_input"].append(dec_input)
data_dict["target"].append(target)
data_dict["dec_len"].append(dec_len)
data_dict["article_oovs"].append(article_oovs)
data_dict["enc_input_extend_vocab"].append(
enc_input_extend_vocab)
dataset.add_field("enc_input", data_dict["enc_input"])
dataset.add_field("nbr_inputs", data_dict["nbr_inputs"])
dataset.add_field("dec_input", data_dict["dec_input"])
dataset.add_field("target", data_dict["target"])
dataset.add_field("dec_len", data_dict["dec_len"])
dataset.add_field("article_oovs", data_dict["article_oovs"])
dataset.add_field("enc_input_extend_vocab",
data_dict["enc_input_extend_vocab"])
dataset.add_field("graph", data_dict["graph"])
dataset.set_ignore_type(
'graph') # without this line, there may be some errors
dataset.set_input("graph")
dataset.set_input("nbr_inputs_len", "nbr_inputs", "enc_len",
"enc_input", "enc_len_mask", "dec_input",
"dec_len", "article_oovs", "nodes_num",
"enc_input_extend_vocab")
dataset.set_target("target", "article_oovs", "abstract_sentences")
dataset.delete_field('graph_inputs_wd')
dataset.delete_field('pad_graph_inputs_wd')
dataset.delete_field('input_wd')
dataset.delete_field('pad_input_wd')
logger.info("------load dataset over---------")
return dataInfo, vocabs
def change_tag(ins):
words = ['[unused14]'] + ins['words'][1:]
return words
for target in target_list:
all_data[target]['POS-ctb9'].apply(change_tag, new_field_name='words')
print(all_data['train']['POS-ctb9'][0]['words'][:1])
for task in all_data['train'].keys():
if task.startswith('Parsing'):
continue
dataset = all_data['train'][task]
for word_lst in dataset['words']:
chars_vocab.add_word_lst(word_lst)
pos_idx = chars_vocab.to_index('[unused14]')
print(pos_idx)
label_vocab['POS'] = Vocabulary().from_dataset(
all_data['train']['POS-ctb9'], field_name='target')
label_vocab['CWS'] = Vocabulary().from_dataset(
all_data['train']['CWS-pku'], field_name='target')
label_vocab['NER'] = Vocabulary().from_dataset(
all_data['train']['NER-msra'], field_name='target')
label_vocab['Parsing'] = torch.load('vocab/parsing_vocab')
label_vocab['pos'] = Vocabulary().from_dataset(
all_data['train']['Parsing-ctb9'], field_name='pos')
for target in target_list:
dev_file = None
test_file = None
for f in v:
# example of f: 'health.dev'
data_type = f.split('.')[1]
if data_type == 'train':
train_file = os.path.join(data_path, f)
elif data_type == 'dev':
dev_file = os.path.join(data_path, f)
elif data_type == 'test':
test_file = os.path.join(data_path, f)
else:
raise ValueError('unknown dataset type')
train_set = read_instances_from_file(train_file)
train_set.add_field('task_id', [task_id] * len(train_set))
train_set.apply(lambda x: vocab.add_word_lst(x['words']))
dev_set = read_instances_from_file(dev_file)
dev_set.add_field('task_id', [task_id] * len(dev_set))
dev_set.apply(lambda x: vocab.add_word_lst(x['words']))
test_set = read_instances_from_file(test_file)
test_set.add_field('task_id', [task_id] * len(test_set))
# test_set.apply(lambda x: vocab.add_word_lst(x['words']))
task = Task(task_id, k, train_set, dev_set, test_set)
task_lst.append(task)
logger.info('Building vocabulary...')
vocab.build_vocab()
logger.info('Finished. Size of vocab: {}.'.format(len(vocab)))
from fastNLP.io import WeiboSenti100kPipe
from fastNLP.embeddings import BertEmbedding
from fastNLP.io.pipe.qa import CMRC2018Loader
from fastNLP.io import CNXNLILoader
from fastNLP.io import WeiboNERLoader
from fastNLP.embeddings import StaticEmbedding
from fastNLP import Vocabulary
if __name__ == "__main__":
# 下载情感分析-分类数据
data_bundle = WeiboSenti100kPipe().process_from_file()
data_bundle.rename_field('chars', 'words')
# 下载bert
embed = BertEmbedding(data_bundle.get_vocab('words'),
model_dir_or_name='cn-wwm',
include_cls_sep=True)
# 问答数据
data_bundle = CMRC2018Loader().load()
# 文本匹配
data_bundle = CNXNLILoader().load()
# NER
data_bundle = WeiboNERLoader().load()
# embedding
vocab = Vocabulary()
vocab.add_word_lst("你 好 .".split())
embed = StaticEmbedding(vocab, model_dir_or_name='cn-sgns-literature-word')
