def load_from_file(cls,
model_folder,
context: mx.Context = None,
**kwargs):
if context is None:
context = mxnet_prefer_gpu()
config_path = os.path.join(model_folder, 'config.pkl')
with open(config_path, 'rb') as f:
config = pickle.load(f)
with context:
embedding_types = [
WordEmbeddings(
'{}data/embedding/fasttext100.vec.txt'.format(
kwargs.get('word_embedding_path', ''))),
# comment in this line to use character embeddings
# CharacterEmbeddings(),
# comment in these lines to use contextual string embeddings
CharLMEmbeddings('{}data/model/lm-news-forward'.format(
kwargs.get('word_embedding_path', '')),
context=context),
CharLMEmbeddings('{}data/model/lm-news-backward'.format(
kwargs.get('word_embedding_path', '')),
context=context),
]
embeddings = StackedEmbeddings(embeddings=embedding_types)
model = SequenceTagger(hidden_size=config['hidden_size'],
embeddings=embeddings,
tag_dictionary=config['tag_dictionary'],
tag_type=config['tag_type'],
use_crf=config['use_crf'],
use_rnn=config['use_rnn'],
rnn_layers=config['rnn_layers'])
model.load_parameters(os.path.join(model_folder, 'model.bin'),
ctx=context)
return model
train_file='train.tsv',
test_file='test.tsv',
dev_file='dev.tsv',
tag_to_biloes='ner')
# 2. what tag do we want to predict?
tag_type = 'ner'
# 3. make the tag dictionary from the corpus
tag_dictionary = corpus.make_tag_dictionary(tag_type=tag_type)
print(tag_dictionary.idx2item)
# 4. initialize embeddings
with mx.Context(mxnet_prefer_gpu()):
embedding_types = [
WordEmbeddings('data/embedding/glove/glove.6B.100d.txt'),
BERTEmbeddings([
'data/embedding/bert_large_sum/conll03.train.bert',
'data/embedding/bert_large_sum/conll03.dev.bert',
'data/embedding/bert_large_sum/conll03.test.bert'
]),
# comment in this line to use character embeddings
# CharacterEmbeddings(),
# comment in these lines to use contextual string embeddings
CharLMEmbeddings('data/model/lm-news-forward'),
CharLMEmbeddings('data/model/lm-news-backward'),
]
embeddings = StackedEmbeddings(embeddings=embedding_types)
corpus = NLPTaskDataFetcher.fetch_column_corpus('data/ctb5.1-pos',
columns,
train_file='train.short.tsv',
test_file='test.short.tsv',
dev_file='dev.short.tsv')
# 2. what tag do we want to predict?
tag_type = 'pos'
# 3. make the tag dictionary from the corpus
tag_dictionary = corpus.make_tag_dictionary(tag_type=tag_type)
print(tag_dictionary.idx2item)
# 4. initialize embeddings
with mx.Context(mxnet_prefer_gpu()):
embedding_types = [
WordEmbeddings('data/embedding/ctb.pos.fasttext.300.txt'),
# BERTEmbeddings(['data/embedding/bert_base_sum/ctb.pos.train.bert',
# 'data/embedding/bert_base_sum/ctb.pos.dev.bert',
# 'data/embedding/bert_base_sum/ctb.pos.test.bert'])
# CharLMEmbeddings('data/model/lm-news-forward'),
# CharLMEmbeddings('data/model/lm-news-backward'),
]
embeddings = StackedEmbeddings(embeddings=embedding_types)
# 5. initialize sequence tagger
tagger = SequenceTagger(hidden_size=256,
embeddings=embeddings,
tag_dictionary=tag_dictionary,
tag_type=tag_type,
use_crf=True)