def __init__(self, out=3, out_src=0, bert_type='bert'):
super(Bert_finetuning_CNN_Entity, self).__init__()
_, self.bert_model = Bert.get_bert(bert_type=bert_type)
self.cnn1 = CNN(inc=768,
Dr=[0.25, 0.25],
nb_filtres=[3],
FC_L_size=[100],
ker_size=[3, 2],
padding=False,
maxpooling_dir=True,
out=100,
out_src=out_src)
self.cnn2 = CNN(inc=768,
Dr=[0.25, 0.25],
nb_filtres=[3],
FC_L_size=[100],
ker_size=[3, 2],
padding=False,
maxpooling_dir=True,
out=100,
out_src=out_src)
self.cnn3 = CNN(inc=768,
Dr=[0.25, 0.25],
nb_filtres=[3],
FC_L_size=[100],
ker_size=[3, 2],
padding=False,
maxpooling_dir=True,
out=100,
out_src=out_src)
self.cnn4 = CNN(inc=768,
Dr=[0.25, 0.25],
nb_filtres=[3],
FC_L_size=[100],
ker_size=[3, 2],
padding=False,
maxpooling_dir=True,
out=100,
out_src=out_src)
self.cnn5 = CNN(inc=768,
Dr=[0.25, 0.25],
nb_filtres=[3],
FC_L_size=[100],
ker_size=[3, 2],
padding=False,
maxpooling_dir=True,
out=100,
out_src=out_src)
self.fc = nn.Sequential(nn.LayerNorm(500),
nn.Dropout(p=0.1, inplace=False),
nn.Linear(500, 100), nn.Hardtanh(),
nn.LayerNorm(100), nn.Linear(100, out),
nn.Hardtanh())
def __init__(self, out=3, out_src=0, bert_type='bert'):
super(Bert_finetuning, self).__init__()
_, self.bert_model = Bert.get_bert(bert_type=bert_type)
self.trensfer = out_src != 0
self.out_layers = nn.Linear(768, out)
self.out_src_layers = nn.Linear(768,
out_src) if self.trensfer else None
def __init__(self, out=3, out_src=0, bert_type='bert'):
super(Bert_finetuning_CNN_residual, self).__init__()
_, self.bert_model = Bert.get_bert(bert_type=bert_type)
self.cnn = CNN(inc=768,
nb_filtres=[3],
FC_L_size=[768],
ker_size=[3, 2],
padding=False,
maxpooling_dir=True,
out=out,
out_src=0,
w_size=768)
def __init__(self, out=11, bert_type='bert'):
super(BertRecNER, self).__init__()
_, self.bert_model = Bert.get_bert(bert_type=bert_type)
#self.level1=SeqClassifier()
#self.level2=SeqClassifier(emb_size=768+out)
#self.level3=SeqClassifier(emb_size=768+out)
#self.level1=nn.Linear(768,out)
#self.level2=nn.Linear(768+out,out)
#self.level3=nn.Linear(768+out,out)
self.level1 = MLP(inputs=768)
self.level2 = MLP(inputs=768 + out)
self.level3 = MLP(inputs=768 + out)
def __init__(self, out=3, out_src=0, bert_type='bert'):
super(Bert_finetuning_Mixed, self).__init__()
_, self.bert_model = Bert.get_bert(bert_type=bert_type)
self.cnn = CNN(inc=768,
Dr=[0.5, 0.5],
nb_filtres=[8],
FC_L_size=[768],
ker_size=[3, 2],
padding=False,
maxpooling_dir=True,
out=out,
out_src=out_src)
self.out_layers = nn.Linear(768, out)
def _train(config):
model = Bert.load_model(config.reload_model_name, CUDA, gpu_ids)
Bert.train(PawsDataSet, model, config)
"""
This module offering word embedding using BERT model
"""
import torch
import Bert
from Parameters import global_param
bert_type = global_param.model_param['bert']
tokenizer, model = Bert.get_bert(bert_type=bert_type)
def Text2tokens(text):
"""
This function tokenize the input text
:param text: input text
:return: tow list of tokens and segmentation respectively
"""
marked_text = "[CLS] " + text + " [SEP]"
tokenized_text = tokenizer.tokenize(marked_text)
indexed_tokens = tokenizer.convert_tokens_to_ids(tokenized_text)
segments_ids = [0] * len(tokenized_text)
return indexed_tokens, segments_ids
def Token2tonsor(token):
"""
This function convert the tuple (list tokens , list segmentation) to tuple of torch tensor
:param token: tuple (list tokens ,list segmentation)