def create_model(bert_config,
is_training,
input_ids,
input_mask,
segment_ids,
labels,
num_labels,
use_one_hot_embeddings,
dropout_rate=1.0,
lstm_size=1,
cell='lstm',
num_layers=1):
"""
创建X模型
:param bert_config: bert 配置
:param is_training:
:param input_ids: 数据的idx 表示
:param input_mask:
:param segment_ids:
:param labels: 标签的idx 表示
:param num_labels: 类别数量
:param use_one_hot_embeddings:
:return:
"""
# 使用数据加载BertModel,获取对应的字embedding
import tensorflow as tf
from bert_base.bert import modeling
model = modeling.BertModel(config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
# 获取对应的embedding 输入数据[batch_size, seq_length, embedding_size]
# embedding Tensor("bert/encoder/Reshape_13:0", shape=(56, 202, 768), dtype=float32)
embedding = model.get_sequence_output()
# max_seq_length 202
max_seq_length = embedding.shape[1].value
# 算序列真实长度
# input_ids Tensor("IteratorGetNext:0", shape=(56, 202), dtype=int32)
# tf.abs Tensor("Abs_1:0", shape=(56, 202), dtype=int32)
# used Tensor("Sign:0", shape=(56, 202), dtype=int32)
# lengths Tensor("Sum:0", shape=(56,), dtype=int32) 1行56列（/56行1列？）的数组 每个batch下序列的真实长度
used = tf.sign(tf.abs(input_ids))
lengths = tf.reduce_sum(
used, reduction_indices=1) # [batch_size] 大小的向量，包含了当前batch中的序列长度
# 添加CRF output layer
blstm_crf = BLSTM_CRF(embedded_chars=embedding,
hidden_unit=lstm_size,
cell_type=cell,
num_layers=num_layers,
dropout_rate=dropout_rate,
initializers=initializers,
num_labels=num_labels,
seq_length=max_seq_length,
labels=labels,
lengths=lengths,
is_training=is_training)
rst = blstm_crf.add_blstm_crf_layer(crf_only=False)
return rst
def create_model(bert_config,
is_training,
input_ids,
input_mask,
segment_ids,
labels,
num_labels,
use_one_hot_embeddings,
dropout_rate=1.0,
lstm_size=1,
cell='lstm',
num_layers=1):
"""
创建模型
:param bert_config: bert配置
:param is_training: 是否训练
:param input_ids: 数据的idx表示
:param input_mask:
:param segment_ids:
:param labels: 标签的idx表示
:param num_labels: 类别数量
:param use_one_hot_embeddings:
:param dropout_rate:
:param lstm_size:
:param cell:
:param num_layers:
:return:
"""
# 使用数据加载BertModel，获取对应的字embedding
model = modeling.BertModel(config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
# 获取对应的embedding，输入数据[batch_size, seq_length, embedding_size]
embedding = model.get_sequence_output()
max_seq_length = embedding.shape[1].value
# 序列的真实长度
used = tf.sign(tf.abs(input_ids))
lengths = tf.reduce_sum(
used, reduction_indices=1) # [batch_size]大小的向量，包含了当前的batch中的序列长度
# 添加CRF output layer
blstm_crf = BLSTM_CRF(embedd_chars=embedding,
hidden_unit=lstm_size,
cell_type=cell,
num_layers=num_layers,
dropout_rate=dropout_rate,
initializers=initializers,
num_labels=num_labels,
seq_length=max_seq_length,
labels=labels,
lengths=lengths,
is_training=is_training)
res = blstm_crf.add_blstm_crf_layer(crf_only=True)
return res
def create_model(bert_config,
is_training,
input_ids,
input_mask,
segment_ids,
labels,
num_labels,
use_one_hot_embeddings,
dropout_rate=1.0,
lstm_size=1,
cell='lstm',
num_layers=1):
"""
创建X模型
:param bert_config: bert 配置
:param is_training:
:param input_ids: 数据的idx 表示
:param input_mask:
:param segment_ids:
:param labels: 标签的idx 表示
:param num_labels: 类别数量
:param use_one_hot_embeddings:
:return:
"""
# 使用数据加载BertModel,获取对应的字embedding
import tensorflow as tf
from bert_base.bert import modeling
model = modeling.BertModel(config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
# 获取对应的embedding 输入数据[batch_size, seq_length, embedding_size]
label_vocab = {"O": 0, "B": 1, "I": 2, "X": 3, "[CLS]": 4, "[SEP]": 5}
SPAN_TYPE = "IOB2"
mask = transition_mask(label_vocab, SPAN_TYPE, label_vocab["[CLS]"],
label_vocab["[SEP]"])
embedding = model.get_sequence_output()
max_seq_length = embedding.shape[1].value
# 算序列真实长度
used = tf.sign(tf.abs(input_ids))
lengths = tf.reduce_sum(
used, reduction_indices=1) # [batch_size] 大小的向量，包含了当前batch中的序列长度
# 添加CRF output layer
blstm_crf = BLSTM_CRF(embedded_chars=embedding,
hidden_unit=lstm_size,
cell_type=cell,
num_layers=num_layers,
dropout_rate=dropout_rate,
initializers=initializers,
num_labels=num_labels,
seq_length=max_seq_length,
labels=labels,
lengths=lengths,
is_training=is_training)
rst = blstm_crf.add_blstm_crf_layer(crf_only=True)
return rst
def create_model(bert_config,
is_training,
input_ids,
input_mask,
segment_ids,
labels,
num_labels,
use_one_hot_embeddings,
dropout_rate=1.0,
lstm_size=1,
cell='lstm',
num_layers=1,
crf_only=False,
lstm_only=False):
"""
创建X模型
:param bert_config: bert 配置
:param is_training:
:param input_ids: 数据的idx 表示
:param input_mask:
:param segment_ids:
:param labels: 标签的idx 表示
:param num_labels: 类别数量
:param use_one_hot_embeddings:
:return:
"""
# 使用数据加载BertModel,获取对应的字embedding
import tensorflow as tf
from bert_base.bert import modeling
model = modeling.BertModel(config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
# 获取对应的embedding 输入数据[batch_size, seq_length, embedding_size]
embedding = model.get_sequence_output()
max_seq_length = embedding.shape[1].value
# 算序列真实长度
used = tf.sign(tf.abs(input_ids))
lengths = tf.reduce_sum(
used, reduction_indices=1) # [batch_size] 大小的向量，包含了当前batch中的序列长度
# 添加CRF output layer
with tf.variable_scope('finetune'):
blstm_crf = BLSTM_CRF(embedded_chars=embedding,
input_mask=input_mask,
hidden_unit=lstm_size,
cell_type=cell,
num_layers=num_layers,
dropout_rate=dropout_rate,
initializers=initializers,
num_labels=num_labels,
seq_length=max_seq_length,
labels=labels,
lengths=lengths,
is_training=is_training)
rst = blstm_crf.add_blstm_crf_layer(crf_only=crf_only,
lstm_only=lstm_only)
return rst
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
labels, num_labels, use_one_hot_embeddings):
"""
创建X模型
:param bert_config: bert 配置
:param is_training:
:param input_ids: 数据的idx 表示
:param input_mask:
:param segment_ids:
:param labels: 标签的idx 表示
:param num_labels: 类别数量
:param use_one_hot_embeddings
:return:
"""
# 使用数据加载BertModel,获取对应的字embedding
model = modeling.BertModel(config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
print("获取到bert的输出")
# 获取对应的embedding 输入数据[batch_size, seq_length, embedding_size]
embedding = model.get_sequence_output()
max_seq_length = embedding.shape[1].value
used = tf.sign(tf.abs(input_ids))
lengths = tf.reduce_sum(
used, reduction_indices=1) # [batch_size] 大小的向量，包含了当前batch中的序列长度
# 将embedding作为LSTM结构的输入
# 加载BLSTM-CRF模型对象
blstm_crf = BLSTM_CRF(embedded_chars=embedding,
hidden_unit=FLAGS.lstm_size,
cell_type=FLAGS.cell,
num_layers=FLAGS.num_layers,
dropout_rate=FLAGS.dropout_rate,
initializers=initializers,
num_labels=num_labels,
seq_length=max_seq_length,
labels=labels,
lengths=lengths,
is_training=is_training)
# 获取添加我们自己网络结构后的结果，这些结果包括loss, logits, trans, pred_ids
# 因为BERT里面已经存在双向编码，所以LSTM并不是必须的,可以将BERT最后一层的结构直接扔给CRF进行解码
try:
rst = blstm_crf.add_blstm_crf_layer()
except Exception as e:
print(str(e))
return rst
def create_model(bert_config,
is_training,
input_ids,
input_mask,
segment_ids,
labels,
num_labels,
use_one_hot_embeddings,
pos_ids,
dropout_rate=1.0,
lstm_size=1,
cell='lstm',
num_layers=1,
add_pos_embedding=True):
"""
创建X模型
:param bert_config: bert 配置
:param is_training:
:param input_ids: 数据的idx 表示
:param input_mask:
:param segment_ids:
:param pos_ids: 词性的idx 表示
:param labels: 标签的idx 表示
:param num_labels: 类别数量
:param use_one_hot_embeddings:
:return:
"""
# 使用数据加载BertModel,获取对应的字embedding
import tensorflow as tf
from bert_base.bert import modeling
model = modeling.BertModel(config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
# 获取对应的embedding 输入数据[batch_size, seq_length, embedding_size]
embedding = model.get_sequence_output()
# jzhang add
if add_pos_embedding:
embedding = embedding_addpos(embedding, pos_ids)
max_seq_length = embedding.shape[1].value
# 算序列真实长度
used = tf.sign(tf.abs(input_ids))
lengths = tf.reduce_sum(
used, reduction_indices=1) # [batch_size] 大小的向量，包含了当前batch中的序列长度
# 添加CRF output layer
blstm_crf = BLSTM_CRF(embedded_chars=embedding,
hidden_unit=lstm_size,
cell_type=cell,
num_layers=num_layers,
dropout_rate=dropout_rate,
initializers=initializers,
num_labels=num_labels,
seq_length=max_seq_length,
labels=labels,
lengths=lengths,
is_training=is_training)
# jzhang: 设置参数crf_only=True，控制模型只使用CRF进行解码
# 如果设置crf_only=False，模型将使用BiLSTM-CRF作为Decoder
rst = blstm_crf.add_blstm_crf_layer(crf_only=False)
return rst