import tensorflow as tf
import numpy as np
import common_utils
import optimization
from models.tf_metrics import precision, recall, f1
from models.utils import dice_dsc_loss
from models.layers.lstm_layers import BLSTM
logger = common_utils.set_logger('NER Training...')
class BILSTMONLY(object):
def __init__(self, params: dict):
self.char_embedding = tf.Variable(np.load(params["embedding_path"]),
dtype=tf.float32,
name='input_char_embedding')
self.word_embedding = tf.Variable(np.load(
params["word_embedding_path"]),
dtype=tf.float32,
name="input_word_embedding")
# 丢弃概率
self.dropout_rate = params["dropout_prob"]
self.num_labels = params["num_labels"]
self.rnn_size = params["rnn_size"]
self.num_layers = params["num_layers"]
self.hidden_units = params["hidden_units"]
def __call__(self,
input_ids=None,
input_word_ids=None,
labels=None,
from models.bert_event_type_classification import bert_classification_model_fn_builder
from data_processing.data_utils import *
from data_processing.event_prepare_data import EventRolePrepareMRC, EventTypeClassificationPrepare
# from data_processing.event_prepare_data import EventRoleClassificationPrepare
from data_processing.event_prepare_data import event_input_bert_mrc_mul_fn, event_index_class_input_bert_fn
from data_processing.event_prepare_data import event_binclass_input_bert_fn
from models.bert_event_type_classification import bert_binaryclassification_model_fn_builder
from data_processing.event_prepare_data import event_input_verfify_mrc_fn
from models.event_verify_av import event_verify_mrc_model_fn_builder
from configs.event_config import event_config
# import horovod.tensorflow as hvd
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
logger = set_logger("[run training]")
# logger = logging.getLogger('train')
# logger.setLevel(logging.INFO)
# os.environ['TF_ENABLE_AUTO_MIXED_PRECISION']='1'
def serving_input_receiver_fn():
"""Serving input_fn that builds features from placeholders
Returns
-------
tf.estimator.export.ServingInputReceiver
"""
words = tf.placeholder(dtype=tf.int32, shape=[None, None], name='words')
nwords = tf.placeholder(dtype=tf.int32, shape=[None], name='text_length')
