def train(self):
x_items, train_y = read_message()
# 获取bert字向量
bert = BERTEmbedding(self.bert_place, sequence_length=256)
model = CNNModel(bert)
# 输入模型训练数据 标签 步数
model.fit(x_items,
train_y,
epochs=200,
batch_size=32,
fit_kwargs={'callbacks': [tf_board_callback]})
# 保存模型
model.save("output/classification-model")
model.evaluate(x_items, train_y)
def train(self):
x_train, train_y = self.read_message('../data/西药执业药师/train.txt')
x_dev, dev_y = self.read_message('../data/西药执业药师/test.txt')
x_test, test_y = self.read_message('../data/西药执业药师/dev.txt')
# 获取bert字向量
bert = BERTEmbedding('bert-base-chinese', sequence_length=100)
# 获取词向量
# embedding = WordEmbeddings('sgns.weibo.bigram.bz2', 50)
long_model = CNNModel(bert)
# 输入模型训练数据 标签 步数
long_model.fit(x_train,
train_y,
x_dev,
dev_y,
epochs=20,
batch_size=128,
fit_kwargs={'callbacks': [tf_board_callback]})
# 保存模型
long_model.save("../classification-model")
result = long_model.evaluate(x_test, test_y)
return result
import tqdm
import jieba
from kashgari.tasks.classification import CNNModel
def read_data_file(path):
lines = open(path, 'r', encoding='utf-8').read().splitlines()
x_list = []
y_list = []
for line in tqdm.tqdm(lines):
rows = line.split('\t')
if len(rows) >= 2:
y_list.append(rows[0])
x_list.append(list(jieba.cut('\t'.join(rows[1:]))))
else:
print(rows)
return x_list, y_list
test_x, test_y = read_data_file('cnews/cnews.test.txt')
train_x, train_y = read_data_file('cnews/cnews.train.txt')
val_x, val_y = read_data_file('cnews/cnews.val.txt')
model = CNNModel()
model.fit(train_x, train_y, val_x, val_y, batch_size=128)
result = model.evaluate(test_x, test_y)
model.save('model/kashgari/cnn')
tf_board_callback = keras.callbacks.TensorBoard(log_dir='tf_dir', update_freq=10)
from kashgari.tasks.classification import CNNLSTMModel, CNNModel
save = ModelCheckpoint(
os.path.join('model_dir', 'CNNModel_bert.h5'),
monitor='val_acc',
verbose=1,
save_best_only=True,
mode='auto'
)
early_stopping = EarlyStopping(
monitor='val_acc',
min_delta=0,
patience=8,
verbose=1,
mode='auto'
)
model = CNNModel(embed)
# ------------ build model ------------
model.fit(
train_features, train_labels,
valid_features, valid_labels,
epochs=60,
batch_size=256,
callbacks=[tf_board_callback, save, early_stopping]
)
model.evaluate(test_features, test_labels)
y_list.append(rows[0])
x_list.append(list(jieba.cut('\t'.join(rows[1:]))))
else:
print(rows)
return x_list, y_list
test_x, test_y = read_data_file('cnews/cnews.test.txt')
train_x, train_y = read_data_file('cnews/cnews.train.txt')
val_x, val_y = read_data_file('cnews/cnews.val.txt')
# 初始化 word2vec embedding
import kashgari
# 初始化 word2vec embedding
from kashgari.embeddings import WordEmbedding
from kashgari.tasks.classification import CNNModel
model = CNNModel()
# 初始化 BERT embedding
# from kashgari.embeddings import BERTEmbedding
# # embedding = BERTEmbedding('bert-base-chinese', sequence_length=600)
# #
# # # 使用 embedding 初始化模型
# # from kashgari.tasks.classification import CNNModel
# # model = CNNModel(embedding)
model.fit(train_x, train_y, val_x, val_y, batch_size=128)
model.evaluate(test_x, test_y)
model.save('./model_cnn')
