def evaluate(sess, x, y, seq_lens, batch_size):
data_len = len(x)
batch_eval = batch_iter(x, y, seq_lens, batch_size)
total_loss = 0.0
total_acc = 0.0
for x_batch, y_batch, seq_lens_batch in batch_eval:
batch_len = len(x_batch)
feed_dict = {
model.content: x_batch,
model.label: y_batch,
model.sequence_lengths: seq_lens_batch
}
loss, acc = sess.run([model.loss, model.accuracy], feed_dict=feed_dict)
total_loss += loss * batch_len
total_acc += acc * batch_len
return total_loss / data_len, total_acc / data_len
def train():
# 配置TensorBoard和Saver
print('Configuring TensorBoard and Saver...')
saver = tf.train.Saver()
if not os.path.exists(save_dir):
os.makedirs(save_dir)
if not os.path.exists(tensorboard_dir):
os.makedirs(tensorboard_dir)
tf.summary.scalar('loss', model.loss)
tf.summary.scalar('accuracy', model.accuracy)
merged_summary = tf.summary.merge_all()
writer = tf.summary.FileWriter(tensorboard_dir)
# 处理数据
print('Loading training data and validation data...')
start_time = time.time()
x_train, y_train, seq_lens_train = process_file(train_dir, word_to_id,
label_to_id,
config.seq_length)
x_val, y_val, seq_lens_val = process_file(val_dir, word_to_id, label_to_id,
config.seq_length)
print('Time usage:', get_time_dif(start_time))
# 创建session
session = tf.Session()
session.run(tf.global_variables_initializer())
# 将图添加到TensorBoard中
writer.add_graph(session.graph)
# 开始训练
print('Start training...')
start_time = time.time()
best_acc_val = 0.0
total_batch = 0
for epoch in range(config.epoch):
batch_train = batch_iter(x_train, y_train, seq_lens_train,
config.batch_size)
for x_batch, y_batch, seq_lens_batch in batch_train:
feed_dict = {
model.content: x_batch,
model.label: y_batch,
model.sequence_lengths: seq_lens_batch
}
# 将训练结果写如到TensorBoard中
if total_batch % config.save_pre_batch == 0:
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
# 输出训练集和验证集的结果，并保存最好的模型
if total_batch % config.print_pre_batch == 0:
loss_train, acc_train = session.run(
[model.loss, model.accuracy], feed_dict=feed_dict)
loss_val, acc_val = evaluate(session, x_val, y_val,
seq_lens_val, config.batch_size)
# 每次只保存最好的模型
if acc_val > best_acc_val:
best_acc_val = acc_val
saver.save(session, save_path)
improved_str = '*'
else:
improved_str = ''
time_dif = get_time_dif(start_time)
msg = 'Iter: {0:>2}, Train Loss: {1:>2.2f}, Train Acc: {2:>2.2%}, ' \
'Val Loss: {3:>2.2f}, Val Acc: {4:>2.2%}, Time: {5} {6}'
print(
msg.format(total_batch, loss_train, acc_train, loss_val,
acc_val, time_dif, improved_str))
session.run(model.optimizer, feed_dict=feed_dict)
total_batch += 1