def evaluate(sess, x_, y_, wv_model=None):
'''评估数据的准确率和损失'''
data_len = len(x_)
batch_eval = batch_iter(x_, y_, wv_model, config.batch_size)
total_loss = 0.0
total_acc = 0.0
for x_batch, y_batch in batch_eval:
batch_len = len(x_batch)
feed_dict = feed_data(x_batch, y_batch, 1.0)
loss, acc = sess.run([model.loss, model.acc], feed_dict=feed_dict)
total_loss += loss * batch_len
total_acc += acc * batch_len
return total_loss / data_len, total_acc / data_len
def train2(restore=False):
logging.info('Configuring TensorBoard and Saver...')
# 配置tensor board
tensorboard_dir = 'text_cnn/tmp'
if not os.path.exists(tensorboard_dir):
os.makedirs(tensorboard_dir)
tf.summary.scalar('loss', model.loss)
tf.summary.scalar('accuracy', model.acc)
merged_summary = tf.summary.merge_all()
writer = tf.summary.FileWriter(tensorboard_dir)
#配置Saver
saver = tf.train.Saver()
if not os.path.exists(save_dir):
os.makedirs(save_dir)
logging.info('Loading training and validation data...')
# 载入训练集与验证集
start_time = time.time()
train_set_data = process_file(train_set_file, config.num_classes,
config.seq_length)
x_train, y_train = train_set_data['train_set']
x_val, y_val = train_set_data['validate_set']
del train_set_data
logging.info('Time usage: {}'.format(get_time_dif(start_time, )))
# 创建session
conf = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True))
session = tf.Session(config=conf)
if restore and os.path.exists(save_dir + "checkpoint"):
logging.info("Restoring Variables from Checkpoint for cnn model.")
saver.restore(session, tf.train.latest_checkpoint(save_dir))
else:
logging.info('first training cnn model, Initializing Variables')
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
logging.info('Training and evaluating...')
start_time = time.time()
best_acc_val = 0.0
flag = False
# wv_model = get_wordvec_model()
for epoch in range(1, config.num_epochs + 1):
logging.info('Epoch: {}'.format(epoch))
batch_train = batch_iter(x_train, y_train, wv_model, config.batch_size)
total_batch = 0
for x_batch, y_batch in batch_train:
feed_dict = feed_data(x_batch, y_batch, config.dropout_keep_prob)
if total_batch % config.print_per_batch == 0:
loss_train, acc_train = session.run([model.loss, model.acc],
feed_dict=feed_dict)
time_dif = get_time_dif(start_time)
logging.info(
'epoch:{4: >3}, Iter: {0:>6}, Train Loss: {1:>6.8}, Train Acc: {2:>7.8%}, Time: {3}'
.format(total_batch, loss_train, acc_train, time_dif,
epoch))
# 运行优化
session.run(model.optim, feed_dict=feed_dict)
total_batch += 1
feed_dict[model.keep_prob] = 1.0
loss_train, acc_train = session.run([model.loss, model.acc],
feed_dict=feed_dict)
loss_val, acc_val = evaluate(session, x_val, y_val, wv_model=wv_model)
if acc_val > best_acc_val:
best_acc_val = acc_val
last_improved = total_batch
saver.save(sess=session, save_path=save_path)
improved_str = '*'
else:
improved_str = ''
time_dif = get_time_dif(start_time)
msg = 'epoch:{0: >3}, Iter: {1:>6}, Train Loss: {2:>6.8}, Train Acc: {3:>7.8%},' \
'Val Loss: {4:>6.8}, Val Acc: {5:>7.8%}, Time: {6} {7}'
logging.info(
msg.format(epoch, total_batch, loss_train, acc_train, loss_val,
acc_val, time_dif, improved_str))
if epoch % config.save_per_epoch == 0:
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
# epoch_name = os.path.join(save_dir, "epoch_{0}".format(epoch))
# saver.save(sess=session, save_path=epoch_name)
session.close()
def train(restore=False):
logging.info('Configuring TensorBoard and Saver...')
# 配置tensor board
tensorboard_dir = 'textcnn/tensorboard'
if not os.path.exists(tensorboard_dir):
os.makedirs(tensorboard_dir)
tf.summary.scalar('loss', model.loss)
tf.summary.scalar('accuracy', model.acc)
merged_summary = tf.summary.merge_all()
writer = tf.summary.FileWriter(tensorboard_dir)
#配置Saver
saver = tf.train.Saver()
if not os.path.exists(save_dir):
os.makedirs(save_dir)
logging.info('Loading training and validation data...')
# 载入训练集与验证集
start_time = time.time()
train_set_data = process_file(train_set_file, config.num_classes, config.seq_length)
x_train, y_train = train_set_data['train_set']
x_val, y_val = train_set_data['validate_set']
logging.info('Time usage: {}'.format(get_time_dif(start_time,)))
random_vector_generate(train_set_file)
# 创建session
conf = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True))
session = tf.Session(config=conf)
if restore and os.path.exists(save_dir+"checkpoint"):
logging.info("Restoring Variables from Checkpoint for cnn model.")
saver.restore(session, tf.train.latest_checkpoint(save_dir))
else:
logging.info('first training cnn model, Initializing Variables')
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
logging.info('Training and evaluating...')
start_time = time.time()
total_batch = 0
best_acc_val = 0.0
last_improved = 0
require_improvement = 10000
flag = False
for epoch in range(config.num_epochs):
logging.info('Epoch: {}'.format(epoch + 1,))
batch_train = batch_iter(x_train, y_train, config.batch_size)
for x_batch, y_batch in batch_train:
feed_dict = feed_data(x_batch, y_batch, config.dropout_keep_prob)
if total_batch % config.save_per_batch == 0:
# 训练结果写入tensorboard轮数
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 输出训练集和验证集性能轮数
feed_dict[model.keep_prob] = 1.0
loss_train, acc_train = session.run([model.loss, model.acc], feed_dict=feed_dict)
time_dif = get_time_dif(start_time)
msg = 'Iter: {0:>6}, Train Loss: {1:>6.4}, Train Acc: {2:>8.4%}, ' \
'Time: {3:>8.4}'
logging.info(msg.format(total_batch, loss_train, acc_train, time_dif))
# 运行优化
session.run(model.optim, feed_dict=feed_dict)
total_batch += 1
# if total_batch - last_improved > require_improvement:
# # 验证集正确率长期不提升，提前结束训练
# logging.info('No optimization for a long time, auto-stopping...')
# flag = True
# break
# 输出训练集和验证集性能轮数
feed_dict[model.keep_prob] = 1.0
loss_train, acc_train = session.run([model.loss, model.acc], feed_dict=feed_dict)
loss_val, acc_val = evaluate(session, x_val, y_val)
if acc_val > best_acc_val:
best_acc_val = acc_val
last_improved = total_batch
saver.save(sess=session, save_path=save_path)
improved_str = '*'
else:
improved_str = ''
time_dif = get_time_dif(start_time)
msg = 'Iter: {0:>6}, Train Loss: {1:>6.4}, Train Acc: {2:>7.4%},' \
'Val Loss: {3:>6.4}, Val Acc: {4:>7.4%}, Time: {5} {6}'
logging.info(msg.format(total_batch, loss_train, acc_train, loss_val, acc_val,
time_dif, improved_str))
if flag:
break
