def retrain():
x_train, y_train = process_file(newfile, word_to_id, cat_to_id,
config.seq_length)
session = tf.Session()
session.run(tf.global_variables_initializer())
saver = tf.train.Saver()
saver.restore(sess=session, save_path=save_path)
batch_size = 8
data_len = len(x_train)
num_batch = int((data_len - 1) / batch_size) + 1
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id,
config.seq_length)
loss_in, acc_in = evaluate(session, x_val, y_val)
print("val loss" + str(loss_in))
print("val acc" + str(acc_in))
print("start to deal with file")
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)
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)
msg = ' Train Loss: {0:>6.2}, Train Acc: {1:>7.2%},' \
+ ' Val Loss: {2:>6.2}, Val Acc: {3:>7.2%}'
print(msg.format(loss_train, acc_train, loss_val, acc_val))
def evaluate(sess, x_, y_):
data_len = len(x_)
batch_eval = batch_iter(x_, y_, 128)
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 evaluate(sess):
"""评估在某一数据上的准确率和损失"""
data_len = len(topics_pad)
batch_eval = batch_iter(val_dir, vocab_dir, topics_pad, config.neg_num, config.seq_length, config.batch_size)
total_loss = 0.0
total_acc = 0.0
for topic_pos_batch, topic_neg_batch, blog_batch in batch_eval:
batch_len = len(topic_pos_batch)
feed_dict = feed_data(topic_pos_batch, topic_neg_batch, blog_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 evaluate(sess, x_, y_):
"""评估在某一数据上的准确率和损失"""
data_len = len(x_)
batch_eval = batch_iter(x_, y_, 128)
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 evaluate(sess, x_, y_):
"""评估在某一数据上的准确率和损失"""
data_len = len(x_)
batch_eval = batch_iter(x_, y_, 64)
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, x_batch.shape[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 evaluate(sess, x, y):
"""评估在某一数据上的准确率和损失"""
data_len = len(x)
batch_eval = batch_iter(x, y, batch_size=128)
total_loss = 0.0
total_acc = 0.0
for batch_x, batch_y in batch_eval:
batch_len = len(batch_x)
feed_dict = feed_data(batch_x, batch_y, keep_prob=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 evaluate(sess, x_, y_):
"""评估在某一数据上的准确率和损失"""
data_len = len(x_)
batch_eval = batch_iter(x_, y_, 128)
total_loss = 0.0
total_acc = 0.0
for x_batch, y_batch in batch_eval:
# print("---------------------",x_batch)
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)
# savefile(batch_len, loss, acc)
total_loss += loss * batch_len
total_acc += acc * batch_len
# print("**********************",total_acc)
return total_loss / data_len, total_acc / data_len
def evaluate(sess, cur, source='tb_test_text'):
"""评估在某一数据上的准确率和损失"""
sql_count = 'select count(*) from {0}'.format(source)
cur.execute(sql_count)
data_len = cur.fetchone()[0]
batch_eval = batch_iter(cur, batch_size=config.batch_size, source=source)
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 evaluate(total_batch, writer, sess, merged_summary_train, x_, y_):
"""评估在某一数据上的准确率和损失"""
data_len = len(x_)
batch_eval = batch_iter(x_, y_, 128)
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)
s = sess.run(merged_summary_train, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
total_loss += loss * batch_len
total_acc += acc * batch_len
writer.add_summary(s, total_batch)
return total_loss / data_len, total_acc / data_len
def evaluate(sess, x_, y_, model, loss, acc):
"""评估在某一数据上的准确率和损失"""
data_len = len(x_)
batch_eval = batch_iter(x_, y_, 128)
total_loss = 0.0
total_acc = 0.0
for x_batch, y_batch in batch_eval:
batch_len = len(x_batch)
feed_dict = {
model.input_x: x_batch,
model.input_y: y_batch,
model.keep_prob: 1.0
}
loss_, acc_ = sess.run([loss, 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 eval(x_val, y_val, model, config):
"""验证model效果"""
model.eval()
corrects, avg_loss = 0, 0
batch_train = batch_iter(x_val, y_val, config.batch_size)
for x_batch, y_batch in batch_train:
x_batch = Variable(torch.LongTensor(x_batch)).cuda()
y_batch = Variable(torch.LongTensor(y_batch)).cuda()
output = model(x_batch)
y_batch = torch.argmax(y_batch, dim=1)
loss = F.cross_entropy(output, y_batch)
avg_loss += loss.data.float()
corrects += (torch.max(output, 1)[1].view(
y_batch.size()).data == y_batch.data).sum()
size = x_val.shape[0]
avg_loss /= size
accuracy = float(corrects) / size
return accuracy
def train(model, file_config):
print(file_config)
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
tensorboard_dir = 'tensorboard/textcnn'
if not os.path.exists(tensorboard_dir):
os.makedirs(tensorboard_dir)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
# my change
config = model.config
# 模型计算
y_, logit = model.inference(model.input_x)
loss = model.loss(logit, model.input_y)
train_op = model.train(loss)
acc = model.accuracy(y_)
# tensorboard定义
tf.summary.scalar("loss", loss)
tf.summary.scalar("accuracy", acc)
merged_summary = tf.summary.merge_all()
# tensorboard文件写入对象
writer = tf.summary.FileWriter(tensorboard_dir)
# 配置 Saver
saver = tf.train.Saver()
if not os.path.exists(file_config.save_path):
os.makedirs(file_config.save_path)
print("Loading training and validation data...")
# 载入训练集与验证集
start_time = time.time()
# 读取训练和测试数据，TODO
x_train, y_train = process_file(file_config.train_path, word_to_id,
cat_to_id, config.seq_length)
x_val, y_val = process_file(file_config.val_path, word_to_id, cat_to_id,
config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# 创建session
session = tf.Session()
session.run(tf.global_variables_initializer())
# 将图写入tensorboard
writer.add_graph(session.graph)
for epoch in range(config.num_epochs):
print('Epoch:', epoch + 1)
# 每一个batch重新定义一个迭代器
batch_train = batch_iter(x_train, y_train, config.batch_size)
for x_batch, y_batch in batch_train:
feed_dict = {
model.input_x: x_batch,
model.input_y: y_batch,
model.keep_prob: config.dropout_keep_prob
}
# tensorboard写入
if total_batch % config.save_per_batch == 0:
# 每多少轮次将训练结果写入tensorboard scalar
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
feed_dict[model.keep_prob] = 1.0
loss_train, acc_train = session.run([loss, acc],
feed_dict=feed_dict)
# TODO
loss_val, acc_val = evaluate(session, x_val, y_val, model,
loss, acc)
if acc_val > best_acc_val:
# 保存最好结果
best_acc_val = acc_val
last_improved = total_batch
saver.save(sess=session, save_path=file_config.save_path)
improved_str = '*'
else:
improved_str = ''
time_dif = get_time_dif(start_time)
msg = 'Iter: {0:>6}, Train Loss: {1:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print(
msg.format(total_batch, loss_train, acc_train, loss_val,
acc_val, time_dif, improved_str))
session.run(train_op, feed_dict=feed_dict) # 运行优化
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
print("No optimization for a long time, auto-stopping...")
flag = True
break # 跳出循环
if flag: # 同上
break
def train():
# 这个是可视化的参数保存处，也就是每次训练的时候我们都可以在这里看参数的边化
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
tensorboard_dir = 'tensorboard/textcnn'
if not os.path.exists(tensorboard_dir):
os.makedirs(tensorboard_dir)
# 摘要/日志信息
tf.summary.scalar("loss", model.loss) # 够保存训练过程以及参数分布图并在tensorboard显示。
tf.summary.scalar("accuracy", model.acc)
# merge_all 可以将所有summary全部保存到磁盘，以便tensorboard显示。如果没有特殊要求，
# 一般用这一句就可一显示训练时的各种信息了。
merged_summary = tf.summary.merge_all()
writer = tf.summary.FileWriter(tensorboard_dir) # 指定一个文件用来保存图
# 配置 Saver
saver = tf.train.Saver() # 也就是我们说的checkpoint存放处,这个是参数存放处，可以继续训练或者保存最好的模型
if not os.path.exists(save_dir):
os.makedirs(save_dir)
print("Loading training and validation data...")
# 载入训练集与验证集
start_time = time.time()
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id,
config.seq_length)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id,
config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# 创建会话session
session = tf.Session()
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
acc_val = 0
for epoch in range(config.num_epochs):
print('Epoch:', epoch + 1)
batch_train = batch_iter(x_train, y_train, config.batch_size)
for x_batch, y_batch in batch_train: # x_batch:64*600; y_batch:64*10
# 将三个数据和标签放在一块，是model的传参
feed_dict = feed_data(x_batch, y_batch, config.dropout_keep_prob)
# print("x_batch is {}".format(x_batch.shape))
if total_batch % config.save_per_batch == 0:
# 每10轮次将训练结果写入tensorboard scalar
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每100轮次输出在训练集和验证集上的性能
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) # todo
if acc_val > best_acc_val:
# 保存最好结果
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:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print(
msg.format(total_batch, loss_train, acc_train, loss_val,
acc_val, time_dif, improved_str))
session.run(model.optim,
feed_dict=feed_dict) # 运行优化 真正开始运行,因为是相互依赖，倒着找的
total_batch += 1
if total_batch - last_improved > require_improvement or acc_val > 0.98:
# 验证集正确率长期不提升，提前结束训练
print("No optimization for a long time, auto-stopping...")
flag = True
break # 跳出循环
if flag: # 同上
break
def train(model, config):
"""training text cnn"""
# 载入训练集和验证集
start_time = time.time()
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id,
config.seq_length)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id,
config.seq_length)
time_df = get_time_dif(start_time)
print("load train and eval data done, time：", time_df)
optimizer = torch.optim.Adam(model.parameters(), lr=config.learning_rate)
start_time = time.time()
steps = 0
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
improved_str = "*"
flag = False
loss_function = nn.CrossEntropyLoss()
for epoch in range(1, config.num_epochs + 1):
print("Epoch:", epoch)
batch_train = batch_iter(x_train, y_train, config.batch_size)
for x_batch, y_batch in batch_train:
x_batch = Variable(torch.LongTensor(x_batch)).cuda()
y_batch = Variable(torch.LongTensor(y_batch)).cuda()
output = model(x_batch)
y_batch = torch.argmax(y_batch, dim=1)
loss = F.cross_entropy(output, y_batch)
optimizer.zero_grad()
loss.backward()
optimizer.step()
steps += 1
if steps % config.print_per_batch == 0:
eval_acc = eval(x_val, y_val, model, config)
corrects = (torch.max(output, 1)[1].view(
y_batch.size()).data == y_batch.data).sum()
train_acc = float(corrects) / config.batch_size
if eval_acc > best_acc_val:
best_acc_val = eval_acc
last_improved = steps
improved_str = "*"
else:
improved_str = ""
time_dif = get_time_dif(start_time)
msg = 'Iter: {0},\tTrain Loss: {1:.3},\tTrain Acc: {2:.3}\tVal Acc: {3:.3}, \tTime: {4} {5}'
print(
msg.format(steps, loss.float(), train_acc, eval_acc,
time_dif, improved_str))
if steps - last_improved > config.require_improvement:
print(
"No optimization for a long time, auto-stopping... pytorch"
)
flag = True
break # 跳出循环
if flag:
break
def restore_train():
"""
restore the model and then continue train
:return:
"""
# 载入训练集与验证集
start_time = time.time()
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id,
config.seq_length)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id,
config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
print("Restore model")
session = tf.Session()
session.run(tf.global_variables_initializer())
saver = tf.train.Saver()
saver.restore(sess=session, save_path=save_path)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
for epoch in range(config.num_epochs):
print('Epoch:', 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:
# # 每多少轮次将训练结果写入tensorboard scalar
# s = session.run(merged_summary, feed_dict=feed_dict)
# writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
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) # todo
if acc_val > best_acc_val:
# 保存最好结果
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:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print(
msg.format(total_batch, loss_train, acc_train, loss_val,
acc_val, time_dif, improved_str))
session.run(model.optim, feed_dict=feed_dict) # 运行优化
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
print("No optimization for a long time, auto-stopping...")
flag = True
break # 跳出循环
if flag: # 同上
break
def train():
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
tensorboard_dir = 'tensorboard/textcnn'
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)
# 创建session
session = tf.Session()
session.run(tf.global_variables_initializer())
# saver.restore(sess=session, save_path=save_path)
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
for epoch in range(config.num_epochs):
print('Epoch:', epoch + 1)
batch_train = batch_iter(train_dir, vocab_dir, topics_pad, config.neg_num, config.seq_length, config.batch_size)
for topic_pos_batch, topic_neg_batch, blog_batch in batch_train:
feed_dict = feed_data(topic_pos_batch, topic_neg_batch, blog_batch, config.dropout_keep_prob)
if total_batch % config.save_per_batch == 0:
# 每多少轮次将训练结果写入tensorboard scalar
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
feed_dict[model.keep_prob] = 1.0
a = session.run(model.test, feed_dict=feed_dict)
loss_train, acc_train = session.run([model.loss, model.acc], feed_dict=feed_dict)
loss_val, acc_val = evaluate(session)
if acc_val > best_acc_val:
# 保存最好结果
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:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print(msg.format(total_batch, loss_train, acc_train, loss_val, acc_val, time_dif, improved_str))
session.run(model.optim, feed_dict=feed_dict) # 运行优化
total_batch += 1
def test():
print("Loading test data...")
x_test, y_test = process_file(test_dir, word_to_id, cat_to_id, config.seq_length)
session = tf.Session()
session.run(tf.global_variables_initializer())
meta_graph_def = sm.loader.load(session, tags=[sm.tag_constants.TRAINING], export_dir=pd_path)
signature = meta_graph_def.signature_def
x_tensor_name = signature[signature_key].inputs['input_x'].name
y_tensor_name = signature[signature_key].inputs['input_y'].name
kp_tensor_name = signature[signature_key].inputs['keep_prob'].name
out_tensor_name = signature[signature_key].outputs['output'].name
acc_tensor_name = signature[signature_key].outputs['acc'].name
loss_tensor_name = signature[signature_key].outputs['loss'].name
x = session.graph.get_tensor_by_name(x_tensor_name)
y = session.graph.get_tensor_by_name(y_tensor_name)
kp = session.graph.get_tensor_by_name(kp_tensor_name)
out = session.graph.get_tensor_by_name(out_tensor_name)
acc = session.graph.get_tensor_by_name(acc_tensor_name)
loss = session.graph.get_tensor_by_name(loss_tensor_name)
print(out_tensor_name)
print('Testing...')
data_len = len(x_test)
batch_eval = batch_iter(x_test, y_test, 128)
total_loss = 0.0
total_acc = 0.0
for x_batch, y_batch in batch_eval:
batch_len = len(x_batch)
feed_dict = {x: x_batch, y: y_batch, kp: 1.0}
loss_, acc_ = session.run([loss, acc], feed_dict=feed_dict)
total_loss += loss_ * batch_len
total_acc += acc_ * batch_len
loss_test, acc_test = total_loss / data_len, total_acc / data_len
msg = 'Test Loss: {0:>6.2}, Test Acc: {1:>7.2%}'
print(msg.format(loss_test, acc_test))
batch_size = 128
data_len = len(x_test)
num_batch = int((data_len - 1) / batch_size) + 1
y_test_cls = np.argmax(y_test, 1)
y_pred_cls = np.zeros(shape=len(x_test), dtype=np.int32) # 保存预测结果
for i in range(num_batch): # 逐批次处理
start_id = i * batch_size
end_id = min((i + 1) * batch_size, data_len)
feed_dict = {
x: x_test[start_id:end_id],
kp: 1.0
}
y_pred_cls[start_id:end_id] = session.run(out, feed_dict=feed_dict)
# 评估
print("Precision, Recall and F1-Score...")
print(metrics.classification_report(y_test_cls, y_pred_cls, target_names=categories))
# 混淆矩阵
print("Confusion Matrix...")
cm = metrics.confusion_matrix(y_test_cls, y_pred_cls)
print(cm)
def train(model,data):
if
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
tensorboard_dir = 'tensorboard/textcnn'
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)
print("Loading training and validation data...")
# 载入训练集与验证集
start_time = time.time()
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id, config.seq_length)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id, config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# 创建session
session = tf.Session()
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
for epoch in range(config.num_epochs):
print('Epoch:', 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:
# 每多少轮次将训练结果写入tensorboard scalar
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
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) # todo
if acc_val > best_acc_val:
# 保存最好结果
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:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print(msg.format(total_batch, loss_train, acc_train, loss_val, acc_val, time_dif, improved_str))
session.run(model.optim, feed_dict=feed_dict) # 运行优化
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
print("No optimization for a long time, auto-stopping...")
flag = True
break # 跳出循环
if flag: # 同上
break
def test():
print("Loading test data...")
start_time = time.time()
x_test, y_test = process_file(test_dir, word_to_id, cat_to_id, config.seq_length)
session = tf.Session()
session.run(tf.global_variables_initializer())
saver = tf.train.Saver()
saver.restore(sess=session, save_path=save_path) # 读取保存的模型
print('Testing...')
loss_test, acc_test = evaluate(session, x_test, y_test)
msg = 'Test Loss: {0:>6.2}, Test Acc: {1:>7.2%}'
print(msg.format(loss_test, acc_test))
batch_size = 128
data_len = len(x_test)
num_batch = int((data_len - 1) / batch_size) + 1
y_test_cls = np.argmax(y_test, 1)
y_pred_cls = np.zeros(shape=len(x_test), dtype=np.int32) # 保存预测结果
for i in range(num_batch): # 逐批次处理
start_id = i * batch_size
end_id = min((i + 1) * batch_size, data_len)
feed_dict = {
model.input_x: x_test[start_id:end_id],
model.keep_prob: 1.0
}
y_pred_cls[start_id:end_id] = session.run(model.y_pred_cls, feed_dict=feed_dict)
# 评估
print("Precision, Recall and F1-Score...")
print(metrics.classification_report(y_test_cls, y_pred_cls, target_names=categories))
# 混淆矩阵
print("Confusion Matrix...")
cm = metrics.confusion_matrix(y_test_cls, y_pred_cls)
print(cm)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
if __name__ == '__main__':
#if len(sys.argv) != 2 or sys.argv[1] not in ['train', 'test']:
# raise ValueError("""usage: python run_cnn.py [train / test]""")
print('Configuring CNN model...')
config = TCNNConfig()
if not os.path.exists(vocab_dir): # 如果不存在词汇表，重建
build_vocab(train_dir, vocab_dir, config.vocab_size)
categories, cat_to_id = read_category()
words, word_to_id = read_vocab(vocab_dir)
config.vocab_size = len(words)
model = TextCNN(config)
#if sys.argv[1] == 'train':
# train()
#else:
# test()
train()
def train():
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
tensorboard_dir = 'tensorboard/textrnn'
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)
print("Loading training and validation data...")
# 载入训练集与验证集
start_time = time.time()
# x 是id的集合
# 比如 x = [单词1的id, 单词5的id，单词1的id，单词3的id，单词n的id。。。]
# y 是 one-hot 表示的类别
# 比如 y = [0,1,0,0,0,0]
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id,
config.seq_length)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id,
config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# 创建session
session = tf.Session()
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
for epoch in range(config.num_epochs):
print('Epoch:', epoch + 1)
# 将所有的数据进行分批，每一Epoch都会将所有的批次训练完
batch_train = batch_iter(x_train, y_train, config.batch_size)
for x_batch, y_batch in batch_train:
# 逐批进行训练
# 喂养数据给tf的Variable，x y dropout
feed_dict = feed_data(x_batch, y_batch, config.dropout_keep_prob)
if total_batch % config.save_per_batch == 0:
# 每多少轮次将训练结果写入tensorboard scalar
# 记录图表，只需要loss和acc，为什么还需要原始数据？
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
# 验证过程需要保留所有的数据
feed_dict[model.keep_prob] = 1.0
# 在训练集上验证一次
loss_train, acc_train = session.run([model.loss, model.acc],
feed_dict=feed_dict)
# 在测试集上验证一次
loss_val, acc_val = evaluate(session, x_val, y_val) # todo
if acc_val > best_acc_val:
# 保存最好结果
best_acc_val = acc_val
last_improved = total_batch
saver.save(sess=session, save_path=save_path)
# 如果结果相对历史情况，有所提升，则标志 *
improved_str = '*'
else:
improved_str = ''
time_dif = get_time_dif(start_time)
msg = 'Iter: {0:>6}, Train Loss: {1:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print(
msg.format(total_batch, loss_train, acc_train, loss_val,
acc_val, time_dif, improved_str))
# Training
session.run(model.optim, feed_dict=feed_dict) # 运行优化
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
print("No optimization for a long time, auto-stopping...")
flag = True
break # 跳出循环
if flag: # 同上
break
def train():
print("Configuring TensorBoard and Saver...")
tensorboard_dir = 'tensorboard/textcnn'
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)
saver = tf.train.Saver()
if not os.path.exists(save_dir):
os.makedirs(save_dir)
print("Loading training and validation data...")
start_time = time.time()
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id,
config.seq_length)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id,
config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
session = tf.Session()
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0
best_acc_val = 0.0
last_improved = 0
require_improvement = 1000
flag = False
for epoch in range(config.num_epochs):
print('Epoch:', 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:
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
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:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print(
msg.format(total_batch, loss_train, acc_train, loss_val,
acc_val, time_dif, improved_str))
feed_dict[model.keep_prob] = config.dropout_keep_prob
session.run(model.optim, feed_dict=feed_dict)
total_batch += 1
if total_batch - last_improved > require_improvement:
print("No optimization for a long time, auto-stopping...")
flag = True
break
if flag:
break
def train():
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
tensorboard_dir = 'tensorboard/textrnn'
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)
print("Loading training and validation data...")
# 载入训练集与验证集
start_time = time.time()
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id, config.seq_length)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id, config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# 创建session
session = tf.Session()
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
for epoch in range(config.num_epochs):
print('Epoch:', 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:
# 每多少轮次将训练结果写入tensorboard scalar
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
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) # todo
if acc_val > best_acc_val:
# 保存最好结果
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:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print(msg.format(total_batch, loss_train, acc_train, loss_val, acc_val, time_dif, improved_str))
session.run(model.optim, feed_dict=feed_dict) # 运行优化
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
print("No optimization for a long time, auto-stopping...")
flag = True
break # 跳出循环
if flag: # 同上
break
def train():
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
if not os.path.exists(tensorboard_path):
os.makedirs(tensorboard_path)
tf.summary.scalar("loss", model.loss)
tf.summary.scalar("accuracy", model.acc)
merged_summary = tf.summary.merge_all()
writer = tf.summary.FileWriter(tensorboard_path)
# 配置 Saver
saver = tf.train.Saver()
if not os.path.exists(save_path):
os.makedirs(save_path)
print("Loading training data...")
# 载入训练集与验证集
start_time = time.time()
train_x, train_y = process_file(train_path, word_to_id, cat_to_id,
config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
print("Loading validation data...")
start_time = time.time()
val_x, val_y = process_file(val_path, word_to_id, cat_to_id,
config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# 创建session
session = tf.Session()
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_val_acc = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
jump_flag = False
for epoch in range(config.num_epochs):
print('Epoch:', epoch + 1)
batch_train = batch_iter(train_x, train_y, config.batch_size)
for batch_x, batch_y in batch_train:
feed_dict = feed_data(batch_x,
batch_y,
keep_prob=config.dropout_keep_prob,
is_train=True)
session.run(model.optim, feed_dict=feed_dict) # 运行优化
if total_batch % config.save_per_batch == 0:
# 每多少轮次将训练结果写入tensorboard scalar
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
feed_dict[model.keep_prob] = 1.0
train_loss, train_acc = session.run([model.loss, model.acc],
feed_dict=feed_dict)
val_loss, val_acc = evaluate(session, val_x, val_y)
if val_acc > best_val_acc:
# 保存最好结果
best_val_acc = val_acc
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:>8}, Train Loss: {1:>8.2}, Train Acc: {2:>8.2%},' \
+ ' Val Loss: {3:>8.2}, Val Acc: {4:>8.2%}, Time: {5} {6}'
print(
msg.format(total_batch, train_loss, train_acc, val_loss,
val_acc, time_dif, improved_str))
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
print("No optimization for a long time, auto-stopping...")
jump_flag = True
break # 跳出循环
if jump_flag: # 同上
break
def train():
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
tensorboard_dir = 'tensorboard/textcnn'
# 判断文件存放的路径是否存在，不存在则创建
if not os.path.exists(tensorboard_dir):
os.makedirs(tensorboard_dir)
# loss ：损失函数，交叉熵
# acc：准确率
# tf.summary.scalar:用来显示标量信息
# 标量：只有数值大小，没有方向，也叫“无向量”
tf.summary.scalar("loss", model.loss)
tf.summary.scalar("accuracy", model.acc)
# 将所有summary全部保存到磁盘，以便 tensorboard 显示
merged_summary = tf.summary.merge_all()
# 指定一个文件用来保存图。
writer = tf.summary.FileWriter(tensorboard_dir)
# 配置 Saver
# 保存训练结果的对象
# save_dir：保存结果的路径:'checkpoints/textcnn'
saver = tf.train.Saver()
if not os.path.exists(save_dir):
os.makedirs(save_dir)
print("Loading training and validation data...")
# 载入训练集与验证集
# 开始时间为当前时间
start_time = time.time()
# train_dir：训练集的文件：cnews.train.txt
# val_dir：验证集的文件：cnews.val.txt
# word_to_id：词汇表对应的字典
# cat_to_id：目录对应的字典
# config.seq_length：序列长度:600
# process_file：将文件转换为id表示
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id,
config.seq_length)
print("x_train" + x_train)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id,
config.seq_length)
"""获取已使用时间"""
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# 创建session
session = tf.Session()
# 全局变量初始化
session.run(tf.global_variables_initializer())
# 添加图？不是很懂
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
# num_epochs = 10 # 总迭代轮次
for epoch in range(config.num_epochs):
# 因为epoch从0开始
print('Epoch:', epoch + 1)
# batch_size = 64 # 每批训练大小
"""batch_iter:生成批次数据"""
batch_train = batch_iter(x_train, y_train, config.batch_size)
for x_batch, y_batch in batch_train:
# dropout_keep_prob = 0.5 # dropout保留比例
# feed_data：将数据转换成字典
feed_dict = feed_data(x_batch, y_batch, config.dropout_keep_prob)
# save_per_batch = 10 # 每多少轮存入tensorboard
# % 取模，即取余，也就是save_per_batch被total_batch整除时候写入tensorboard scalar
if total_batch % config.save_per_batch == 0:
# 每多少轮次将训练结果写入tensorboard scalar
# 原来的写法s = session.run(merged_summary, feed_dict=feed_dict)
# 可以用以下来代替
s = session.run(merged_summary, feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
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) # todo
# 如果本次计算的结果比之前记录的最佳结果更好，就要更新
# last_improved:记录上一次提升批次
if acc_val > best_acc_val:
# 保存最好结果
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:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print("msg.format" +
msg.format(total_batch, loss_train, acc_train, loss_val,
acc_val, time_dif, improved_str))
session.run(model.optim, feed_dict=feed_dict) # 运行优化
total_batch += 1
# require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
if total_batch - last_improved > require_improvement:
# 验证集正确率长期不提升，提前结束训练
print("No optimization for a long time, auto-stopping...")
flag = True
break # 跳出循环
if flag: # 同上
break
def train():
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
tensorboard_dir = 'tensorboard/textcnn'
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)
print("Loading training and validation data...")
# 载入训练集与验证集
start_time = time.time()
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id,
config.seq_length)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id,
config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# 创建session
session = tf.Session()
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
for epoch in range(config.num_epochs):
print('Epoch:', 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:
# 每多少轮次将训练结果写入tensorboard scalar
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
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) # todo
xx.append(total_batch)
yy_train.append(acc_train)
yy_val.append(acc_val)
yy1.append(loss_train)
yy2.append(loss_val)
if acc_val > best_acc_val:
# 保存最好结果
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:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print(
msg.format(total_batch, loss_train, acc_train, loss_val,
acc_val, time_dif, improved_str))
session.run(model.optim, feed_dict=feed_dict) # 运行优化
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
print("No optimization for a long time, auto-stopping...")
flag = True
break # 跳出循环
if flag: # 同上
break
def train():
tf.get_variable_scope().reuse_variables()
#当前变量作用域可以用tf.get_variable_scope()进行检索并且reuse设置为True .
print("Configuring TensorBoard and Saver...")
# 配置 Tensorboard，重新训练时，请将tensorboard文件夹删除，不然图会覆盖
tensorboard_dir = 'tensorboard/textrnn'
tf.summary.scalar("loss", model.loss) #用来显示标量信息，一般在画loss,accuary时会用到这个函数。
tf.summary.scalar("accuracy", model.acc) #生成准确率标量图
merged_summary = tf.summary.merge_all() #merge_all可以将所有summary全部保存到磁盘，以便tensorboard显示。
writer = tf.summary.FileWriter(tensorboard_dir) #指定一个文件用来保存图。定义一个写入summary的目标文件，dir为写入文件地址
# 配置 Saver
saver = tf.train.Saver()#创建一个Saver对象用来保存模型
print("Loading training and validation data...")
# 载入训练集与验证集
start_time = time.time()
x_train, y_train = process_file(train_dir, word_to_id, cat_to_id, config.seq_length)
x_val, y_val = process_file(val_dir, word_to_id, cat_to_id, config.seq_length)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
# 创建session
session = tf.Session()
session.run(init)
writer.add_graph(session.graph)#写入数据流图
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次提升批次
require_improvement = 1000 # 如果超过1000轮未提升，提前结束训练
flag = False
for epoch in range(config.num_epochs):#轮次
print('Epoch:', epoch + 1)
batch_train = batch_iter(x_train, y_train, config.batch_size)#每批训练大小
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:
# 每多少轮次将训练结果写入tensorboard scalar
s = session.run(merged_summary, feed_dict=feed_dict)
writer.add_summary(s, total_batch)
if total_batch % config.print_per_batch == 0:
# 每多少轮次输出在训练集和验证集上的性能
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) # todo
if acc_val > best_acc_val:
# 保存最好结果
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:>6.2}, Train Acc: {2:>7.2%},' \
+ ' Val Loss: {3:>6.2}, Val Acc: {4:>7.2%}, Time: {5} {6}'
print(msg.format(total_batch, loss_train, acc_train, loss_val, acc_val, time_dif, improved_str))
session.run(model.optim, feed_dict=feed_dict) # 运行优化
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
print("No optimization for a long time, auto-stopping...")
flag = True
break # 跳出循环
if flag: # 同上
break