def test(model, config, word_to_id, cat_to_id, id_to_cat):
print("Loading test data_loaders...")
start_time = time.time()
x_test, y_test = process_file(config.test_dir, word_to_id, cat_to_id,
config.seq_length)
with tf.Session() as session:
session.run(tf.global_variables_initializer())
saver = tf.train.Saver()
# 读取保存的模型
saver.restore(sess=session, save_path=config.save_dir)
print('Testing...')
loss_test, acc_test = evaluate(session, model, 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)
# 用于存储预测结果
y_pred_cls = np.zeros(shape=len(x_test), dtype=np.int32)
# 逐批次处理
for i in range(num_batch):
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("PRF:")
print(
metrics.classification_report(y_test_cls,
y_pred_cls,
target_names=id_to_cat))
# 混淆矩阵
print("Confusion Matrix:")
print("Truth \ Prediction")
cm = metrics.confusion_matrix(y_test_cls, y_pred_cls)
print(cm)
time_dif = get_time_dif(start_time)
print("Time usage:", time_dif)
def train():
with open(os.path.join(saves_dir, 'word2id.pkl'), 'rb') as f:
vocab = pickle.load(f)
f.close()
with open(os.path.join(saves_dir, 'id2word.pkl'), 'rb') as f:
id2word = pickle.load(f)
f.close()
with open(os.path.join(saves_dir, 'word_embedding.pkl'), 'rb') as f:
word_embedding = pickle.load(f)
f.close()
train_loader, char_max_len = get_trainloader('./data/processed/train.pkl')
valid_loader = get_validloader('./data/processed/valid.pkl')
# test_loader = get_testloader('./data/processed/play.pkl')
char2idx = json.loads(open('./data/processed/char2idx.js').read())
char_data = {
'char_vocab_size': len(char2idx) + 1,
'char_emb_dim': CHAR_EMB_DIM,
'char_hid_dim': CHAR_HID_DIM,
'char_len': char_max_len
}
print('=========model architecture==========')
model = RNNModel(args.model,
len(vocab),
args.emdim,
args.hidim,
args.nlayers,
args.tied,
args.dropout,
device,
word_embedding,
use_ch=args.use_ch,
use_he=args.use_he,
use_i=args.use_i,
use_h=args.use_h,
use_g=args.use_g,
**char_data).to(device)
print(model)
print('=============== end =================')
loss_fn = nn.CrossEntropyLoss(ignore_index=0)
parameters = filter(lambda p: p.requires_grad, model.parameters())
optimizer = torch.optim.Adam(parameters, args.lr, weight_decay=args.wdecay)
print('Training and evaluating...')
start_time = time.time()
if not os.path.exists(model_save_dir):
os.makedirs(model_save_dir)
best_ppl = 9999999
last_improved = 0
require_improvement = 5
for epoch in range(args.epochs):
model.training = True
total_loss = 0.0
loss_epoch = []
for batch, inp in enumerate(
tqdm(train_loader,
desc='Epoch: %03d' % (epoch + 1),
leave=False)):
data = {
'word':
torch.tensor(inp['word'], dtype=torch.long).to(device),
'seq':
torch.tensor(torch.t(inp['seq']), dtype=torch.long).to(device),
'chars':
torch.tensor(inp['chars'], dtype=torch.long).to(device),
'hnym':
torch.tensor(inp['hnym'], dtype=torch.long).to(device),
'hnym_weights':
torch.tensor(inp['hnym_weights'], dtype=torch.float).to(device)
}
targets = torch.tensor(torch.t(inp['target']),
dtype=torch.long).to(device)
output, hidden = model(data, None)
loss = loss_fn(output, targets.view(-1))
optimizer.zero_grad()
total_loss += loss.item()
loss_epoch.append(loss.item())
loss.backward()
# `clip_grad_norm`
torch.nn.utils.clip_grad_norm_(model.parameters(), args.clip)
optimizer.step()
train_loss = np.mean(loss_epoch)
train_ppl = np.exp(train_loss)
valid_loss, valid_ppl = evaluate(model, valid_loader, device)
# generate(model, test_loader, id2word, epoch + 1, 20)
if valid_ppl < best_ppl:
best_ppl = valid_ppl
last_improved = epoch
torch.save(
model.state_dict(), model_save_dir +
'defseq_model_params_%s_min_ppl.pkl' % (epoch + 1))
improved_str = '*'
else:
torch.save(
model.state_dict(), model_save_dir +
'defseq_model_params_%s_ppl.pkl' % (epoch + 1))
improved_str = ''
time_dif = get_time_dif(start_time)
msg = 'Epoch: {0:>6},Train Loss: {1:>6.6}, Train Ppl: {2:>6.6},' \
+ ' Val loss: {3:>6.6}, Val Ppl: {4:>6.6},Time:{5} {6}'
print(
msg.format(epoch + 1, train_loss, train_ppl, valid_loss, valid_ppl,
time_dif, improved_str))
if epoch - last_improved > require_improvement:
print("No optimization for a long time, auto-stopping...")
break
return 1
def train(model, config, word_to_id, cat_to_id):
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 = config.tensorboard_dir
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(config.save_dir):
os.makedirs(config.save_dir)
print("Loading training and validation data_loaders...")
# 载入训练集与验证集
start_time = time.time()
x_train, y_train = process_file(config.train_dir, word_to_id, cat_to_id,
config.seq_length)
x_val, y_val = process_file(config.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
with tf.Session() as session:
session.run(tf.global_variables_initializer())
writer.add_graph(session.graph)
print('Training and evaluating...')
start_time = time.time()
total_batch = 0 # 已训练的总批次
best_acc_val = 0.0 # 最佳验证集准确率
last_improved = 0 # 记录上一次有提升是第几个批次
require_improvement = 1000 # 如果超过1000个批次未提升，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 = create_feed_dict(model, x_batch, y_batch,
config.dropout_keep_prob)
session.run(model.optim, feed_dict=feed_dict) # 运行优化
total_batch += 1
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, model, x_val, y_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=config.save_dir)
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))
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