def train(args): train_iter, dev_iter = data_processor.load_data(args) # 将数据分为训练集和验证集 print('加载数据完成') model = TextRNN(args) Cuda = torch.cuda.is_available() if Cuda and args.cuda: model.cuda() """ Q5: Please give optimizer here Add lr_scheduler to adjust learning rate. """ optimizer = torch.optim.Adam(model.parameters(), lr = args.lr) scheduler = torch.optim.lr_scheduler.ExponentialLR(optimizer, 0.8) steps = 0 best_acc = 0 last_step = 0 model.train() for epoch in range(1, args.epoch + 1): for batch in train_iter: feature, target = batch.text, batch.label # t_()函数表示将(max_len, batch_size)转置为(batch_size, max_len) with torch.no_grad(): feature.t_(), target.sub_(1) # target减去1 if args.cuda and Cuda: feature, target = feature.cuda(), target.cuda() optimizer.zero_grad() logits = model(feature) loss = F.cross_entropy(logits, target) loss.backward() optimizer.step() steps += 1 if steps % args.log_interval == 0: # torch.max(logits, 1)函数:返回每一行中最大值的那个元素,且返回其索引(返回最大元素在这一行的列索引) corrects = (torch.max(logits, 1)[1] == target).sum() train_acc = 100.0 * corrects / batch.batch_size sys.stdout.write( '\rBatch[{}] - loss: {:.6f} acc: {:.4f}%({}/{})'.format(steps, loss.item(), train_acc, corrects, batch.batch_size)) if steps % args.test_interval == 0: dev_acc = eval(dev_iter, model, args) if dev_acc > best_acc: best_acc = dev_acc last_step = steps if args.save_best: print('Saving best model, acc: {:.4f}%\n'.format(best_acc)) save(model, args.save_dir, 'best', steps) else: scheduler.step() print('lr decayed to {}'.format(optimizer.state_dict()['param_groups'][0]['lr'])) if steps - last_step >= args.early_stopping: print('\nearly stop by {} steps, acc: {:.4f}%'.format(args.early_stopping, best_acc)) raise KeyboardInterrupt
def train(args): train_iter, dev_iter = data_processor.load_data(args) # 将数据分为训练集和验证集 print('加载数据完成') model = TextRNN(args) if args.cuda: model.cuda() """ Q5: Please give optimizer here """ optimizer = torch.optim.Adam(model.parameters()) steps = 0 best_acc = 0 last_step = 0 model.train() for epoch in range(1, args.epoch + 1): for batch in train_iter: feature, target = batch.text, batch.label # t_()函数表示将(max_len, batch_size)转置为(batch_size, max_len) with torch.no_grad(): #feature.t_() target.sub_(1) # target减去1 #print(feature.shape) if args.cuda: feature, target = feature.cuda(), target.cuda() optimizer.zero_grad() logits = model(feature) #print(logits.shape) loss = F.cross_entropy(logits, target) loss.backward() torch.nn.utils.clip_grad_norm_(model.parameters(), 1) optimizer.step() steps += 1 if steps % args.log_interval == 0: # torch.max(logits, 1)函数:返回每一行中最大值的那个元素,且返回其索引(返回最大元素在这一行的列索引) corrects = (torch.max(logits, 1)[1] == target).sum() train_acc = 100.0 * corrects / batch.batch_size sys.stdout.write( '\rBatch[{}] - loss: {:.6f} acc: {:.4f}%({}/{})'.format( steps, loss.item(), train_acc, corrects, batch.batch_size)) if steps % args.test_interval == 0: dev_acc = eval(dev_iter, model, args) if dev_acc > best_acc: best_acc = dev_acc last_step = steps if args.save_best: print('Saving best model, acc: {:.4f}%\n'.format( best_acc)) save(model, args.save_dir, 'best', steps) else: if steps - last_step >= args.early_stopping: print('\nearly stop by {} steps, acc: {:.4f}%'.format( args.early_stopping, best_acc)) raise KeyboardInterrupt
def train(): # 配置文件 cf = Config('./config.yaml') # 有GPU用GPU device = torch.device("cuda" if torch.cuda.is_available() else "cpu") # 训练数据 train_data = NewsDataset("./data/cnews_final_train.txt", cf.max_seq_len) train_dataloader = DataLoader(train_data, batch_size=cf.batch_size, shuffle=True) # 测试数据 test_data = NewsDataset("./data/cnews_final_test.txt", cf.max_seq_len) test_dataloader = DataLoader(test_data, batch_size=cf.batch_size, shuffle=True) # 预训练词向量矩阵 embedding_matrix = get_pre_embedding_matrix("./data/final_vectors") # 模型 model = TextRNN(cf, torch.tensor(embedding_matrix)) # 优化器用adam optimizer = Adam(filter(lambda p: p.requires_grad, model.parameters())) # 把模型放到指定设备 model.to(device) # 让模型并行化运算 if torch.cuda.device_count() > 1: model = torch.nn.DataParallel(model) # 训练 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 model.train() for epoch_id in trange(cf.epoch, desc="Epoch"): # for step,batch in enumerate(tqdm(train_dataloader,"batch",total=len(train_dataloader))): for step, batch in enumerate(train_dataloader): label_id = batch['label_id'].squeeze(1).to(device) seq_len = batch["seq_len"].to(device) segment_ids = batch['segment_ids'].to(device) # 将序列按长度降序排列 seq_len, perm_idx = seq_len.sort(0, descending=True) label_id = label_id[perm_idx] segment_ids = segment_ids[perm_idx].transpose(0, 1) loss = model(segment_ids, seq_len, label_id) loss.backward() optimizer.step() optimizer.zero_grad() total_batch += 1 if total_batch % cf.print_per_batch == 0: model.eval() with torch.no_grad(): loss_train, acc_train = model.get_loss_acc( segment_ids, seq_len, label_id) loss_val, acc_val = evaluate(model, test_dataloader, device) if acc_val > best_acc_val: # 保存最好结果 best_acc_val = acc_val last_improved = total_batch torch.save(model.state_dict(), "./output/model.bin") 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)) model.train() if total_batch - last_improved > require_improvement: print("长时间未优化") flag = True break if flag: break
model = TextRNN() # 损失函数:这里用交叉熵 criterion = nn.CrossEntropyLoss() # 优化器 这里用SGD optimizer = optim.Adam(model.parameters(), lr=0.001) # device : GPU or CPU device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") model.to(device) # 训练 for epoch in range(EPOCH): start_time = time.time() for i, data in enumerate(train_loader): model.train() inputs, labels = data inputs, labels = inputs.to(device), labels.to(device) # 前向传播 outputs = model(inputs) # 计算损失函数 loss = criterion(outputs, labels) # 清空上一轮梯度 optimizer.zero_grad() # 反向传播 loss.backward() # 参数更新 optimizer.step() accuracy = torch.mean((torch.argmax(outputs, 1) == labels.data).float()) print('epoch{} loss:{:.4f} acc:{:.4f} time:{:.4f}'.format(