完整项目
最后,介绍一下模型的训练、验证以及测试流程。
目录
1. 训练
2. 验证
3. 测试
1. 训练
def train(config, model, train_iter, dev_iter, test_iter):start_time = time.time()model.train()#训练模式param_optimizer = list(model.named_parameters())#下列参数 不进行正则化(权重衰减)no_decay = ['bias', 'LayerNorm.bias', 'LayerNorm.weight']optimizer_grouped_parameters = [{'params': [p for n, p in param_optimizer if not any(nd in n for nd in no_decay)], 'weight_decay': 0.01},{'params': [p for n, p in param_optimizer if any(nd in n for nd in no_decay)], 'weight_decay': 0.0}]# optimizer = torch.optim.Adam(model.parameters(), lr=config.learning_rate)#优化器optimizer = BertAdam(optimizer_grouped_parameters,lr=config.learning_rate,warmup=0.05,t_total=len(train_iter) * config.num_epochs)total_batch = 0 # 记录进行到多少batchdev_best_loss = float('inf')#跟踪验证集最小的loss 或 最大f1-score或准确率last_improve = 0 # 记录上次验证集loss下降的batch数flag = False # 记录是否很久没有效果提升model.train() #训练模式for epoch in range(config.num_epochs):print('Epoch [{}/{}]'.format(epoch + 1, config.num_epochs))for i, (trains, labels) in enumerate(train_iter):outputs = model(trains)#前向传播 获取输出model.zero_grad()#清空梯度loss = F.cross_entropy(outputs, labels)#计算交叉熵损失(内部包含softmax log等操作) 可以用nn里面的函数 也可以用F中的函数 labels为整数索引 内部会自动转换为one-hotloss.backward()#计算梯度optimizer.step()#更新参数if total_batch % 100 == 0:#每100个batch 计算一下在验证集上的指标 或者像之前项目中那样 每一个epoch在验证集上计算相关指标true = labels.data.cpu()predic = torch.max(outputs.data, 1)[1].cpu()train_acc = metrics.accuracy_score(true, predic)#当前batch上训练集的准确率 因为是类别均衡数据集,所以可以直接用准确率作评估指标dev_acc, dev_loss = evaluate(config, model, dev_iter)#计算此时模型在验证集上的损失和准确率if dev_loss < dev_best_loss:dev_best_loss = dev_loss#更新验证集最小损失torch.save(model.state_dict(), config.save_path)#保存在验证集上损失最小的参数improve = '*'#效果提升标志last_improve = total_batchelse:improve = ''time_dif = get_time_dif(start_time)msg = 'Iter: {0:>6}, Train Loss: {1:>5.2}, Train Acc: {2:>6.2%}, Val Loss: {3:>5.2}, Val Acc: {4:>6.2%}, Time: {5} {6}'print(msg.format(total_batch, loss.item(), train_acc, dev_loss, dev_acc, time_dif, improve))model.train()#回到训练模式total_batch += 1if total_batch - last_improve > config.require_improvement:# 如果长期没有提高 就提前终止# 验证集loss超过1000batch没下降,结束训练print("No optimization for a long time, auto-stopping...")flag = Truebreakif flag:breaktest(config, model, test_iter)#模型训练结束后 进行测试
2. 验证
def evaluate(config, model, data_iter, test=False):model.eval()#测试模式loss_total = 0predict_all = np.array([], dtype=int)#存储验证集所有batch的预测结果labels_all = np.array([], dtype=int)#存储验证集所有batch的真实标签with torch.no_grad():for texts, labels in data_iter:outputs = model(texts)loss = F.cross_entropy(outputs, labels)loss_total += losslabels = labels.data.cpu().numpy()predic = torch.max(outputs.data, 1)[1].cpu().numpy()labels_all = np.append(labels_all, labels)predict_all = np.append(predict_all, predic)acc = metrics.accuracy_score(labels_all, predict_all)#计算验证集准确率if test:#如果是测试集的话 计算一下分类报告report = metrics.classification_report(labels_all, predict_all, target_names=config.class_list, digits=4)confusion = metrics.confusion_matrix(labels_all, predict_all)#计算混淆矩阵return acc, loss_total / len(data_iter), report, confusionreturn acc, loss_total / len(data_iter)#返回准确率和每个batch的平均损失
3. 测试
def test(config, model, test_iter):# testmodel.load_state_dict(torch.load(config.save_path))#加载使验证集损失最小的参数model.eval()#测试模式start_time = time.time()test_acc, test_loss, test_report, test_confusion = evaluate(config, model, test_iter, test=True)#计算测试集准确率,每个batch的平均损失 分类报告、混淆矩阵msg = 'Test Loss: {0:>5.2}, Test Acc: {1:>6.2%}'print(msg.format(test_loss, test_acc))print("Precision, Recall and F1-Score...")print(test_report)print("Confusion Matrix...")print(test_confusion)time_dif = get_time_dif(start_time)print("Time usage:", time_dif)