基于TNEWS‘ 今日头条中文新闻（短文本）分类

数据部分

内容

TNEWS’今日头条中文新闻数据集来自今日头条的新闻版块，共提取了15个类别的新闻，包括旅游，教育，金融，军事等。数据量：训练集(53,360)，验证集(10,000)，测试集(10,000)

数据处理

将文本数据转换成id

# 转换成id的函数
def convert_example(example, tokenizer):
    encoded_inputs = tokenizer(text=example["sentence"], max_seq_len=128, pad_to_max_seq_len=True)
    return tuple([np.array(x, dtype="int64") for x in [
            encoded_inputs["input_ids"], encoded_inputs["token_type_ids"], [example["label"]]]])
            
# 把训练集合转换成id
train_ds = train_ds.map(partial(convert_example, tokenizer=tokenizer))
# 把验证集合转换成id
dev_ds = dev_ds.map(partial(convert_example, tokenizer=tokenizer))
# 构建训练集合的dataloader
train_batch_size=32
dev_batch_size=32
train_batch_sampler = paddle.io.DistributedBatchSampler(dataset=train_ds, batch_size=train_batch_size, shuffle=True)
train_data_loader = paddle.io.DataLoader(dataset=train_ds, batch_sampler=train_batch_sampler, return_list=True)

# 针对验证集数据加载，我们使用单卡进行评估，所以采用 paddle.io.BatchSampler 即可
# 定义验证集的dataloader
dev_batch_sampler = paddle.io.BatchSampler(dev_ds, batch_size=dev_batch_size, shuffle=False)

dev_data_loader = paddle.io.DataLoader(
        dataset=dev_ds,
        batch_sampler=dev_batch_sampler,
        return_list=True)    

模型构建

class ShortTextClassification(nn.Layer):
    def __init__(self, pretrained_model,num_class,dropout=None):
        super().__init__()
        self.ptm = pretrained_model
        self.dropout = nn.Dropout(dropout if dropout is not None else 0.1)

        # num_labels = 2 (similar or dissimilar)
        self.classifier = nn.Linear(self.ptm.config["hidden_size"], num_class)

    def forward(self,
                input_ids,
                token_type_ids=None,
                position_ids=None,
                attention_mask=None):

        _, cls_embedding = self.ptm(input_ids, token_type_ids, position_ids,
                                    attention_mask)

        cls_embedding = self.dropout(cls_embedding)
        logits = self.classifier(cls_embedding)

        return logits

model = ShortTextClassification(pretrained_model,num_class=len(train_ds.label_list))

训练配置

epochs = 3
num_training_steps = len(train_data_loader) * epochs

# 定义 learning_rate_scheduler，负责在训练过程中对 lr 进行调度
lr_scheduler = LinearDecayWithWarmup(2E-5, num_training_steps, 0.0)

# 训练结束后，存储模型参数
save_dir ="checkpoint"
# 创建保存的文件夹
os.makedirs(save_dir,exist_ok=True)

# Generate parameter names needed to perform weight decay.
# All bias and LayerNorm parameters are excluded.
decay_params = [
    p.name for n, p in model.named_parameters()
    if not any(nd in n for nd in ["bias", "norm"])
]

# 定义 Optimizer
optimizer = paddle.optimizer.AdamW(
    learning_rate=lr_scheduler,
    parameters=model.parameters(),
    weight_decay=0.0,
    apply_decay_param_fun=lambda x: x in decay_params)
# 交叉熵损失
criterion = paddle.nn.loss.CrossEntropyLoss()
# 评估的时候采用准确率指标
metric = paddle.metric.Accuracy()

模型训练和预测

定义评估函数

# 因为训练过程中同时要在验证集进行模型评估，因此我们先定义评估函数
@paddle.no_grad()
def evaluate(model, criterion, metric, data_loader, phase="dev"):
    model.eval()
    metric.reset()
    losses = []
    for batch in data_loader:
        input_ids, token_type_ids, labels = batch
        probs = model(input_ids=input_ids, token_type_ids=token_type_ids)
        # 计算损失
        loss = criterion(probs, labels)
        losses.append(loss.numpy())
        # 计算准确率
        correct = metric.compute(probs, labels)
        #准确率更新
        metric.update(correct)
        accu = metric.accumulate()
    print("eval {} loss: {:.5}, accu: {:.5}".format(phase,
                                                    np.mean(losses), accu))
    model.train()
    metric.reset()
    return np.mean(losses),accu

训练

def do_train(model, criterion, metric, dev_data_loader,train_data_loader):
    
    global_step = 0
    tic_train = time.time()
    best_accuracy=0.0

    for epoch in range(1, epochs + 1):
        for step, batch in enumerate(train_data_loader, start=1):

            input_ids, token_type_ids, labels = batch
            probs = model(input_ids=input_ids, token_type_ids=token_type_ids)
            loss = criterion(probs, labels)
            correct = metric.compute(probs, labels)
            metric.update(correct)
            acc = metric.accumulate()

            global_step += 1 
            
            # 每间隔 100 step 输出训练指标
            if global_step % 100 == 0:
                print(
                    "global step %d, epoch: %d, batch: %d, loss: %.5f, accu: %.5f, speed: %.2f step/s"
                    % (global_step, epoch, step, loss, acc,
                        10 / (time.time() - tic_train)))
                tic_train = time.time()
            loss.backward()
            optimizer.step()
            lr_scheduler.step()
            optimizer.clear_grad()

            # 每间隔 100 step 在验证集和测试集上进行评估
            if global_step % 500 == 0:
                eval_loss,eval_accu=evaluate(model, criterion, metric, dev_data_loader, "dev")
                if(best_accuracy<eval_accu):
                    best_accuracy=eval_accu
                    # 保存模型
                    save_param_path = os.path.join(save_dir, 'model_best.pdparams')
                    paddle.save(model.state_dict(), save_param_path)
                    # 保存tokenizer
                    tokenizer.save_pretrained(save_dir)

do_train(model, criterion, metric, dev_data_loader,train_data_loader) 

预测

state_dict=paddle.load('checkpoint/model_best.pdparams')
model.load_dict(state_dict)

# 测试集可以选择 test，test1.0两个
test_ds = load_dataset('clue', task_name, splits=['test1.0'])

def do_predict(model,example):
    # 把文本转换成input_ids,token_type_ids
    # example=test_ds[0]
    encoded_text = tokenizer(text=example["sentence"], max_seq_len=512, pad_to_max_seq_len=True)
    # 把input_ids变成paddle tensor
    input_ids = paddle.to_tensor([encoded_text['input_ids']])
    # 把token_type_ids变成paddle tensor
    segment_ids = paddle.to_tensor([encoded_text['token_type_ids']])
    # 模型预测
    pooled_output = model(input_ids, segment_ids)
    # 取概率值最大的索引
    out2 = paddle.argmax(pooled_output, axis=1)fgh
    # print('预测的label标签为 {}'.format(out2.numpy()[0]))
    # print('真实的label标签为 {}'.format(test_ds[0]['label']))
    return out2.numpy()[0]

predict_label=[]
for i in tqdm(range(len(test_ds))):
    example=test_ds[i]
    label_pred=do_predict(model,example)
    predict_label.append(label_pred)

output_submit_file = "tnews10_predict.json"
label_map = {i: label for i, label in enumerate(train_ds.label_list)}
# 保存标签结果
with open(output_submit_file, "w") as writer:
    for i, pred in enumerate(predict_label):
        json_d = {}
        json_d['id'] = i
        json_d['label'] = str(label_map[pred])
        writer.write(json.dumps(json_d) + '\n')