Created by: zhangxu3486432

可以使用 task.load_parameters 加载 best_model，

但是构建 task 实例过于繁琐，只是预测任务的话，会加载很多不必要的参数，有没有更加简便的方法？

import paddlehub as hub

from paddlehub.dataset.base_nlp_dataset import BaseNLPDataset

class TrafficNews(BaseNLPDataset):
    def __init__(self):
        # 数据集存放位置
        self.dataset_dir = "/home/traffic_news/dataset"
        super(TrafficNews, self).__init__(
            base_path=self.dataset_dir,
            train_file="train.tsv",
            dev_file="valid.tsv",
            test_file="test.tsv",
            train_file_with_header=True,
            dev_file_with_header=True,
            test_file_with_header=True,
            # 数据集类别集合
            label_list=['1','2','3','4']
        )
dataset = TrafficNews()

module = hub.Module(name='chinese-electra-base')

reader = hub.reader.ClassifyReader(
    dataset=dataset,
    vocab_path=module.get_vocab_path(),
    sp_model_path=module.get_spm_path(),
    word_dict_path=module.get_word_dict_path(),
    max_seq_len=128
)

inputs, outputs, program = module.context(trainable=True, max_seq_len=128)

# Use "pooled_output" for classification tasks on an entire sentence.
pooled_output = outputs["pooled_output"]

feed_list = [
    inputs["input_ids"].name,
    inputs["position_ids"].name,
    inputs["segment_ids"].name,
    inputs["input_mask"].name,
]

strategy = hub.AdamWeightDecayStrategy(
    weight_decay=0.01,
    warmup_proportion=0.1,
    learning_rate=1e-4
)

config = hub.RunConfig(
    use_cuda=False,
    num_epoch=10,
    checkpoint_dir="hub_electra",
    batch_size=16,
    eval_interval=50,
    strategy=strategy
)

task = hub.TextClassifierTask(
        data_reader=reader,
        feature=pooled_output,
        feed_list=feed_list,
        num_classes=dataset.num_labels,
        config=config
)

task.load_parameters('hub_electra/best_model/')