### 数据集划分 为了将测试数据集进一步划分为验证集和测试集，在Python中可以采用类似于训练集与验证集分割的方法。给定一个完整的数据集`test_dataset`以及期望的比例来定义验证集大小，可以通过以下方式实现： ```python import numpy as np from torch.utils.data import random_split def test_valid_split(test_dataset, valid_ratio=0.5, seed=42): ''' Split provided test data into validation set and test set. Parameters: test_dataset (Dataset): The dataset to be split. valid_ratio (float): Proportion of the dataset that should become the validation set. Default is 0.5 which means half will go to each subset. seed (int): Seed value used for reproducibility during splitting. Returns: tuple: A pair containing two datasets representing the validation set and test set respectively. ''' # Calculate sizes based on ratio total_size = len(test_dataset) valid_set_size = int(total_size * valid_ratio) test_set_size = total_size - valid_set_size # Perform actual split using PyTorch's utility function while ensuring deterministic behavior via manual seeding valid_subset, test_subset = random_split( test_dataset, [valid_set_size, test_set_size], generator=torch.Generator().manual_seed(seed) ) return np.array(valid_subset), np.array(test_subset)[^1] ``` 此函数接受三个参数：待拆分的数据集、用于创建验证子集的比例（默认为一半），以及随机种子以确保结果可重复。 通过调整`valid_ratio`参数，可以根据具体需求灵活控制分配给验证集和最终测试集样本数量之间的比例关系。