test_save.py

import os
import numpy as np
import torch
from pathlib import Path

def test_saved_files(content_path, num_epochs=3):
    """测试保存的文件是否符合要求
    
    Args:
        content_path: 训练过程的根目录
        num_epochs: 要测试的epoch数量
    """
    # 检查目录结构
    required_dirs = [
        'model',
        'dataset/representation',
        'dataset/prediction',
        'dataset/label'
    ]
    
    for dir_path in required_dirs:
        full_path = os.path.join(content_path, dir_path)
        if not os.path.exists(full_path):
            print(f"错误: 目录不存在: {full_path}")
            return False
    
    # 检查模型文件
    print("\n检查模型文件...")
    for epoch in range(1, num_epochs + 1):
        model_path = os.path.join(content_path, 'model', f'{epoch}.pth')
        if not os.path.exists(model_path):
            print(f"错误: 模型文件不存在: {model_path}")
            return False
        try:
            # 尝试加载模型文件以验证其有效性
            state_dict = torch.load(model_path, map_location='cpu')
            print(f"✓ {epoch}.pth 格式正确")
        except Exception as e:
            print(f"错误: 无法加载模型文件 {model_path}: {str(e)}")
            return False
    
    # 检查特征向量文件
    print("\n检查特征向量文件...")
    prev_samples = None
    for epoch in range(1, num_epochs + 1):
        repr_path = os.path.join(content_path, 'dataset', 'representation', f'{epoch}.npy')
        if not os.path.exists(repr_path):
            print(f"错误: 特征向量文件不存在: {repr_path}")
            return False
        try:
            features = np.load(repr_path)
            samples, dim = features.shape
            if not (512 <= dim <= 1024):
                print(f"警告: 特征维度 {dim} 不在预期范围[512, 1024]内")
            if prev_samples is not None and samples != prev_samples:
                print(f"错误: epoch {epoch} 的样本数量与之前不一致")
                return False
            prev_samples = samples
            print(f"✓ {epoch}.npy 格式正确 [样本数: {samples}, 特征维度: {dim}]")
        except Exception as e:
            print(f"错误: 无法加载特征向量文件 {repr_path}: {str(e)}")
            return False
    
    # 检查预测结果文件
    print("\n检查预测结果文件...")
    for epoch in range(1, num_epochs + 1):
        pred_path = os.path.join(content_path, 'dataset', 'prediction', f'{epoch}.npy')
        if not os.path.exists(pred_path):
            print(f"错误: 预测结果文件不存在: {pred_path}")
            return False
        try:
            predictions = np.load(pred_path)
            samples, classes = predictions.shape
            if samples != prev_samples:
                print(f"错误: 预测结果的样本数量与特征向量不一致")
                return False
            if classes != 10:  # CIFAR-10 有10个类别
                print(f"警告: 类别数量 {classes} 不等于10")
            print(f"✓ {epoch}.npy 格式正确 [样本数: {samples}, 类别数: {classes}]")
        except Exception as e:
            print(f"错误: 无法加载预测结果文件 {pred_path}: {str(e)}")
            return False
    
    # 检查标签文件
    print("\n检查标签文件...")
    label_path = os.path.join(content_path, 'dataset', 'label', 'labels.npy')
    if not os.path.exists(label_path):
        print(f"错误: 标签文件不存在: {label_path}")
        return False
    try:
        labels = np.load(label_path)
        if len(labels.shape) != 1:
            print(f"错误: 标签文件维度不正确，应为1维数组，实际为{len(labels.shape)}维")
            return False
        if labels.shape[0] != prev_samples:
            print(f"错误: 标签数量与样本数量不一致")
            return False
        if not np.all((labels >= 0) & (labels < 10)):  # CIFAR-10 的标签范围是0-9
            print("错误: 存在超出范围的标签值")
            return False
        print(f"✓ labels.npy 格式正确 [样本数: {labels.shape[0]}]")
    except Exception as e:
        print(f"错误: 无法加载标签文件 {label_path}: {str(e)}")
        return False
    
    print("\n✓ 所有文件格式检查通过！")
    return True

if __name__ == "__main__":
    # 设置要测试的目录路径
    content_path = "/home/ruofei/RRF/ttvnet/Image/AlexNet/model/0"
    
    # 运行测试
    test_saved_files(content_path, num_epochs=44)