从H5文件创建数据集遇阻：HDF5插件目录读取报错求助

错误信息 #

加载H5文件时触发以下错误：

OSError: Can't read data (can't open directory: /opt/conda/envs/AE4353/lib/hdf5/plugin)

加载代码 #

确认文件路径正确且依赖已安装，但运行以下代码仍报错：

def load_h5_file(h5_file_path):
    with h5py.File(h5_file_path, "r") as file: 
        images = file["images"][:] 
        targets = file["targets"][:] 

        Images = torch.tensor(images)  
        Targets = torch.tensor(targets)  

    return Images, Targets

load_h5_file(h5_file_path)

已尝试的操作 #

1. conda安装hdf5-plugin失败：

   (AE4353_user_env) vscode ➜ /workspaces/AE4353-Y24 (main) $ conda install -c conda-forge hdf5-plugin
   Channels:
    - conda-forge
    - defaults
   Platform: linux-64
   Collecting package metadata (repodata.json): done
   Solving environment: failed
   
   PackagesNotFoundError: The following packages are not available from current channels:
   
     - hdf5-plugin
   
   Current channels:
   
     - https://conda.anaconda.org/conda-forge
     - defaults
   

2. pip安装h5py[hdf5plugin]出现警告：

   Requirement already satisfied: h5py[hdf5plugin] in /home/kibangadas/.local/lib/python3.8/site-packages (3.11.0)
     WARNING: h5py 3.11.0 does not provide the extra 'hdf5plugin'
   Requirement already satisfied: numpy>=1.17.3 in /home/kibangadas/.local/lib/python3.8/site-packages (from h5py[hdf5plugin]) (1.24.4)
   

3. 单独安装hdf5plugin成功，但问题依旧：

   Collecting hdf5plugin
     Downloading hdf5plugin-5.0.0-py3-none-manylinux_2_17_x86_64.manylinux2014_x86_64.whl (45.6 MB)
        |████████████████████████████████| 45.6 MB 14.1 MB/s
   Requirement already satisfied: h5py>=3.0.0 in /home/kibangadas/.local/lib/python3.8/site-packages (from hdf5plugin) (3.11.0)
   Requirement already satisfied: numpy>=1.17.3 in /home/kibangadas/.local/lib/python3.8/site-packages (from h5py>=3.0.0->hdf5plugin) (1.24.4)
   Installing collected packages: hdf5plugin
   Successfully installed hdf5plugin-5.0.0
   

解决方案 #

1. 手动指定HDF5_PLUGIN_PATH环境变量 #

hdf5plugin安装后，需要让HDF5库找到插件路径，可在代码开头添加：

import os
import hdf5plugin
os.environ['HDF5_PLUGIN_PATH'] = hdf5plugin.PLUGIN_PATH

或者在终端运行代码前设置环境变量：

export HDF5_PLUGIN_PATH=$(python -c "import hdf5plugin; print(hdf5plugin.PLUGIN_PATH)")

2. 确保conda环境使用正确的hdf5plugin安装 #

检查VSCode的Python解释器是否选中了AE4353_user_env，然后激活该环境重新安装：

conda activate AE4353_user_env
pip install hdf5plugin

3. 创建缺失的插件目录并软链接 #

直接补全报错中缺失的目录，将hdf5plugin的插件文件链接过去：

mkdir -p /opt/conda/envs/AE4353/lib/hdf5/plugin
ln -s $(python -c "import hdf5plugin; print(hdf5plugin.PLUGIN_PATH)")/* /opt/conda/envs/AE4353/lib/hdf5/plugin/

4. 调整h5py版本兼容 #

部分h5py版本与hdf5plugin存在兼容性问题，尝试降级h5py：

pip install h5py==3.9.0

内容的提问来源于stack exchange，提问作者Anon