cuda 9.0 安装torch 0.4_PyTorch geometric 安装中libcusparse.so.10 error的解决策略

准备工作（自己的电脑）：

• OS: Ubuntu 16.04
• Python version: 3.6
• PyTorch version: 1.6.0
• CUDA version: 10.1

安装流程（pip）：

1. 首先确保PyTorch 1.4.0 及以上版本

$ python -c "import torch; print(torch.__version__)"
>>> 1.6.0

2. 查询GPU版PyTorch的CUDA版本

$ python -c "import torch; print(torch.version.cuda)"
>>> 10.1

3. 安装依赖包

$ pip install torch-scatter==latest+cu101 -f https://pytorch-geometric.com/whl/torch-1.6.0.html
$ pip install torch-sparse==latest+cu101 -f https://pytorch-geometric.com/whl/torch-1.6.0.html
$ pip install torch-cluster==latest+cu101 -f https://pytorch-geometric.com/whl/torch-1.6.0.html
$ pip install torch-spline-conv==latest+cu101 -f https://pytorch-geometric.com/whl/torch-1.6.0.html
$ pip install torch-geometric

其中“cu101” 和 “torch-1.6.0” 可依据自己的配置来选择

4. 测试是否安装成功

$ python
>>> import torch
>>> from torch_geometric.data import Data

OSError: libcusparse.so.10: cannot open shared object file: No such file or directory

显然，未能成功安装。

修复流程

1. 检查 nvcc

$ nvcc --version

-base：/usr/lib/command-not-found：/usr/bin/python3: bad interpreter: No such file or directory

又出问题！既然不识别命令，那就安装呗～

2. 打开terminal

$ nano /home/username/.bashrc

 # or

$ nano /home/$USER/.bashrc

3. 添加以下命令行

 export PATH="/usr/local/cuda-10.1/bin:$PATH"
 export LD_LIBRARY_PATH="/usr/local/cuda-10.1/lib64:$LD_LIBRARY_PATH"

4. 保存并关闭terminal

 On you keyboard press the following: 

 ctrl + o             --> save 
 enter or return key  --> accept changes
 ctrl + x             --> close editor

5. 运行nvcc(一定要确保PyTorch和CUDA的版本一致)

$ nvcc --version
>>> 10.1

$ python -c "import torch; print(torch.version.cuda)"
>>> 10.1

6. 再次测试是否安装成功

import torch
from torch_geometric.data import Data

edge_index = torch.tensor([[0, 1, 1, 2],
                           [1, 0, 2, 1]], dtype=torch.long)
x = torch.tensor([[-1], [0], [1]], dtype=torch.float)

data = Data(x=x, edge_index=edge_index)

>>> print(data.keys)
>>> ['x', 'edge_index']

恭喜，安装成功:)