Troubleshooting the problem of pytorch geometric torch scatter and torch spark installation errors

problem

In the installation torch-scatter、torch-sparse Waiting for a wrong report
Report errors subprocess.CalledProcessError: Command ‘[‘which‘, ‘g++‘]‘ returned non-zero exit status 1.

Traceback (most recent call last):
    File "<string>", line 1, in <module>
    File "/tmp/pip-install-56zzyn27/torch-sparse/setup.py", line 46, in <module>
      packages=find_packages(),
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/site-packages/setuptools/__init__.py", line 153, in setup
      return distutils.core.setup(**attrs)
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/distutils/core.py", line 148, in setup
      dist.run_commands()
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/distutils/dist.py", line 955, in run_commands
      self.run_command(cmd)
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/distutils/dist.py", line 974, in run_command
      cmd_obj.run()
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/site-packages/wheel/bdist_wheel.py", line 290, in run
      self.run_command('build')
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/distutils/cmd.py", line 313, in run_command
      self.distribution.run_command(command)
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/distutils/dist.py", line 974, in run_command
      cmd_obj.run()
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/distutils/command/build.py", line 135, in run
      self.run_command(cmd_name)
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/distutils/cmd.py", line 313, in run_command
      self.distribution.run_command(command)
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/distutils/dist.py", line 974, in run_command
      cmd_obj.run()
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/site-packages/setuptools/command/build_ext.py", line 79, in run
      _build_ext.run(self)
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/distutils/command/build_ext.py", line 339, in run
      self.build_extensions()
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/site-packages/torch/utils/cpp_extension.py", line 232, in build_extensions
      self._check_abi()
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/site-packages/torch/utils/cpp_extension.py", line 370, in _check_abi
      check_compiler_abi_compatibility(compiler)
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/site-packages/torch/utils/cpp_extension.py", line 162, in check_compiler_abi_compatibility
      if not check_compiler_ok_for_platform(compiler):
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/site-packages/torch/utils/cpp_extension.py", line 138, in check_compiler_ok_for_platform
      which = subprocess.check_output(['which', compiler], stderr=subprocess.STDOUT)
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/subprocess.py", line 356, in check_output
      **kwargs).stdout
    File "/home/aha/anaconda3/envs/geo/lib/python3.6/subprocess.py", line 438, in run
      output=stdout, stderr=stderr)
  subprocess.CalledProcessError: Command '['which', 'g++']' returned non-zero exit status 1.
  ----------------------------------------
  ERROR: Failed building wheel for torch-sparse

Solution

step 1: install gcc and g++

This is because gcc and g++ No installation , You can install it with the following command

sudo apt-get install build-essential

Ubuntu By default , It didn't provide C/C++ Compiler environment for , Therefore, manual installation is also required . If installed separately gcc as well as g++ More trouble , Fortunately, , To be able to compile Ubuntu The kernel of ,Ubuntu Provides a build-essential software package .
Installation completed , Some users re execute pip Command to install scatter and sparse That's all right. , But I have a problem again .

step 2: solve torch_scatter Installation version problem

Let's start with the solution , Is to specify in the installation command torch_scatter Version of
pip install torch-scatter==2.0.6 -f https://data.pyg.org/whl/torch-1.6.0+cu101.html
You can use only
pip install torch-scatter==2.0.6

Let's talk about the process of solving the problem ：
First, I run the following commands under the guidance of the official website ：
pip install -f https://data.pyg.org/whl/torch-1.6.0+cu101.html
An error has occurred , Some error reports are as follows :

csrc/scatter.cpp:135:30: error: no matching function for call to ‘at::Tensor::div_(at::Tensor&, const char [6])’
           out.div_(count, "floor");
                                  ^
    In file included from /opt/conda/lib/python3.7/site-packages/torch/include/ATen/Tensor.h:3:0,
                     from /opt/conda/lib/python3.7/site-packages/torch/include/ATen/Context.h:4,
                     from /opt/conda/lib/python3.7/site-packages/torch/include/ATen/ATen.h:5,
                     from /opt/conda/lib/python3.7/site-packages/torch/include/torch/csrc/api/include/torch/types.h:3,
                     from /opt/conda/lib/python3.7/site-packages/torch/include/torch/script.h:3,
                     from csrc/scatter.cpp:2:
    /opt/conda/lib/python3.7/site-packages/torch/include/ATen/core/TensorBody.h:649:12: note: candidate: at::Tensor& at::Tensor::div_(const at::Tensor&) const
       Tensor & div_(const Tensor & other) const;
                ^~~~
    /opt/conda/lib/python3.7/site-packages/torch/include/ATen/core/TensorBody.h:649:12: note:   candidate expects 1 argument, 2 provided
    /opt/conda/lib/python3.7/site-packages/torch/include/ATen/core/TensorBody.h:651:12: note: candidate: at::Tensor& at::Tensor::div_(c10::Scalar) const
       Tensor & div_(Scalar other) const;
                ^~~~
    /opt/conda/lib/python3.7/site-packages/torch/include/ATen/core/TensorBody.h:651:12: note:   candidate expects 1 argument, 2 provided
    error: command 'gcc' failed with exit status 1

Take a closer look at this command. At the beginning, the execution is :

That is, trying to install torch_scatter 2.0.9 edition .
however , When I go to the combination given on the official website , I found that my system should not use 2.0.9 edition
https://data.pyg.org/whl/
Click... In the above link torch-1.6.0+cu101

List the following torch_scatter edition

The highest is 2.0.6 edition , according to 2.0.9 The version does not match my system , Therefore, an error is reported . As for why this happens , I don't know . But we can specify torch_scatter To solve the problem .
pip install torch-scatter==2.0.6 -f https://data.pyg.org/whl/torch-1.6.0+cu101.html

Refer to the post ：
https://blog.csdn.net/klarclm/article/details/8550931
https://blog.csdn.net/Iv_zzy/article/details/109310965