This topic describes how to prepare the Mars runtime environment.

To run Mars in MaxCompute, you must prepare the Mars runtime environment by using one of the following methods:
  • DataWorks
    1. Create a DataWorks PyODPS 3 node, which provides features of PyODPS and Mars.
      You can run the following commands in the new PyODPS 3 node to check the versions of PyODPS and Mars. Make sure that the versions meet the requirements.
      from odps import __version__ as odps_version
      from mars import __version__ as mars_version
      print(odps_version)
      print(mars_version)

      odps_version: the version of PyODPS. Make sure that the PyODPS version is V0.9.3.1 or later. mars_version: the version of Mars. Make sure that the Mars version is V0.4.4 or later.

    2. Initialize a MaxCompute entry.

      You can use the MaxCompute entry provided by the DataWorks PyODPS 3 node.

  • Other environments
    1. Install pip. After pip is installed, install PyODPS and Mars by running the pip install command in a command-line interface such as the Command Prompt in Windows. For more information about how to install pip, see Installation in the pip documentation. The following commands show an example of how to use the pip install command:
      pip install -U pip  # Optional. Make sure that pip is in the latest version.
      pip install pyodps -i https://mirrors.aliyun.com/pypi/simple/ # Install the latest version of PyODPS. In the command, https://mirrors.aliyun.com/pypi/simple/ is the URL of the Python Package Index (PyPI) mirror that Alibaba Cloud provides to accelerate package download.
      pip install pymars -i https://mirrors.aliyun.com/pypi/simple/ # Install the latest version of Mars.
      pip install protobuf -i https://mirrors.aliyun.com/pypi/simple/ # Install the latest version of protocol buffers.
      pip install pyarrow -i https://mirrors.aliyun.com/pypi/simple/ # Optional. Install the latest version of PyArrow to accelerate job execution in Mars.
    2. Initialize a MaxCompute entry.

      You must use your AccessKey ID and AccessKey secret to initialize the MaxCompute entry. For more information about how to initialize a MaxCompute entry, see PyODPS: ODPS Python SDK and data analysis framework.