Script Requirement

File type.Create training task by compressing your entire folder as one single .zip file. The compressed file will be sent directly to local container(virtual environment) in Orion GPU computing node and unzipped for execution.

Programming language.The only programming language Orion GPU client miner will accept and execute is python3.6.

Executed file.Directory structure for packaging your scripts and data within your task folder(e.g. /mytask) could be flexible, however, only one python script will be executed in working node. This executed file, must be located directly under your workspace directory as /mytask/file_name.py.

Parameter configuration.It is mandatory for users to submit a params.json file, to specify your task configuration information to computing node. This json file must be located directly under your workspace directory as /mytask/params.json.

Users need to define four parameters in params.json file:

  • "exec_file_name": the name of python file to be executed.
  • "data_uri": users are allowed to provide a public accessible external URL for data-downloading. Orion computing node will download content from URL and save original content into directory in "data_path". If not applicable, enter "".
  • "data_path": local directory to which data from "data_uri" will be saved. If not applicable, enter "".
  • "output_path": local directory to which task result will be saved. All contents from output directory will be compressed and retrievable by task provider.

Path.When importing data or modules under workspace directory /mytask, relative path is accepted. Path pointing upwards beyond /mytask will not be recognized and will lead to task termination.

Module Requirements.Users will need to list all required third-party python packages in a requirements text file requirements.txt. Orion will pip install the listed libraries in your client computing machine container as your task-execution environment. With one line per library, list the package name and optionally the version for the package (defaults to latest, if version not defined): 

    numpy==1.14.5
    pandas
    matplotlib>=3.0.0
    scikit-learn==0.19.2
    nltk==3.3
Note: Frameworks like Tensorflow and pytorch are already pre-configured in Docker image. Users are suggested to comply with the default version of these frameworks.

Suggested Task Directory.When importing data or modules under workspace directory /mytask, relative path is accepted. Path pointing upwards beyond /mytask will not be recognized and will lead to task termination.

  • Directory structure
    • directory
    Note: * mandatory, name could be self-defined; **mandatory, name is fixed directory files
  • Params.json for suggested directory Code snippet.
    •     {
        "exec_file_name": "file_name.py",
        "data_uri": "http://dataURL/data.zip" ,
        "data_path": "datadir",
        "output_path":"result"
    }