

Workflow

Non-Scalar

Some jobs, such as in machine learning, can result in a new workflow being generated and placed in the user's account.

Creation during ML Jobs

If any unit in the workflow has the workflow:pyml_predict property, Express will be called to construct the new predict workflow. The following process is performed to convert a workflow from "Train" to "Predict" mode:

• The IS_WORKFLOW_RUNNING_TO_PREDICT flag is set to True
• Any units which contain a workflow:pyml_predict property have this property removed (to prevent new workflows from being generated when the predict job is run)
• IO units are configured to download any python objects saved by the user with the context manager in settings.py. These files are saved inside of the .job_context folder.

Example

In the below example, a user has run a machine learning training workflow, which has generated a workflow named workflow:pyml_predict.

Schema

The JSON schema and an example implementation can be found here