Engines

machinable provides different execution engines that enable seamless parallel and remote execution. You can also implement your own engine to customize the execution to your needs.

Local

By default, the local engine is used. The engine supports parallel execution using Python's multiprocessing.

ml.execute(ml.Experiment().component('demo'), driver='local')
# use a maximum of 10 processes in parallel
ml.execute(ml.Experiment().component('demo'), driver='local:10')

Ray

Ray is a powerful distributed framework that supports local as well as remote execution on clusters like AWS and Google Cloud instances. Please refer to Ray's documentation to learn more.

To execute machinable tasks on the connected Ray backend use the Ray driver:

ml.execute(ml.Experiment().component('demo'), driver='ray')

Custom engines

You can implement a custom engine as a subclass of machinable.Engine and pass them as an argument.

class CustomEngine(ml.Engine):

    # implement abstract methods

ml.execute(ml.Experiment().component('demo'), driver=CustomDriver())