Converging Workflow

Get a copy of this example

$ cylc get-resources examples/converging-workflow

A workflow which runs a pattern of tasks over and over until a convergence condition has been met.

• The increment task runs some kind of model or process which increments us toward the solution.

• The check_convergence task, checks if the convergence condition has been met.

[meta]
    title = Converging Workflow
    description = """
        A workflow which runs a pattern of tasks over and over until a
        convergence condition has been met.
    """

[scheduling]
    cycling mode = integer
    initial cycle point = 1
    [[graph]]
        P1 = """
            # run "increment" then check the convergence condition
            check_convergence[-P1]:not_converged? => increment => check_convergence

            # if the workflow has converged, then do nothing
            check_convergence:converged?
        """

[runtime]
    [[increment]]
        # a task which evolves the data
    [[check_convergence]]
        # a task which checks whether the convergence condition has been met
        script = """
            if (( CYLC_TASK_CYCLE_POINT == 4 )); then
                # for the purpose of example, assume convergence at cycle point 4
                cylc message -- 'convergence condition met'
            else
                cylc message -- 'convergence condition not met'
            fi
        """
        [[[outputs]]]
            converged = 'convergence condition met'
            not_converged = 'convergence condition not met'

Run it with:

$ cylc vip converging-workflow

Example - Genetic algorithms

An example of a converging workflow might be a genetic algorithm, where you “breed” entities, then test their “fitness”, and breed again, over and over until you end up with an entity which is able to satisfy the requirement.