To gather more, and especially more diverse, execution data for the training of the cost functions, it would be very useful to generate arbitrary ExecutionPlanss/ExecutionTaskFlows from a given RheemPlan. In general, this can be done by an appropriate instrumentation of the respective Rheem app. A more general approach, though, is to have Rheem create such arbitrary plans. And a simple way to do so is to provide a PlanEnumerationPruningStrategy that prunes randomly.
From @sekruse on October 21, 2016 11:5
To gather more, and especially more diverse, execution data for the training of the cost functions, it would be very useful to generate arbitrary
ExecutionPlanss/ExecutionTaskFlows from a givenRheemPlan. In general, this can be done by an appropriate instrumentation of the respective Rheem app. A more general approach, though, is to have Rheem create such arbitrary plans. And a simple way to do so is to provide aPlanEnumerationPruningStrategythat prunes randomly.Copied from original issue: daqcri/rheem#31