Turning on multiprocess in the ParallelProcessor. To enable this, we have to:
replace all local/inner functions/lambdas/classes with public functions/classes.
replace metric decorator with modified MetricWrapper
Note that turning on parallel processing on small dataset is actually counterproductive since the overhead of spawning processes is high. In this case it's better to use single process. This will be a TODO to let user select whether to use multiprocess. In the future we may want to let the framework decide what to do.
Description
Turning on multiprocess in the ParallelProcessor. To enable this, we have to:
Note that turning on parallel processing on small dataset is actually counterproductive since the overhead of spawning processes is high. In this case it's better to use single process. This will be a TODO to let user select whether to use multiprocess. In the future we may want to let the framework decide what to do.