kgal-core/kgal.distributed/parallelismConfig

parallelismConfig

inline fun DistributedConfigScope<*, *>.parallelismConfig(config: ParallelismConfigScope.() -> Unit)

Creates ParallelismConfig with ParallelismConfigScope and apply it to the current DistributedConfig.

Achieves almost linear speedup with distributed launching! It is recommended to use parallelism in DistributedGA.

Distributed Parallelism is based on the idea of distributed launching DistributedGA.children. After which the results are synchronized and interaction between subpopulations is carried out. It is assumed that each child GA running independently in their own coroutine. That's why workerCount is recommended to be equal to children.size.

Base (and default for evolve) launchChildren stage function support Distributed Parallelism by default using process function - it's safe to use!

Example:

dGA {
    // configure specific params of DistributedGA

    // configure Distributed parallelism
    parallelismConfig {
        workersCount = 4 // set it equal to children.size
    }

    // create and add children GAs
    +pGAs(
        count = 4, // island count equal to workersCount
        population = { population(size = 50) }, // island population
    ) {
        // create evolution strategy for island here
    }

    // set distributed evolution strategy:
    // launches children with ga.start() (default == true) before each iteration
    // each child will be running in their own coroutine
    evolve(useDefault = true) {
        stopBy(maxIteration = 5)
        migration(percent = 0.1) // no parallelism here
    }
}

parallelismConfig

See also