Hands-On Neuroevolution with Python

Creating the evaluation workers

One of the primary responsibilities of the ConcurrentWorkers class is to create and manage RLEvalutionWorker instances. This is done in the class constructor as follows:

    self.workers = [RLEvalutionWorker(make_env_f, *args,          ref_batch=ref_batch,          **dict(kwargs, device=gpus[i])) for i in range(len(gpus))]    self.model = self.workers[0].model    self.steps_counter = sum([w.steps_counter for w in self.workers])    self.async_hub = AsyncTaskHub()    self.hub = WorkerHub(self.workers, self.async_hub.input_queue,                             self.async_hub)

Here, we create the number of RLEvalutionWorker instances that are correlated to the number of GPU devices that are available in the system. After that, we initialize the selected ANN graph model and ...