Hands-On Neuroevolution with Python

The asynchronous task runner

This function registers a specific task to be evaluated by a worker in the GPU device context. It takes the ID of the task, the task object holder, and the callback to be executed on task completion as input. This function is defined under the name run_async and is implemented as follows:

1. First, it extracts the corresponding data from the task object and loads it into the current TensorFlow session:

    theta, extras, max_frames=task    self.model.load(self.sess, task_id, theta, extras)    if max_frames is None:        max_frames = self.env.env_default_timestep_cutoff

Here, theta is an array with all the connection weights in the controller ANN model, extras holds a random seeds list of the corresponding genome, and max_frames ...