Hands-On Neuroevolution with Python

The graph evaluation loop

This is the loop that we use to evaluate the previously created network graph over multiple games in parallel – the number of games that can be evaluated simultaneously is determined by the batch_size parameter.

The evaluation loop is defined in the _loop function and is implemented as follows:

1. First, we start with the creation of arrays to hold game evaluation values over multiple episodes:

running = np.zeros((self.batch_size,), dtype=np.bool)cumrews = np.zeros((self.batch_size, ), dtype=np.float32)cumlen = np.zeros((self.batch_size, ), dtype=np.int32)

2. Next, we start the loop and set the corresponding indices of the running array we just created to True:

    while True:        # nothing loaded, block if not any(running): ...