Hands-On Neuroevolution with Python

The game observation function

This function obtains the current game state from the Atari environment as a game screen buffer. This function is implemented as follows:

    def observation(self, indices=None, name=None):        if indices is None:            indices = np.arange(self.batch_size)        with tf.variable_scope(name, default_name='AtariObservation'):            with tf.device('/cpu:0'):                obs = gym_tensorflow_module.environment_observation(                                   self.instances, indices, T=tf.uint8)            obs = tf.gather(tf.constant(self.color_pallete),                                                 tf.cast(obs,tf.int32))            obs = tf.reduce_max(obs, axis=1)            obs = tf.image.resize_bilinear(obs, self.warp_size,                                                    align_corners=True)            obs.set_shape((None,) + self.warp_size + (1,))            return obs

This function acquires a screengrab from the Atari environment ...