look at figure 2-3. It shows a sine wave and an overlapped quantized sine wave. The sine wave is continuous, a high precision representation. The quantized sine wave is a low precision representation which which takes integer values in the range [0, 5]. As a result, the quantized wave requires low transmission bandwidth. Figure 2-3: Quantization of sine waves. Let’s dig deeper into its mechanics using an

CHILD_PARAMS = dict( epochs=6, batch_size=128, learning_rate=0.001, train_ds=train_ds, val_ds=val_ds, rolling_accuracies_window=20, max_branch_length=2, blocks=5, cells=2, initial_width=1, initial_channels=4 self.vds = CHILD_PARAMS['val_ds'].batch(256) self.past_accuracies = deque( maxlen=CHILD_PARAMS['rolling_accuracies_window'] ) self.past_accuracies.append(DATASET_PARAMS['baseline_accuracy']) self.layers train(model) self.past_accuracies.append(accuracy) rolling_accuracy = (sum(self.past_accuracies)/len(self.past_accuracies)) reward = accuracy - rolling_accuracy return reward, accuracy The get_rewards()

for Artificial Intelligence Imperfect Information Huge State and Action Space Long-Term Planning Temporal and Spatial Reasoning Adversarial Real-time Strategy Multiagent Cooperation StarCraft Agents 2 Dropships and 2 tanks vs. 1 Ultralisk Hierarchical Reinforcement Learning Strategy & Planning Combat Economy Information Imitation Learning Supervised Learning Reinforcement Learning Continual

度，并且窗口的长度始终为10个单位长度， 直至移动到末端。 由此可知，通过滑动窗口统计的指标会更加 平稳一些，数据上下浮动的范围会比较小。 57 数据统计—滑动窗口 Pandas中提供了一个窗口方法rolling()。 rolling(window, min_periods=None, center=False, win_ty pe=None, on=None, axis=0, closed=None) ➢ window