1、RNN来作为这里的controller，然后用policy gradient来最大化controller
    实现：https://github.com/wallarm/nascell-automl

    • auto-regressive :未来和历史序列相关
    • space：
    • 策略：

    image.png
    Example:
    Input:
    “layers:[1,3,5,7]”,”activation_functions=[tanh,sigmoid,relu,leaky relu]”,”output_activation=[sotmax]”,”optimizer=[Adam,RMS Prop,SGD]”
    Output:
    “layers:5,”activation_functions=relu”,”output_activation=softmax”,”optimizer=Adam”
    image.png
    https://medium.com/@abinesh.mba13/neural-architecture-search-nas-the-future-of-deep-learning-4b35ca473b9