Supplementary code.

Code in Python and Matlab necessary to reproduce the figures. Code for Figs 1–4 is written in Python 3.6. The networks in Figs 1, 2 and 4 are coded using the standard Numpy package, and the networks in Fig 3 make use of Keras with a TensorFlow backend (keras.io). The application of dendritic normalisation in Keras with TensorFlow allows for immediate inclusion in Keras-based deep learning models. The normalisation requires a custom layer, constraint, and optimiser. Fig 5 uses code written in Matlab 2020b, using the freely available Trees Toolbox package [69]. (ZIP)用于复现本文所有图表的Python与Matlab代码。其中图1至图4的代码基于Python 3.6编写：图1、2、4中的神经网络采用标准Numpy库编码实现，图3的神经网络则使用以TensorFlow为后端的Keras（keras.io）编写。基于TensorFlow的Keras框架中的树突归一化方法可直接集成至基于Keras的深度学习模型中，该归一化操作需自定义层、约束项与优化器。图5的代码使用Matlab 2020b编写，并依托免费可用的Trees Toolbox工具包[69]。（ZIP）