Research data supporting "Archetypal landscapes for deep neural networks"

This archive contains input and data files to obtain the results published in: Archetypal landscapes for deep neural networks P.C. Verpoort, A.A. Lee, D.J. Wales Accepted in: Proceedings of the National Academy of Sciences of the USA The folder data_files contains the training data (and testing data where produced) for the energy landscapes reported in the article. The names of the subfolders correspond to the names given to these datasets in the article. The LJAT19 dataset was created specifically for this publication, and has not been reported elsewhere. The other two datasets were taken from the UCI Machine Learning Repository and can also be obtained from there; links to the original data source (accessed on April 24th, 2020) are provided in the README files in each subfolder. For completeness and because these data had to be processed to serve as inputs for our landscape analysis software, the data files used for the present work are also contained within this archive. The folder input_files contains the instruction input files for the GMIN, OPTIM and PATHSAMPLE programs. This is just a starting point, and more manual refinement of parameters and connectivity jobs to run is required in order to fully reproduce the results presented in the article.

本归档文件包含用于复现以下已发表研究成果的输入文件与数据集：论文《面向深度神经网络的原型景观（Archetypal landscapes for deep neural networks）》，作者为P.C. Verpoort、A.A. Lee、D.J. Wales，已被《美国国家科学院院刊（Proceedings of the National Academy of Sciences of the USA）》接收。data_files文件夹包含本文中提及的能量景观所用的训练数据（若生成了测试数据则一并包含），子文件夹的命名与本文中对应数据集的名称一致。LJAT19数据集专为本次发表工作定制，此前未在其他场合公开报道。其余两个数据集取自UCI机器学习库（UCI Machine Learning Repository），也可从该库直接获取；各子文件夹的README文件中均提供了原始数据源的链接（访问时间为2020年4月24日）。为保证研究完整性，同时由于本研究所需数据需经过处理才能作为景观分析软件的输入，本工作使用的数据集文件也一并包含在本归档中。input_files文件夹包含GMIN、OPTIM与PATHSAMPLE程序的指令输入文件。本文件仅为初始配置，若要完整复现本文所述成果，还需手动优化参数并配置待运行的连通性任务。