The main code generating the results table in the paper is contained in the three evaluate_benchmark Jupyter notebooks, whereas the corresponding neural networks are trained via train_benchmark.py. powerscaled_net_investigation features further senstivity analyses using a powerscaled network and generates all figures of the main text.