Fig. 3. (a–e) The ten main data sets used in our neural network and their training/testing splits are shown. “A” and “B” refer to data sets with the label dxy energy and dxy–π* energy gap, respectively. Data sets 1A/1B, 2A/2B, 3A/3B each have ten input Hammett parameters, while sets 4A/4B and 5A/5B have less than ten input Hammett parameters due to the use of feature selection methods to create them.
