Architecture of our CAE for feature extraction. The 1 × 1 convolutional layer is at the center to act as the bottleneck. The values obtained at this bottleneck are the result of the encoder part, and constitute the outputs of the NN, the last layer from Fig. 1. We note that C1D are 1D convolutions and C1TD are 1D transposed convolutions. The activation after each convolutional layer function is always an exponential linear unit (ELU). For convolutional layers the heights were scaled to the number of channels and the depths were scaled to the resulting number of kernels per channel. The detailed code available at https://github.com/cwestend/iNNterpol