I’m using an RNN, and my input for each sample is currently a 2D STFT. I’d like to add more input features, but each is of different shape/size. Is there a better approach than padding to make them concatenateable? Would PyTorch’s packing be the right approach?