This is an experiment to generate symbolic music with LSTM and the representation of the symbolic music is computed through a positive definite kernel. See here (https://archive.org/details/measuring-note-similarity-with-positive-definite-kernels) for more details for the positive definite kernels on notes.