We discuss results for bottomonium at nonzero temperature obtained using NRQCD on Fastsum Generation 2L ensembles, as part of the Fastsum collaboration’s programme to determine the spectrum of the bottomonium system as a function of temperature using a variety of approaches. Here we give an update on results for spectral functions obtained using Kernel Ridge Regression. We pay in particular attention to the generation of training data and introduce the notion of using lattice QCD ensembles to learn how to improve the generation of training data. A practical implementation is given.