After discussion with @ollieparish and @young-x-skyee: Because there's a limit of about 4 concurrent queue jobs at the CBU, and because loading the data takes quite a long time, it makes more sense to run different functions in sequence after loading the data once, and sbatch-parallelise a different set of parameters (or e.g. sbatch over ANN layers, with nodes within the layers run in sequence).
After discussion with @ollieparish and @young-x-skyee: Because there's a limit of about 4 concurrent queue jobs at the CBU, and because loading the data takes quite a long time, it makes more sense to run different functions in sequence after loading the data once, and
sbatch-parallelise a different set of parameters (or e.g. sbatch over ANN layers, with nodes within the layers run in sequence).