Currently maps are loaded, a subset taken using the splits variable, and a new file created with this subset. This file is then passed to the dataloader. This can create issues with multiple runs simulatenously using different sizes of training sets, adding an extra layer of bookkeeping for these additional "temporary" files. It might make more sense to simplify this and just pass the path to the full sim set straight to the dataloader. This is certainly simpler in the case of a single field, but might be more complicated in the case of multiple fields.
Currently maps are loaded, a subset taken using the
splits
variable, and a new file created with this subset. This file is then passed to the dataloader. This can create issues with multiple runs simulatenously using different sizes of training sets, adding an extra layer of bookkeeping for these additional "temporary" files. It might make more sense to simplify this and just pass the path to the full sim set straight to the dataloader. This is certainly simpler in the case of a single field, but might be more complicated in the case of multiple fields.