It seems like loading can be optimized for datasets, specifically a 100k sample dataset will take much longer to load than it probably needs to.