Can autosklearn generate a model file that can be reused for classifying new data? Would be useful for classifying big data streams.