Default assumption is that all persistables are fully preconfigured (all relevant state configurations passed on init) so that hashing and reproducibility are guaranteed (*).
Explicitly change that to have parallel classes for preconfiguration vs traditional persistables that get mutated in a declaritive manner during workflows (e.g. fit(new_dataset, *args, ...))
Default assumption is that all persistables are fully preconfigured (all relevant state configurations passed on init) so that hashing and reproducibility are guaranteed (*).
Explicitly change that to have parallel classes for preconfiguration vs traditional persistables that get mutated in a declaritive manner during workflows (e.g. fit(new_dataset, *args, ...))