Our DAG contains to two implicit loops:
a) the sub-DAG after X and y marked operations can be repeatedly executed during multiple train/validate runs (e.g. when cv=5).
b) in each train/validation run, the sub-DAG is runs iterated during train and once during validation
The un-rolling fit/ predict is not trivial, since right now every operations return only on output, with un-rolling now, we have Transformers / Estimators which return not only their transform / predict output, but also return their state during fit, which makes them Multi Output Operation, which did not have before, this is visualized in this image:
Our DAG contains to two implicit loops:
a) the sub-DAG after X and y marked operations can be repeatedly executed during multiple train/validate runs (e.g. when cv=5).
b) in each train/validation run, the sub-DAG is runs iterated during train and once during validation
The un-rolling fit/ predict is not trivial, since right now every operations return only on output, with un-rolling now, we have Transformers / Estimators which return not only their transform / predict output, but also return their state during fit, which makes them Multi Output Operation, which did not have before, this is visualized in this image: