Generic Task with Lazy Tensor Target #410
Description
To make mlr3torch flexible to use-cases for which no dedicated task type exist, we should implement a flexible task type that allows to have one more more lazy tensor as the target variable. For now, I would still restrict this to supervised learning, but we might one to also think about self-supervised learning and unsupervised learning at some point.