Closed cshoebridge closed 4 months ago
Great idea! I wonder however if we should pre-do this. Iterating over the entire dataset can be super long, so maybe this should be in each DataSet rather than in LIONSolver?
Unsure, actually, so maybe a thing worth discussing.
On second thought, even if it was in the datasets we provide, it would also need to be in the solver to be able to support generic datasets, so it makes sense
This has been tested I assume? happy to merge?
yes good idea, could be worth looking at, though the inner working of the datasets is out of my realm of expertise
and yes, tested, as set up it doesn't break anything else, it's entirely optional.
Zak mentioned that it is often beneficial to normalize inputs. This is done using global dataset statistics, so shouldn't be the job of a LIONModel, who has no knowledge of the dataset, rather the job of the LIONSolver who does.
Added functions to optionally enable normalization and logic in existing solvers to leverage this.