This PR tidies up some implementation details and renames a few files for clarity.
When implementing regularisation penalties, we combined the bare loss and the penalty into a single function penalized_loss(x, y) but a more flexible arrangement is to keep the two separate as long as possible. This is the main change of substance in the PR.
A use case in my research was the impetus for making this change.
This PR tidies up some implementation details and renames a few files for clarity.
When implementing regularisation penalties, we combined the bare loss and the penalty into a single function
penalized_loss(x, y)
but a more flexible arrangement is to keep the two separate as long as possible. This is the main change of substance in the PR.A use case in my research was the impetus for making this change.