refactor!: clarify kw/positional parameters#578

Merged

ValerianRey merged 15 commits intomainfrom

make-some-args-positional-valued-only

Feb 18, 2026

Commits on Feb 18, 2026

Make `output` parameter of `Engine.compute_gramian` positional-only.
PierreQuinton
committed
make `__call__` in `Aggregator`, `Weighting` and `GeneralizedWeighting` take a positional-only parameter.
PierreQuinton
committed
(Private interface) make `GeneralizedWeighting.forward` take a positional-only argument.
PierreQuinton
committed
(Private interface) make `Aggregator.forward` take a single positional only argument.
PierreQuinton
committed
Proposition of minimal separation for backward, could be `tensors` positional only and the rest named-only in my opinion.
PierreQuinton
committed
Minimal separation for mtl_backward, could be stricter (tensors and features positional only and the rest named only). Note that for `tensors` it makes little sense to use named now, before we had …
PierreQuinton
committed
Minimal separation for `jac`, could be a bit stricter with inputs the only positional and named parameter, everything before that positional and after named.
PierreQuinton
committed
Minimal separation for `jac_to_grad`, `aggregator` could also be positional only but this is fine IMO.
PierreQuinton
committed
Make backward stricter.
PierreQuinton
committed
Make mtl_backward stricter. `feature` did not pass the smell test of runing the tests, most call to `mtl_backward` use key-word for `feature` while we would want to make it positional. So I think I…
PierreQuinton
committed
Make `jac` stricter. Problem with `inputs`, it is not supposed to be optional and should therefore be kw+pos
PierreQuinton
committed
Merge branch 'main' into make-some-args-positional-valued-only
ValerianRey
authored
change `outpust` of `jac`
PierreQuinton
committed
Add changelog entry
ValerianRey
committed
Fix formatting
ValerianRey
committed