Closed geoffroyleconte closed 1 year ago
Base: 61.79% // Head: 62.00% // Increases project coverage by +0.21%
:tada:
Coverage data is based on head (
854b51a
) compared to base (0b94cfa
). Patch coverage: 100.00% of modified lines in pull request are covered.
:umbrella: View full report at Codecov.
:loudspeaker: Do you have feedback about the report comment? Let us know in this issue.
If ∆ is not used, it's a bug. Those operators should include the indicator of [ℓ, u] ⋂ ∆𝔹.
If ∆ is not used, it's a bug. Those operators should include the indicator of [ℓ, u] ⋂ ∆𝔹.
You used shifted(h, xk, max.(-Δk, l_bound - xk), min.(Δk, u_bound - xk), Δk, selected)
in
https://github.com/JuliaSmoothOptimizers/RegularizedOptimization.jl/pull/81
Should I understand that you want to use this instead: shifted(h, xk, l_bound - xk, u_bound - xk, Δk, selected)
?
With the current state of this PR it changes to shifted(h, xk, max.(-Δk, l_bound - xk), min.(Δk, u_bound - xk), selected)
.
I think the first solution implies calling max
internally anyways. And some algorithms don't need $\Delta \mathbb{B}$ (R2 for example). I just want to make sure that it is the right thing to implement before changing everything, and my PR for the iprox
is also based on the same design.
You used shifted(h, xk, max.(-Δk, l_bound - xk), min.(Δk, u_bound - xk), Δk, selected)
Ah yes, you're right. The way it is now is right because we can use the same box operators with R2 when there are bound constraints.
thank you
For NormL0Box and NormL1Box.
This will be breaking for box operators because
shifted(h, x, l, u, Δ)
becomesshifted(h, x, l, u)
, andset_radius!
will no longer work (it did not really updated the trust region before but will now return an error message indicating that there is no fieldΔ
,set_bounds!
should be used instead). I think thatΔ
was not used in these shifted operators.