L2 Penalty Algorithm #145
L2 Penalty Algorithm #145
Conversation
| @@ -20,5 +20,8 @@ include("TRDH_alg.jl") | |||
| include("R2_alg.jl") | |||
| include("LM_alg.jl") | |||
| include("LMTR_alg.jl") | |||
| include("R2DH.jl") | |||
| include("R2N.jl") | |||
There was a problem hiding this comment.
Remove R2N and R2DH from here. Also, if you want to update LM, please do it in a separate PR.
| ktol = max(ktol,atol) # Keep ϵ₀ ≥ ϵ | ||
| tol_init = ktol # store value of ϵ₀ | ||
|
|
||
| done = false |
There was a problem hiding this comment.
Should compute the least-squares multipliers here. If by chance, the solver were initialized with a stationary point, we should exit before the main loop.
| n_iter_since_decrease = 0 | ||
|
|
||
| while !done | ||
| model = RegularizedNLPModel(nlp, sub_ψ) |
There was a problem hiding this comment.
This probably allocates a little. We should think of a way to create it once and update sub_ψ.
| end | ||
| set_solution!(stats,x1[1:n]) | ||
|
|
||
| end |
There was a problem hiding this comment.
Shouldn't this function be in ShiftedProximalOperators?
There was a problem hiding this comment.
I don't really know.
I think it should be considered as a JSO solver since this is what R2N will expect as a subsolver.
Therefore, it should be written here instead of ShiftedProximalOperators but I am open to rewrite it elsewhere if necessary.
src/L2Penalty_alg.jl
Outdated
| H1 = [-Q reg_nlp.h.A'] | ||
| H2 = [reg_nlp.h.A αₖ*opEye(m,m)] | ||
| H = [H1;H2] | ||
| x1,_ = minres_qlp(H,u1) |
There was a problem hiding this comment.
Here, we should explore the potential of other methods, such as TriMR.
First, @MohamedLaghdafHABIBOULLAH should revise #153. |
|
I should also add some unit test for R2N and R2DH. |
| @@ -4,6 +4,8 @@ author = ["Robert Baraldi <[email protected]> and Dominique Orban <dominique.orban | |||
| version = "0.1.0" | |||
|
|
|||
| [deps] | |||
| DelimitedFiles = "8bb1440f-4735-579b-a4ab-409b98df4dab" | |||
There was a problem hiding this comment.
Why do we need this? It shouldn't be required by the solver.
| @@ -104,13 +104,15 @@ function LM( | |||
| k = 0 | |||
| Fobj_hist = zeros(maxIter) | |||
| Hobj_hist = zeros(maxIter) | |||
| R = eltype(xk) | |||
| sqrt_ξ1_νInv = one(R) | |||
| @@ -0,0 +1,297 @@ | |||
| export R2DH | |||
There was a problem hiding this comment.
Let's remove R2N and R2DH from this PR.
This is my implementation of the$\ell_2$ exact penalty algorithm.
Details are provided in the documentation.