Move ARTrustRegion to SolverTools (#116)

Author: tmigot

Project.toml

@@ -9,6 +9,7 @@ LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 LinearOperators = "5c8ed15e-5a4c-59e4-a42b-c7e8811fb125"
 NLPModels = "a4795742-8479-5a88-8948-cc11e1c8c1a6"
 SolverCore = "ff4d7338-4cf1-434d-91df-b86cb86fb843"
+SolverTools = "b5612192-2639-5dc1-abfe-fbedd65fab29"
 SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
 SparseMatricesCOO = "fa32481b-f100-4b48-8dc8-c62f61b13870"
 Stopping = "c4fe5a9e-e7fb-5c3d-89d5-7f405ab2214f"
@@ -19,6 +20,7 @@ Krylov = "^0.8.2, 0.9"
 LinearOperators = "1, 2"
 NLPModels = "0.18, 0.19, 0.20"
 SolverCore = "0.2, 0.3"
+SolverTools = "^0.8.8"
 SparseMatricesCOO = "0.1, 0.2"
 Stopping = "0.6.4"
 StoppingInterface = "0.3, 0.4, 0.5"

src/AdaptiveRegularization.jl

@@ -3,14 +3,13 @@ module AdaptiveRegularization
 # stdlib
 using LinearAlgebra, SparseArrays
 # JSO
-using Krylov, LinearOperators, NLPModels, SparseMatricesCOO, SolverCore
+using Krylov, LinearOperators, NLPModels, SparseMatricesCOO, SolverCore, SolverTools
 # Stopping
 using Stopping, StoppingInterface
 
 # Selective includes.
 include("./utils/hessian_rep.jl")
 include("./utils/pdata_struct.jl")
-include("./utils/utils.jl")
 include("./utils/increase_decrease.jl")
 
 path = joinpath(dirname(@__FILE__), "SolveModel")
@@ -48,7 +47,7 @@ Some variants of TRARC are already implemented and listed in `AdaptiveRegulariza
 - `nlp::AbstractNLPModel`: the model solved, see `NLPModels.jl`.
 
 The keyword arguments include
-- `TR::TrustRegion`: structure with trust-region/ARC parameters, see [`TrustRegion`](@ref). Default: `TrustRegion(T(10.0))`.
+- `TR::ARTrustRegion`: structure with trust-region/ARC parameters, see [`SolverTools.jl`](https://github.com/JuliaSmoothOptimizers/SolverTools.jl). Default: `ARTrustRegion(T(10.0))`.
 - `hess_type::Type{Hess}`: Structure used to handle the hessian. The possible values are: `HessDense`, `HessSparse`, `HessSparseCOO`, `HessOp`. Default: `HessOp`.
 - `pdata_type::Type{ParamData}` Structure used for the preprocessing step. Default: `PDataKARC`.
 - `robust::Bool`: `true` implements a robust evaluation of the model. Default: `true`.

src/main.jl

@@ -172,7 +172,7 @@ function SolverCore.solve!(
 
       iter += 1
 
-      r, good_grad, ∇fnext = compute_r(nlp, ft, Δf, Δq, slope, d, xtnext, workspace, robust)
+      r, good_grad, ∇fnext = SolverTools.compute_r(nlp, ft, Δf, Δq, slope, d, xtnext, workspace, robust)
 
       if Δq < 0.0 # very unsucessful
         verbose > 0 &&

src/trarc_solver.jl

@@ -88,7 +88,7 @@ The keyword arguments may include:
 - `stp::NLPStopping`: `Stopping` structure for this algorithm workflow;
 - `meta::ParamData`: see [`ParamData`](@ref);
 - `workspace::TRARCWorkspace`: allocated space for the solver itself;
-- `TR::TrustRegion`: trust-region parameters.
+- `TR::ARTrustRegion`: trust-region parameters.
 
 """
 mutable struct TRARCSolver{
@@ -106,7 +106,7 @@ mutable struct TRARCSolver{
   stp::NLPStopping{Pb, M, SRC, NLPAtX{Score, T, S}, MStp, LoS}
   meta::Data
   workspace::TRARCWorkspace{T, S, Hess}
-  TR::TrustRegion{T}
+  TR::ARTrustRegion{T}
 end
 
 export TRARCSolver
@@ -118,7 +118,7 @@ end
 
 function TRARCSolver(
   stp::NLPStopping{Pb, M, SRC, NLPAtX{Score, T, S}, MStp, LoS};
-  TR::TrustRegion = TrustRegion(T(10.0)),
+  TR::ARTrustRegion = ARTrustRegion(T(10.0)),
   hess_type::Type{Hess} = HessOp,
   pdata_type::Type{ParamData} = PDataKARC,
   kwargs...,

src/utils/increase_decrease.jl

@@ -1,23 +1,23 @@
 """
-    decrease(X::TPData, α::T, TR::TrustRegion)
+    decrease(X::TPData, α::T, TR::ARTrustRegion)
 
 Return a decreased `α`.
 """
-function decrease(X::TPData, α::T, TR::TrustRegion) where {T}
+function decrease(X::TPData, α::T, TR::ARTrustRegion) where {T}
   return α * TR.decrease_factor
 end
 
 """
-    increase(X::TPData, α::T, TR::TrustRegion)
+    increase(X::TPData, α::T, TR::ARTrustRegion)
 
 Return an increased `α`.
 """
-function increase(::TPData, α::T, TR::TrustRegion) where {T}
+function increase(::TPData, α::T, TR::ARTrustRegion) where {T}
   return min(α * TR.increase_factor, TR.max_α)
 end
 
 # X.indmin is between 1 and nshifts
-function decrease(X::PDataKARC, α::T, TR::TrustRegion) where {T}
+function decrease(X::PDataKARC, α::T, TR::ARTrustRegion) where {T}
   X.indmin += 1 # the step wasn't successful so we need to change something
   α2 = max(X.norm_dirs[X.indmin] / X.shifts[X.indmin], eps(T))
 
@@ -41,7 +41,7 @@ function decrease(X::PDataKARC, α::T, TR::TrustRegion) where {T}
   return α2
 end
 
-function decrease(X::PDataTRK, α::T, TR::TrustRegion) where {T}
+function decrease(X::PDataTRK, α::T, TR::ARTrustRegion) where {T}
   X.indmin += 1
   α2 = X.norm_dirs[X.indmin]
 

src/utils/utils.jl

@@ -4,7 +4,7 @@ T = Float64
 S = Vector{T}
 n = 1000
 
-TR = TrustRegion(T(10))
+TR = ARTrustRegion(T(10))
 α = T(100)
 
 # Test increase / decrease

test/allocation_test_utils.jl

@@ -3,7 +3,8 @@ using AdaptiveRegularization
 # stdlib
 using LinearAlgebra, SparseArrays, Test
 # JSO
-using ADNLPModels, NLPModels, OptimizationProblems.ADNLPProblems, SolverCore, SolverTest
+using ADNLPModels, NLPModels, OptimizationProblems.ADNLPProblems
+using SolverCore, SolverTools, SolverTest
 # Stopping
 using Stopping