Upgrade ADNLPModels.jl for SparseMatrixColoring.jl v0.4.0

JuliaSmoothOptimizers · Aug 13, 2024 · 13a399d · 13a399d
1 parent 725413c
commit 13a399d
Show file tree

Hide file tree

Showing 3 changed files with 24 additions and 13 deletions.
diff --git a/Project.toml b/Project.toml
@@ -20,5 +20,5 @@ NLPModels = "0.18, 0.19, 0.20, 0.21"
 Requires = "1"
 ReverseDiff = "1"
 SparseConnectivityTracer = "0.6"
-SparseMatrixColorings = "0.3.5"
+SparseMatrixColorings = "0.4.0"
 julia = "^1.6"
diff --git a/src/sparse_hessian.jl b/src/sparse_hessian.jl
@@ -21,7 +21,7 @@ function SparseADHessian(
   ncon,
   c!;
   x0::AbstractVector = rand(nvar),
-  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(),
+  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(decompression=:direct),
   detector::AbstractSparsityDetector = TracerSparsityDetector(),
   kwargs...,
 )
@@ -36,13 +36,17 @@ function SparseADHessian(
   c!,
   H::SparseMatrixCSC{Bool, Int64};
   x0::S = rand(nvar),
-  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(),
+  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(decompression=:direct),
   kwargs...,
 ) where {S}
   T = eltype(S)
 
-  colors, star_set = symmetric_coloring_detailed(H, coloring)
-  ncolors = maximum(colors)
+  problem = ColoringProblem(; structure=:symmetric, partition=:column)
+  result = coloring(H, problem, coloring, decompression_eltype=T)
+  colors = column_colors(result)
+  groups = column_groups(result)
+  ncolors = length(groups)
+  star_set = result.star_set
 
   d = BitVector(undef, nvar)
 
@@ -125,7 +129,7 @@ function SparseReverseADHessian(
   ncon,
   c!;
   x0::AbstractVector = rand(nvar),
-  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(),
+  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(decompression=:direct),
   detector::AbstractSparsityDetector = TracerSparsityDetector(),
   kwargs...,
 )
@@ -140,11 +144,15 @@ function SparseReverseADHessian(
   c!,
   H::SparseMatrixCSC{Bool, Int64};
   x0::AbstractVector{T} = rand(nvar),
-  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(),
+  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(decompression=:direct),
   kwargs...,
 ) where {T}
-  colors, star_set = symmetric_coloring_detailed(H, coloring)
-  ncolors = maximum(colors)
+  problem = ColoringProblem(; structure=:symmetric, partition=:column)
+  result = coloring(H, problem, coloring, decompression_eltype=T)
+  colors = column_colors(result)
+  groups = column_groups(result)
+  ncolors = length(groups)
+  star_set = result.star_set
 
   d = BitVector(undef, nvar)
 

diff --git a/src/sparse_jacobian.jl b/src/sparse_jacobian.jl
@@ -16,7 +16,7 @@ function SparseADJacobian(
   ncon,
   c!;
   x0::AbstractVector = rand(nvar),
-  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(),
+  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(decompression=:direct),
   detector::AbstractSparsityDetector = TracerSparsityDetector(),
   kwargs...,
 )
@@ -32,12 +32,15 @@ function SparseADJacobian(
   c!,
   J::SparseMatrixCSC{Bool, Int64};
   x0::AbstractVector{T} = rand(nvar),
-  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(),
+  coloring::AbstractColoringAlgorithm = GreedyColoringAlgorithm(decompression=:direct),
   kwargs...,
 ) where {T}
   # TODO: use ADTypes.row_coloring instead if you have the right decompression and some heuristic recommends it
-  colors = ADTypes.column_coloring(J, coloring)
-  ncolors = maximum(colors)
+  problem = ColoringProblem(; structure=:nonsymmetric, partition=:column)
+  result = coloring(J, problem, coloring, decompression_eltype=T)
+  colors = column_colors(result)
+  groups = column_groups(result)
+  ncolors = length(groups)
 
   d = BitVector(undef, nvar)