cleanup using

ZIB-IOL · matbesancon · Nov 16, 2023 · Nov 16, 2023 · Nov 16, 2023 · Nov 16, 2023
commit 987143327644d6f7c9af8eef930112348f99e335
diff --git a/examples/adaptive_step_sizes.jl b/examples/adaptive_step_sizes.jl
@@ -1,9 +1,4 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
 
 import LinearAlgebra
 

diff --git a/examples/adaptive_step_sizes_nontrivial.jl b/examples/adaptive_step_sizes_nontrivial.jl
@@ -1,7 +1,5 @@
 using FrankWolfe
-using ProgressMeter
 using Random
-using Plots
 
 using LinearAlgebra
 

diff --git a/examples/approximate_caratheodory.jl b/examples/approximate_caratheodory.jl
@@ -1,11 +1,5 @@
 
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using LinearAlgebra
 
 n = Int(1e2);

diff --git a/examples/away_step_cg.jl b/examples/away_step_cg.jl
@@ -1,11 +1,5 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
-import LinearAlgebra
+using LinearAlgebra
 
 include("../examples/plot_utils.jl")
 
@@ -17,7 +11,7 @@ xpi = rand(n);
 total = sum(xpi);
 const xp = xpi # ./ total;
 
-f(x) = LinearAlgebra.norm(x - xp)^2
+f(x) = norm(x - xp)^2
 function grad!(storage, x)
     @. storage = 2 * (x - xp)
 end

diff --git a/examples/birkhoff_polytope.jl b/examples/birkhoff_polytope.jl
@@ -1,10 +1,4 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using LinearAlgebra
 using Random
 import GLPK

diff --git a/examples/blended_cg.jl b/examples/blended_cg.jl
@@ -1,10 +1,4 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using LinearAlgebra
 using Random
 using SparseArrays

diff --git a/examples/blended_pairwise_vs_afw_sparity.jl b/examples/blended_pairwise_vs_afw_sparity.jl
@@ -10,12 +10,6 @@ The default for AFW is K = 2.0
 =#
 
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using LinearAlgebra
 using Random
 

diff --git a/examples/large_scale.jl b/examples/large_scale.jl
@@ -11,12 +11,6 @@ NOTE.
 =#
 
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using LinearAlgebra
 
 n = Int(1e7)

diff --git a/examples/lazy_away_step_cg-sparsity.jl b/examples/lazy_away_step_cg-sparsity.jl
@@ -10,12 +10,6 @@ The default for AFW is lazy_tolerance = 2.0
 =#
 
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using LinearAlgebra
 using Random
 

diff --git a/examples/lazy_away_step_cg.jl b/examples/lazy_away_step_cg.jl
@@ -1,11 +1,5 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
-import LinearAlgebra
+using LinearAlgebra
 
 include("../examples/plot_utils.jl")
 
@@ -18,7 +12,7 @@ xpi = rand(n);
 total = sum(xpi);
 const xp = xpi # ./ total;
 
-f(x) = LinearAlgebra.norm(x - xp)^2
+f(x) = norm(x - xp)^2
 function grad!(storage, x)
     @. storage = 2 * (x - xp)
 end

diff --git a/examples/lcg_cache_size.jl b/examples/lcg_cache_size.jl
@@ -1,8 +1,4 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-
 using LinearAlgebra
 
 n = Int(1e4)

diff --git a/examples/lcg_expensive_LMO.jl b/examples/lcg_expensive_LMO.jl
@@ -1,9 +1,5 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
 using Plots
-using DoubleFloats
-using ReverseDiff
 
 using LinearAlgebra
 using Random
@@ -25,7 +21,7 @@ const xp = xpi # ./ total;
 # better for memory consumption as we do coordinate-wise ops
 
 function cf(x, xp)
-    return LinearAlgebra.norm(x .- xp)^2 / n^2
+    return norm(x .- xp)^2 / n^2
 end
 
 function cgrad!(storage, x, xp)

diff --git a/examples/lcg_expensive_LMO_results.jl b/examples/lcg_expensive_LMO_results.jl
@@ -6,12 +6,6 @@
 # compile time of Julia for the first run is not tainting the results
 
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using JSON
 using LaTeXStrings
 results = JSON.Parser.parsefile("lcg_expensive_data.json")

diff --git a/examples/linear_regression.jl b/examples/linear_regression.jl
@@ -4,12 +4,6 @@ using the stochastic version of Frank-Wolfe.
 """
 
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using Random
 using LinearAlgebra
 

diff --git a/examples/lowerbound.jl b/examples/lowerbound.jl
@@ -32,13 +32,7 @@ NOTE:
 =#
 
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
-import LinearAlgebra
+using LinearAlgebra
 
 include("../examples/plot_utils.jl")
 
@@ -49,7 +43,7 @@ k = Int(1e3)
 xp = 1 / n * ones(n);
 
 # definition of objective
-f(x) = LinearAlgebra.norm(x - xp)^2
+f(x) = norm(x - xp)^2
 
 # definition of gradient
 function grad!(storage, x)

diff --git a/examples/moi_optimizer.jl b/examples/moi_optimizer.jl
@@ -5,12 +5,6 @@
 # to get accurate timings it is important to run twice so that the compile time of Julia for the first run
 # is not tainting the results
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using LinearAlgebra
 using LaTeXStrings
 

diff --git a/examples/movielens.jl b/examples/movielens.jl
@@ -1,9 +1,4 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
 
 # download movielens data
 using ZipFile, DataFrames, CSV

diff --git a/examples/movielens_results.jl b/examples/movielens_results.jl
@@ -6,14 +6,9 @@
 # is not tainting the results
 
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using JSON
 using LaTeXStrings
+
 results = JSON.Parser.parsefile("movielens_result.json")
 
 ref_optimum = results["trajectory_arr_lazy_ref"][end][2]

diff --git a/examples/nonconvex_lasso.jl b/examples/nonconvex_lasso.jl
@@ -1,11 +1,5 @@
 
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using LinearAlgebra
 
 

diff --git a/examples/nuclear_norm.jl b/examples/nuclear_norm.jl
@@ -1,10 +1,4 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 import Random
 using SparseArrays, LinearAlgebra
 using Test

diff --git a/examples/polynomial_results.jl b/examples/polynomial_results.jl
@@ -1,11 +1,5 @@
 
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using JSON
 using LaTeXStrings
 

diff --git a/examples/polynomials.jl b/examples/polynomials.jl
@@ -1,10 +1,4 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
 using LinearAlgebra
 import Random
 

diff --git a/examples/trajectory_comparison.jl b/examples/trajectory_comparison.jl
@@ -1,11 +1,5 @@
 using FrankWolfe
-using ProgressMeter
-using Arpack
-using Plots
-using DoubleFloats
-using ReverseDiff
-
-import LinearAlgebra
+using LinearAlgebra
 
 include("../examples/plot_utils.jl")
 
@@ -16,17 +10,16 @@ xpi = rand(n);
 total = sum(xpi);
 const xp = xpi ./ total;
 
-f(x) = LinearAlgebra.norm(x - xp)^2
+f(x) = norm(x - xp)^2
 
 function grad!(storage, x)
     storage .= 2 * (x - xp)
     return nothing
 end
 
 # better for memory consumption as we do coordinate-wise ops
-
 function cf(x, xp)
-    return @. LinearAlgebra.norm(x - xp)^2
+    return norm(x .- xp)^2
 end
 
 # lmo = FrankWolfe.KSparseLMO(100, 1.0)

diff --git a/src/FrankWolfe.jl b/src/FrankWolfe.jl
@@ -2,9 +2,6 @@ module FrankWolfe
 
 using GenericSchur
 using LinearAlgebra
-using Printf
-using ProgressMeter
-using TimerOutputs
 using SparseArrays: spzeros, SparseVector
 import SparseArrays
 import Random

diff --git a/src/afw.jl b/src/afw.jl
@@ -365,7 +365,7 @@ function away_frank_wolfe(
         callback(state, active_set)
     end
 
-    return x, v, primal, dual_gap, traj_data, active_set
+    return (x=x, v=v, primal=primal, dual_gap=dual_gap, traj_data=traj_data, active_set=active_set)
 end
 
 function lazy_afw_step(x, gradient, lmo, active_set, phi, epsilon, d; use_extra_vertex_storage=false, extra_vertex_storage=nothing, lazy_tolerance=2.0, memory_mode::MemoryEmphasis=InplaceEmphasis())