Pass solver constructors instead of instances

ericphanson · ericphanson · commit b7da3f640d03 · 2020-02-28T13:15:29.000-06:00
diff --git a/NEWS.md b/NEWS.md
@@ -1,8 +1,10 @@
-# Changes in v0.13.0
+# Major changes in v0.13.0
 
 * The intermediate layer has changed from MathProgBase.jl to
   [MathOptInterface.jl](https://github.com/JuliaOpt/MathOptInterface.jl)
-  ([#330](https://github.com/JuliaOpt/Convex.jl/pull/330)).
+  ([#330](https://github.com/JuliaOpt/Convex.jl/pull/330)). To solve problems,
+  one should pass a MathOptInterface optimizer constructor, such as
+  `SCS.Optimizer`, or `() -> SCS.Optimizer(verbose=false)`.
 * `lambdamin` and `lambdamax` have been deprecated in favor of `eigmin` and
   `eigmax`. ([#357](https://github.com/JuliaOpt/Convex.jl/pull/357))
 * `evaluate(x::Variable)` and `evaluate(c::Constant)` now return scalars and
diff --git a/README.md b/README.md
@@ -40,7 +40,7 @@ x = Variable(n)
 problem = minimize(sumsquares(A * x - b), [x >= 0])
 
 # Solve the problem by calling solve!
-solve!(problem, SCS.Optimizer())
+solve!(problem, SCS.Optimizer)
 
 # Check the status of the problem
 problem.status # :Optimal, :Infeasible, :Unbounded etc.
diff --git a/docs/examples_literate/general_examples/basic_usage.jl b/docs/examples_literate/general_examples/basic_usage.jl
@@ -8,7 +8,7 @@ end
 
 using SCS
 ## passing in verbose=0 to hide output from SCS
-solver = SCS.Optimizer(verbose=0)
+solver = () -> SCS.Optimizer(verbose=0)
 
 # ### Linear program
 #
@@ -73,7 +73,7 @@ p.optval
 
 x = Variable(4)
 p = satisfy(norm(x) <= 100, exp(x[1]) <= 5, x[2] >= 7, geomean(x[3], x[4]) >= x[2])
-solve!(p, SCS.Optimizer(verbose=0))
+solve!(p, solver)
 println(p.status)
 x.value
 
@@ -82,7 +82,7 @@ x.value
 
 y = Semidefinite(2)
 p = maximize(eigmin(y), tr(y)<=6)
-solve!(p, SCS.Optimizer(verbose=0))
+solve!(p, solver)
 p.optval
 
 #-
@@ -108,7 +108,7 @@ y.value
 using GLPK
 x = Variable(4, :Int)
 p = minimize(sum(x), x >= 0.5)
-solve!(p, GLPK.Optimizer())
+solve!(p, GLPK.Optimizer)
 x.value
 
 #-
diff --git a/docs/examples_literate/general_examples/chebyshev_center.jl b/docs/examples_literate/general_examples/chebyshev_center.jl
@@ -25,7 +25,7 @@ p.constraints += a1' * x_c + r * norm(a1, 2) <= b[1];
 p.constraints += a2' * x_c + r * norm(a2, 2) <= b[2];
 p.constraints += a3' * x_c + r * norm(a3, 2) <= b[3];
 p.constraints += a4' * x_c + r * norm(a4, 2) <= b[4];
-solve!(p, SCS.Optimizer(verbose=0))
+solve!(p, () -> SCS.Optimizer(verbose=0))
 p.optval
 
 # Generate the figure
diff --git a/docs/examples_literate/general_examples/control.jl b/docs/examples_literate/general_examples/control.jl
@@ -121,7 +121,7 @@ push!(constraints, velocity[:, T] == 0)
 
 ## Solve the problem
 problem = minimize(sumsquares(force), constraints)
-solve!(problem, SCS.Optimizer(verbose=0))
+solve!(problem, () -> SCS.Optimizer(verbose=0))
 
 # We can plot the trajectory taken by the object.
 
diff --git a/docs/examples_literate/general_examples/huber_regression.jl b/docs/examples_literate/general_examples/huber_regression.jl
@@ -40,18 +40,18 @@ for i=1:length(p_vals)
     fit = norm(beta - beta_true) / norm(beta_true);
     cost = norm(X' * beta - Y);
     prob = minimize(cost);
-    solve!(prob, SCS.Optimizer(verbose=0));
+    solve!(prob, () -> SCS.Optimizer(verbose=0));
     lsq_data[i] = evaluate(fit);
     
     ## Form and solve a prescient regression problem,
     ## i.e., where the sign changes are known.
     cost = norm(factor .* (X'*beta) - Y);
-    solve!(minimize(cost), SCS.Optimizer(verbose=0))
+    solve!(minimize(cost), () -> SCS.Optimizer(verbose=0))
     prescient_data[i] = evaluate(fit);
     
     ## Form and solve the Huber regression problem.
     cost = sum(huber(X' * beta - Y, 1));
-    solve!(minimize(cost), SCS.Optimizer(verbose=0))
+    solve!(minimize(cost), () -> SCS.Optimizer(verbose=0))
     huber_data[i] = evaluate(fit);
 end
 
diff --git a/docs/examples_literate/general_examples/logistic_regression.jl b/docs/examples_literate/general_examples/logistic_regression.jl
@@ -23,7 +23,7 @@ X = hcat(ones(size(iris, 1)), iris.SepalLength, iris.SepalWidth, iris.PetalLengt
 n, p = size(X)
 beta = Variable(p)
 problem = minimize(logisticloss(-Y.*(X*beta)))
-solve!(problem, SCS.Optimizer(verbose=false))
+solve!(problem, () -> SCS.Optimizer(verbose=false))
 
 # Let's see how well the model fits.
 using Plots
diff --git a/docs/examples_literate/general_examples/max_entropy.jl b/docs/examples_literate/general_examples/max_entropy.jl
@@ -23,7 +23,7 @@ b = rand(m, 1);
 
 x = Variable(n);
 problem = maximize(entropy(x), sum(x) == 1, A * x <= b)
-solve!(problem, SCS.Optimizer(verbose=false))
+solve!(problem, () -> SCS.Optimizer(verbose=false))
 problem.optval
 
 #-
diff --git a/docs/examples_literate/general_examples/optimal_advertising.jl b/docs/examples_literate/general_examples/optimal_advertising.jl
@@ -46,7 +46,7 @@ D = Variable(m, n);
 Si = [min(R[i]*dot(P[i,:], D[i,:]'), B[i]) for i=1:m];
 problem = maximize(sum(Si),
                [D >= 0, sum(D, dims=1)' <= T, sum(D, dims=2) >= c]);
-solve!(problem, SCS.Optimizer(verbose=0));
+solve!(problem, () -> SCS.Optimizer(verbose=0));
 
 #-
 
diff --git a/docs/examples_literate/general_examples/robust_approx_fitting.jl b/docs/examples_literate/general_examples/robust_approx_fitting.jl
@@ -43,18 +43,18 @@ x = Variable(n)
 
 # Case 1: Nominal optimal solution
 p = minimize(norm(A * x - b, 2))
-solve!(p, SCS.Optimizer(verbose=0))
+solve!(p, () -> SCS.Optimizer(verbose=0))
 x_nom = evaluate(x)
 
 # Case 2: Stochastic robust approximation
 P = 1 / 3 * B' * B;
 p = minimize(square(pos(norm(A * x - b))) + quadform(x, Symmetric(P)))
-solve!(p, SCS.Optimizer(verbose=0))
+solve!(p, () -> SCS.Optimizer(verbose=0))
 x_stoch = evaluate(x)
 
 # Case 3: Worst-case robust approximation
 p = minimize(max(norm((A - B) * x - b), norm((A + B) * x - b)))
-solve!(p, SCS.Optimizer(verbose=0))
+solve!(p, () -> SCS.Optimizer(verbose=0))
 x_wc = evaluate(x)
 
 # Plot residuals:
diff --git a/docs/examples_literate/general_examples/svm.jl b/docs/examples_literate/general_examples/svm.jl
@@ -34,7 +34,7 @@ neg_data = rand(MvNormal([-1.0, 2.0], 1.0), M);
 
 #-
 
-function svm(pos_data, neg_data, solver=SCS.Optimizer(verbose=0))
+function svm(pos_data, neg_data, solver=() -> SCS.Optimizer(verbose=0))
     ## Create variables for the separating hyperplane w'*x = b.
     w = Variable(n)
     b = Variable()
diff --git a/docs/examples_literate/general_examples/svm_l1regularization.jl b/docs/examples_literate/general_examples/svm_l1regularization.jl
@@ -34,7 +34,7 @@ beta_vals = zeros(length(beta), TRIALS);
 for i = 1:TRIALS
     lambda = lambda_vals[i];
     problem = minimize(loss/m + lambda*reg);
-    solve!(problem, ECOS.Optimizer(verbose=0));
+    solve!(problem, () -> ECOS.Optimizer(verbose=0));
     ## solve!(problem, SCS.Optimizer(verbose=0,linear_solver=SCS.Direct, eps=1e-3))
     train_error[i] = sum(float(sign.(X*beta_true .+ offset) .!= sign.(evaluate(X*beta - v))))/m;
     test_error[i] = sum(float(sign.(X_test*beta_true .+ offset) .!= sign.(evaluate(X_test*beta - v))))/TEST;
diff --git a/docs/examples_literate/general_examples/trade_off_curves.jl b/docs/examples_literate/general_examples/trade_off_curves.jl
@@ -18,7 +18,7 @@ x = Variable(n);
 for i=1:length(gammas)
     cost = sumsquares(A*x - b) + gammas[i]*norm(x,1);
     problem = minimize(cost, [norm(x, Inf) <= 1]);
-    solve!(problem, SCS.Optimizer(verbose=0));
+    solve!(problem, () -> SCS.Optimizer(verbose=0));
     x_values[:,i] = evaluate(x);
 end
 
diff --git a/docs/examples_literate/general_examples/worst_case_analysis.jl b/docs/examples_literate/general_examples/worst_case_analysis.jl
@@ -18,7 +18,7 @@ w = Variable(n);
 ret = dot(r, w);
 risk = sum(quadform(w, Sigma_nom));
 problem = minimize(risk, [sum(w) == 1, ret >= 0.1, norm(w, 1) <= 2])
-solve!(problem, SCS.Optimizer(verbose=0));
+solve!(problem, () -> SCS.Optimizer(verbose=0));
 wval = vec(evaluate(w))
 
 #-
@@ -31,7 +31,7 @@ problem = maximize(risk, [Sigma == Sigma_nom + Delta,
                     diag(Delta) == 0,
                     abs(Delta) <= 0.2,
                     Delta == Delta']);
-solve!(problem, SCS.Optimizer(verbose=0));
+solve!(problem, () -> SCS.Optimizer(verbose=0));
 println("standard deviation = ", round(sqrt(wval' * Sigma_nom * wval), sigdigits=2));
 println("worst-case standard deviation = ", round(sqrt(evaluate(risk)), sigdigits=2));
 println("worst-case Delta = ");
diff --git a/docs/examples_literate/mixed_integer/aux_files/antidiag.jl b/docs/examples_literate/mixed_integer/aux_files/antidiag.jl
@@ -9,6 +9,8 @@
 # Please read expressions.jl first.
 #############################################################################
 import Convex.sign, Convex.monotonicity, Convex.curvature, Convex.evaluate, Convex.conic_form!
+using Convex: AbstractExpr, Nondecreasing, ConstVexity, UniqueConicForms, has_conic_form, cache_conic_form, get_conic_form
+using LinearAlgebra, SparseArrays
 export antidiag
 
 ### Diagonal
diff --git a/docs/examples_literate/mixed_integer/binary_knapsack.jl b/docs/examples_literate/mixed_integer/binary_knapsack.jl
@@ -24,5 +24,5 @@ n = length(w)
 using Convex, GLPK
 x = Variable(n, :Bin)
 problem = maximize(dot(p, x), dot(w, x) <= C)
-solve!(problem, GLPK.Optimizer())
+solve!(problem, GLPK.Optimizer)
 evaluate(x)
diff --git a/docs/examples_literate/mixed_integer/n_queens.jl b/docs/examples_literate/mixed_integer/n_queens.jl
@@ -16,7 +16,7 @@ constr += Constraint[sum(diag(x, k)) <= 1 for k = -n+2:n-2]
 ## Exactly one queen per row and one queen per column
 constr += Constraint[sum(x, dims=1) == 1, sum(x, dims=2) == 1]
 p = satisfy(constr)
-solve!(p, GLPK.Optimizer())
+solve!(p, GLPK.Optimizer)
 
 # Let us test the results:
 for k = -n+2:n-2
diff --git a/docs/examples_literate/mixed_integer/section_allocation.jl b/docs/examples_literate/mixed_integer/section_allocation.jl
@@ -30,5 +30,5 @@ constraints = [sum(X, dims=2) == 1, sum(X, dims=1) <= 10, sum(X, dims=1) >= 6]
 # students since the ranking of the first choice is 1.
 p = minimize(vec(X)' * vec(P), constraints)
 
-solve!(p, GLPK.Optimizer())
+solve!(p, GLPK.Optimizer)
 p.optval
diff --git a/docs/examples_literate/optimization_with_complex_variables/Fidelity in Quantum Information Theory.jl b/docs/examples_literate/optimization_with_complex_variables/Fidelity in Quantum Information Theory.jl
@@ -33,7 +33,7 @@ Z = ComplexVariable(n,n)
 objective = 0.5*real(tr(Z+Z'))
 constraint = [P Z;Z' Q] ⪰ 0
 problem = maximize(objective,constraint)
-solve!(problem, SCS.Optimizer(verbose=0))
+solve!(problem, () -> SCS.Optimizer(verbose=0))
 computed_fidelity = evaluate(objective)
 
 #-
diff --git a/docs/examples_literate/optimization_with_complex_variables/phase_recovery_using_MaxCut.jl b/docs/examples_literate/optimization_with_complex_variables/phase_recovery_using_MaxCut.jl
@@ -69,7 +69,7 @@ objective = inner_product(U,M)
 c1 = diag(U) == 1 
 c2 = U in :SDP
 p = minimize(objective,c1,c2)
-solve!(p, SCS.Optimizer(verbose=0))
+solve!(p, () -> SCS.Optimizer(verbose=0))
 U.value
 
 #-
diff --git a/docs/examples_literate/optimization_with_complex_variables/povm_simulation.jl b/docs/examples_literate/optimization_with_complex_variables/povm_simulation.jl
@@ -55,7 +55,7 @@ function get_visibility(K)
     constraints += t*K[3] + (1-t)*noise[3] == P[2][2] + P[4][2] + P[6][1]
     constraints += t*K[4] + (1-t)*noise[4] == P[3][2] + P[5][2] + P[6][2]
     p = maximize(t, constraints)
-    solve!(p, SCS.Optimizer(verbose=0))
+    solve!(p, () -> SCS.Optimizer(verbose=0))
     return p.optval
 end
 
diff --git a/docs/examples_literate/optimization_with_complex_variables/power_flow_optimization.jl b/docs/examples_literate/optimization_with_complex_variables/power_flow_optimization.jl
@@ -25,7 +25,7 @@ c3 = real(W[1,1]) == 1.06^2;
 push!(c1, c2)
 push!(c1, c3)
 p = maximize(objective, c1);
-solve!(p, SCS.Optimizer(verbose = 0))
+solve!(p, () -> SCS.Optimizer(verbose = 0))
 p.optval
 #15.125857662600703
 evaluate(objective)
diff --git a/docs/examples_literate/portfolio_optimization/portfolio_optimization.jl b/docs/examples_literate/portfolio_optimization/portfolio_optimization.jl
@@ -56,7 +56,7 @@ p = minimize( risk,
               w_lower <= w, 
               w <= w_upper )
 
-solve!(p, SCS.Optimizer())     #use SCS.Optimizer(verbose = false) to suppress printing
+solve!(p, () -> SCS.Optimizer())     #use SCS.Optimizer(verbose = false) to suppress printing
 
 #-
 
diff --git a/docs/examples_literate/portfolio_optimization/portfolio_optimization2.jl b/docs/examples_literate/portfolio_optimization/portfolio_optimization2.jl
@@ -52,7 +52,7 @@ for i = 1:N
     λ = λ_vals[i]
     p = minimize( λ*risk - (1-λ)*ret,
                   sum(w) == 1 )
-    solve!(p, SCS.Optimizer(verbose = false))
+    solve!(p, () -> SCS.Optimizer(verbose = false))
     MeanVarA[i,:]= [evaluate(ret),evaluate(risk)]
 end
 
@@ -68,7 +68,7 @@ for i = 1:N
                   sum(w) == 1,
                   w_lower <= w,     #w[i] is bounded
                   w <= w_upper )
-    solve!(p, SCS.Optimizer(verbose = false))
+    solve!(p, () -> SCS.Optimizer(verbose = false))
     MeanVarB[i,:]= [evaluate(ret),evaluate(risk)]
 end
 
@@ -95,7 +95,7 @@ for i = 1:N
     p = minimize( λ*risk - (1-λ)*ret,
                   sum(w) == 1,
                   (norm(w, 1)-1) <= Lmax)
-    solve!(p, SCS.Optimizer(verbose = false))
+    solve!(p, () -> SCS.Optimizer(verbose = false))
     MeanVarC[i,:]= [evaluate(ret),evaluate(risk)]
 end
 
diff --git a/docs/examples_literate/supplemental_material/Convex.jl_intro_ISMP2015.jl b/docs/examples_literate/supplemental_material/Convex.jl_intro_ISMP2015.jl
@@ -41,7 +41,7 @@ problem = minimize(square(norm(A * x - b)) + λ*square(norm(x)) + μ*norm(x, 1),
                    x >= 0)
 
 ## Solve the problem by calling solve!
-solve!(problem, SCS.Optimizer(verbose=0))
+solve!(problem, () -> SCS.Optimizer(verbose=0))
 
 println("problem status is ", problem.status) # :Optimal, :Infeasible, :Unbounded etc.
 println("optimal value is ", problem.optval)
@@ -54,7 +54,7 @@ using Interact, Plots
     global A
     problem = minimize(square(norm(A * x - b)) + λ*square(norm(x)) + μ*norm(x, 1),
                    x >= 0)
-    solve!(problem, SCS.Optimizer(verbose=0))
+    solve!(problem, () -> SCS.Optimizer(verbose=0))
     histogram(evaluate(x), xlims=(0,3.5), label="x")
 end
 
@@ -128,7 +128,7 @@ p = minimize(objective, constraint)
 #-
 
 ## solve the problem
-solve!(p, SCS.Optimizer(verbose=0))
+solve!(p, () -> SCS.Optimizer(verbose=0))
 p.status
 
 #-
@@ -152,7 +152,7 @@ evaluate(objective)
 # to solve problem using a different solver, just import the solver package and pass the solver to the `solve!` method: eg
 #
 #     using Mosek
-#     solve!(p, Mosek.Optimizer())
+#     solve!(p, Mosek.Optimizer)
 
 #-
 
@@ -173,9 +173,9 @@ x = Variable(n)
 μ = 1
 problem = minimize(square(norm(A * x - b)) + λ*square(norm(x)) + μ*norm(x, 1), 
                    x >= 0)
-@time solve!(problem, SCS.Optimizer(verbose=0))
+@time solve!(problem, () -> SCS.Optimizer(verbose=0))
 λ = 1.5
-@time solve!(problem, SCS.Optimizer(verbose=0), warmstart = true)
+@time solve!(problem, () -> SCS.Optimizer(verbose=0), warmstart = true)
 
 # # DCP examples
 
diff --git a/docs/examples_literate/supplemental_material/paper_examples.jl b/docs/examples_literate/supplemental_material/paper_examples.jl
@@ -13,7 +13,7 @@ e = 0;
   end
   p = minimize(e, x>=1);
 end
-@time solve!(p, ECOS.Optimizer(verbose=0))
+@time solve!(p, () -> ECOS.Optimizer(verbose=0))
 
 # Indexing.
 println("Indexing example")
@@ -26,7 +26,7 @@ e = 0;
   end
   p = minimize(e, x >= ones(1000, 1));
 end
-@time solve!(p, ECOS.Optimizer(verbose=0))
+@time solve!(p, () -> ECOS.Optimizer(verbose=0))
 
 # Matrix constraints.
 println("Matrix constraint example")
@@ -37,7 +37,7 @@ b = randn(p, n);
 @time begin
   p = minimize(norm(vec(X)), A * X == b);
 end
-@time solve!(p, ECOS.Optimizer(verbose=0))
+@time solve!(p, () -> ECOS.Optimizer(verbose=0))
 
 # Transpose.
 println("Transpose example")
@@ -46,12 +46,12 @@ A = randn(5, 5);
 @time begin
   p = minimize(norm2(X - A), X' == X);
 end
-@time solve!(p, ECOS.Optimizer(verbose=0))
+@time solve!(p, () -> ECOS.Optimizer(verbose=0))
 
 n = 3
 A = randn(n, n);
 #@time begin
   X = Variable(n, n);
   p = minimize(norm(vec(X' - A)), X[1,1] == 1);
-  solve!(p, ECOS.Optimizer(verbose=0))
+  solve!(p, () -> ECOS.Optimizer(verbose=0))
 #end
diff --git a/docs/examples_literate/time_series/time_series.jl b/docs/examples_literate/time_series/time_series.jl
diff --git a/docs/examples_literate/tomography/tomography.jl b/docs/examples_literate/tomography/tomography.jl
diff --git a/docs/src/advanced.md b/docs/src/advanced.md
diff --git a/docs/src/complex-domain_optimization.md b/docs/src/complex-domain_optimization.md
diff --git a/docs/src/problem_depot.md b/docs/src/problem_depot.md
diff --git a/docs/src/quick_tutorial.md b/docs/src/quick_tutorial.md
diff --git a/docs/src/solvers.md b/docs/src/solvers.md
diff --git a/docs/src/types.md b/docs/src/types.md
diff --git a/src/solution.jl b/src/solution.jl
diff --git a/test/runtests.jl b/test/runtests.jl
diff --git a/test/test_utilities.jl b/test/test_utilities.jl
diff --git a/test/warmstart.jl b/test/warmstart.jl
