Merge pull request #196 from jump-dev/jg/tests

Simplify Solvers

Author: joaquimg

Project.toml

@@ -23,19 +23,3 @@ LazyArrays = "0.21, 0.22"
 MathOptInterface = "1"
 MathOptSetDistances = "0.2"
 julia = "1.6"
-
-[extras]
-DelimitedFiles = "8bb1440f-4735-579b-a4ab-409b98df4dab"
-Flux = "587475ba-b771-5e3f-ad9e-33799f191a9c"
-GLPK = "60bf3e95-4087-53dc-ae20-288a0d20c6a6"
-Ipopt = "b6b21f68-93f8-5de0-b562-5493be1d77c9"
-MLDatasets = "eb30cadb-4394-5ae3-aed4-317e484a6458"
-OSQP = "ab2f91bb-94b4-55e3-9ba0-7f65df51de79"
-Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
-Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
-SCS = "c946c3f1-0d1f-5ce8-9dea-7daa1f7e2d13"
-Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
-Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
-
-[targets]
-test = ["Test", "DelimitedFiles", "SCS", "Random", "OSQP", "GLPK", "Ipopt", "Statistics", "Flux", "MLDatasets", "Plots"]

README.md

@@ -23,9 +23,9 @@ julia> ]
 ```julia
 using JuMP
 import DiffOpt
-import GLPK
+import HiGHS
 
-model = JuMP.Model(() -> DiffOpt.diff_optimizer(GLPK.Optimizer))
+model = JuMP.Model(() -> DiffOpt.diff_optimizer(HiGHS.Optimizer))
 ```
 
 2. Define your model and solve it a single line.

docs/Project.toml

@@ -2,14 +2,13 @@
 ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"
 Documenter = "e30172f5-a6a5-5a46-863b-614d45cd2de4"
 Flux = "587475ba-b771-5e3f-ad9e-33799f191a9c"
-GLPK = "60bf3e95-4087-53dc-ae20-288a0d20c6a6"
+HiGHS = "87dc4568-4c63-4d18-b0c0-bb2238e4078b"
 Ipopt = "b6b21f68-93f8-5de0-b562-5493be1d77c9"
 JuMP = "4076af6c-e467-56ae-b986-b466b2749572"
 LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
 Literate = "98b081ad-f1c9-55d3-8b20-4c87d4299306"
 MLDatasets = "eb30cadb-4394-5ae3-aed4-317e484a6458"
 MathOptInterface = "b8f27783-ece8-5eb3-8dc8-9495eed66fee"
-OSQP = "ab2f91bb-94b4-55e3-9ba0-7f65df51de79"
 Plots = "91a5bcdd-55d7-5caf-9e0b-520d859cae80"
 SCS = "c946c3f1-0d1f-5ce8-9dea-7daa1f7e2d13"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"

docs/src/examples/Thermal_Generation_Dispatch_Example.jl

@@ -27,7 +27,7 @@ using Test
 using JuMP
 import DiffOpt
 import LinearAlgebra: dot
-import GLPK
+import HiGHS
 import MathOptInterface
 import Plots
 const MOI = MathOptInterface
@@ -36,7 +36,7 @@ const MOI = MathOptInterface
 
 function generate_model(d::Float64; g_sup::Vector{Float64}, c_g::Vector{Float64}, c_ϕ::Float64)
     ## Creation of the Model and Parameters
-    model = Model(() -> DiffOpt.diff_optimizer(GLPK.Optimizer))
+    model = Model(() -> DiffOpt.diff_optimizer(HiGHS.Optimizer))
     set_silent(model)
     I = length(g_sup)
 
@@ -116,20 +116,20 @@ d = 0.0:0.1:80
 d_size = length(d)
 c_g = [1.0, 3.0, 5.0]
 c_ϕ = 10.0
+;
 
 # Generate models for each demand `d`
-models = generate_model.(d; g_sup = g_sup, c_g = c_g, c_ϕ = c_ϕ)
+models = generate_model.(d; g_sup = g_sup, c_g = c_g, c_ϕ = c_ϕ);
 
 # Get the results of models with the DiffOpt Forward and Backward context
 
 result_forward = diff_forward.(models)
-
 optimize!.(models)
-result_backward = diff_backward.(models)
+result_backward = diff_backward.(models);
 
 # Organization of results to plot
 # Initialize data_results that will contain every result
-data_results = Array{Float64,3}(undef, 2, d_size, 2*(I+1))
+data_results = Array{Float64,3}(undef, 2, d_size, 2*(I+1));
 
 # Populate the data_results array
 for k in 1:d_size

docs/src/examples/autotuning-ridge.jl

@@ -26,7 +26,7 @@
 
 using JuMP     # The mathematical programming modelling language
 import DiffOpt # JuMP extension for differentiable optimization
-import OSQP    # Optimization solver that handles quadratic programs
+import Ipopt    # Optimization solver that handles quadratic programs
 import Plots   # Graphing tool
 import LinearAlgebra: norm, dot
 import Random
@@ -66,7 +66,7 @@ function fit_ridge(model, X, y, α)
         dot(err_term, err_term) / (2 * N * D) + α * dot(w, w) / (2 * D),
     )
     optimize!(model)
-    @assert termination_status(model) == MOI.OPTIMAL
+    @assert termination_status(model) in [MOI.OPTIMAL, MOI.LOCALLY_SOLVED, MOI.ALMOST_LOCALLY_SOLVED]
     return w
 end
 
@@ -76,7 +76,7 @@ end
 αs = 0.00:0.01:0.50
 mse_test = Float64[]
 mse_train = Float64[]
-model = Model(() -> DiffOpt.diff_optimizer(OSQP.Optimizer))
+model = Model(() -> DiffOpt.diff_optimizer(Ipopt.Optimizer))
 (Ntest, D) = size(X_test)
 (Ntrain, D) = size(X_train)
 for α in αs
@@ -146,7 +146,7 @@ function descent(α0, max_iters=100; fixed_step = 0.01, grad_tol=1e-3)
     test_loss = Float64[]
     α = α0
     N, D = size(X_test)
-    model = Model(() -> DiffOpt.diff_optimizer(OSQP.Optimizer))
+    model = Model(() -> DiffOpt.diff_optimizer(Ipopt.Optimizer))
     for iter in 1:max_iters
         w = fit_ridge(model, X_train, y_train, α)
         ŵ = value.(w)

docs/src/examples/chainrules_unit.jl

@@ -14,7 +14,7 @@ using JuMP
 import DiffOpt
 import Plots
 import LinearAlgebra: ⋅
-import GLPK
+import HiGHS
 import ChainRulesCore
 
 # ## Unit commitment problem
@@ -41,7 +41,7 @@ import ChainRulesCore
 
 function unit_commitment(
         load1_demand, load2_demand, gen_costs, noload_costs;
-        model = Model(GLPK.Optimizer), silent=false)
+        model = Model(HiGHS.Optimizer), silent=false)
     MOI.set(model, MOI.Silent(), silent)
 
     ## Problem data
@@ -96,7 +96,7 @@ function unit_commitment(
     return JuMP.value.(p.data)
 end
 
-m = Model(GLPK.Optimizer)
+m = Model(HiGHS.Optimizer)
 @show unit_commitment(
     [1.0, 1.2, 1.4, 1.6], [1.0, 1.2, 1.4, 1.6],
     [1000.0, 1500.0], [500.0, 1000.0],
@@ -142,9 +142,9 @@ function ChainRulesCore.frule(
         (_, Δload1_demand, Δload2_demand, Δgen_costs, Δnoload_costs),
         ::typeof(unit_commitment),
         load1_demand, load2_demand, gen_costs, noload_costs;
-        optimizer=GLPK.Optimizer,
+        optimizer=HiGHS.Optimizer,
         )
-    ## creating the UC model with a DiffOpt optimizer wrapper around GLPK
+    ## creating the UC model with a DiffOpt optimizer wrapper around HiGHS
     model = Model(() -> DiffOpt.diff_optimizer(optimizer))
     ## building and solving the main model
     pv = unit_commitment(
@@ -212,7 +212,7 @@ noload_costs = [500.0, 1000.0];
 function ChainRulesCore.rrule(
         ::typeof(unit_commitment),
         load1_demand, load2_demand, gen_costs, noload_costs;
-        optimizer=GLPK.Optimizer,
+        optimizer=HiGHS.Optimizer,
         silent=false)
     model = Model(() -> DiffOpt.diff_optimizer(optimizer))
     ## solve the forward UC problem
@@ -253,7 +253,7 @@ end
 (pv, pullback_unit_commitment) = ChainRulesCore.rrule(
     unit_commitment,
     load1_demand, load2_demand, gen_costs, noload_costs,
-    optimizer=GLPK.Optimizer,
+    optimizer=HiGHS.Optimizer,
     silent=true,
 )
 dpv = 0 * pv

docs/src/examples/custom-relu.jl

@@ -11,7 +11,7 @@
 
 using JuMP
 import DiffOpt
-import OSQP
+import Ipopt
 import ChainRulesCore
 import Flux
 import Statistics
@@ -23,7 +23,7 @@ import Base.Iterators: repeated
 # Return the solution of the problem.
 function matrix_relu(
     y::AbstractArray{T};
-    model = Model(() -> DiffOpt.diff_optimizer(OSQP.Optimizer))
+    model = Model(() -> DiffOpt.diff_optimizer(Ipopt.Optimizer))
 ) where T
     _x = zeros(size(y))
     N = length(y[:, 1])
@@ -47,7 +47,7 @@ end
 function ChainRulesCore.rrule(
     ::typeof(matrix_relu),
     y::AbstractArray{T};
-    model = Model(() -> DiffOpt.diff_optimizer(OSQP.Optimizer))
+    model = Model(() -> DiffOpt.diff_optimizer(Ipopt.Optimizer))
 ) where T
     pv = matrix_relu(y, model = model)
     function pullback_matrix_relu(dl_dx)

docs/src/examples/sensitivity-analysis-ridge.jl

@@ -35,7 +35,7 @@
 using JuMP
 import DiffOpt
 import Random
-import OSQP
+import Ipopt
 import Plots
 import LinearAlgebra: normalize!, dot
 
@@ -55,13 +55,13 @@ Y = w * X .+ b + 0.8 * randn(N);
 # The helper method `fit_ridge` defines and solves the corresponding model.
 # The ridge regression is modeled with quadratic programming
 # (quadratic objective and linear constraints) and solved in generic methods
-# of OSQP. This is not the standard way of solving the ridge regression problem
+# of Ipopt. This is not the standard way of solving the ridge regression problem
 # this is done here for didactic purposes.
 
 function fit_ridge(X, Y, alpha = 0.1)
     N = length(Y)
-    ## Initialize a JuMP Model with OSQP solver
-    model = Model(() -> DiffOpt.diff_optimizer(OSQP.Optimizer))
+    ## Initialize a JuMP Model with Ipopt solver
+    model = Model(() -> DiffOpt.diff_optimizer(Ipopt.Optimizer))
     set_silent(model)
     @variable(model, w) # angular coefficient
     @variable(model, b) # linear coefficient