maximizer, statistical model and tests

Author: BatyLeo

Project.toml

@@ -27,7 +27,6 @@ SparseArrays = "2f01184e-e22b-5df5-ae63-d93ebab69eaf"
 Statistics = "10745b16-79ce-11e8-11f9-7d13ad32a3b2"
 
 [compat]
-ConstrainedShortestPaths = "0.5.1"
 DataDeps = "0.7"
 Distributions = "0.25"
 DocStringExtensions = "0.9"

src/StochasticVehicleScheduling/StochasticVehicleScheduling.jl

@@ -5,6 +5,7 @@ using DocStringExtensions: TYPEDEF, TYPEDFIELDS, TYPEDSIGNATURES
 using ConstrainedShortestPaths:
     stochastic_routing_shortest_path, stochastic_routing_shortest_path_with_threshold
 using Distributions: Distribution, LogNormal, Uniform
+using Flux: Chain, Dense
 using Graphs:
     AbstractGraph,
     SimpleDiGraph,
@@ -44,6 +45,8 @@ include("solution/solution.jl")
 include("solution/exact_algorithms/mip.jl")
 include("solution/exact_algorithms/column_generation.jl")
 
+include("maximizer.jl")
+
 """
 $TYPEDFIELDS
 
@@ -80,8 +83,12 @@ function Utils.generate_dataset(
     ]
 end
 
-function Utils.generate_maximizer(bench::StochasticVehicleSchedulingBenchmark) end
-function Utils.generate_statistical_model(bench::StochasticVehicleSchedulingBenchmark) end
+function Utils.generate_maximizer(bench::StochasticVehicleSchedulingBenchmark)
+    return vsp_maximizer
+end
+function Utils.generate_statistical_model(bench::StochasticVehicleSchedulingBenchmark)
+    return Chain(Dense(20 => 1; bias=false), vec)
+end
 
 export StochasticVehicleSchedulingBenchmark
 

src/StochasticVehicleScheduling/maximizer.jl

@@ -0,0 +1,45 @@
+"""
+$TYPEDSIGNATURES
+
+Given arcs weights θ, solve the deterministic VSP problem associated to `instance`.
+"""
+function vsp_maximizer(
+    θ::AbstractVector; instance::Instance, model_builder=highs_model, silent=true
+)
+    (; graph) = instance
+
+    model = model_builder()
+    silent && set_silent(model)
+
+    nb_nodes = nv(graph)
+    job_indices = 2:(nb_nodes - 1)
+
+    @variable(model, y[i=1:nb_nodes, j=1:nb_nodes; has_edge(graph, i, j)], Bin)
+
+    @objective(
+        model,
+        Max,
+        sum(θ[i] * y[src(edge), dst(edge)] for (i, edge) in enumerate(edges(graph)))
+    )
+
+    @constraint(
+        model,
+        flow[i in job_indices],
+        sum(y[j, i] for j in inneighbors(graph, i)) ==
+            sum(y[i, j] for j in outneighbors(graph, i))
+    )
+    @constraint(
+        model, demand[i in job_indices], sum(y[j, i] for j in inneighbors(graph, i)) == 1
+    )
+
+    optimize!(model)
+
+    solution = falses(ne(graph))
+    for (i, edge) in enumerate(edges(graph))
+        if isapprox(value(y[edge.src, edge.dst]), 1; atol=1e-3)
+            solution[i] = true
+        end
+    end
+
+    return solution
+end

test/vsp.jl

@@ -0,0 +1,27 @@
+@testitem "Stochastic VSP" begin
+    using DecisionFocusedLearningBenchmarks
+    using DecisionFocusedLearningBenchmarks.StochasticVehicleScheduling
+    using Graphs
+
+    b = StochasticVehicleSchedulingBenchmark(; nb_tasks=25, nb_scenarios=10)
+
+    N = 50
+    dataset = generate_dataset(b, N)
+    @test length(dataset) == N
+
+    maximizer = generate_maximizer(b)
+    model = generate_statistical_model(b)
+
+    for sample in dataset
+        x = sample.x
+        instance = sample.instance
+        E = ne(instance.graph)
+        @test size(x) == (20, E)
+        θ = model(x)
+        @test length(θ) == E
+        y = maximizer(θ; instance=instance)
+        @test length(y) == E
+        solution = StochasticVehicleScheduling.Solution(y, instance)
+        @test is_feasible(solution, instance)
+    end
+end