Subsampled tests on hardcoded solutions

src/solve.jl

@@ -244,7 +244,7 @@ function pwq_dp_constrained{T}(ℓ::AbstractTransitionCost{T}, V_N::QuadraticPol
 
     N = size(ℓ, 2)
 
-    @assert M-1 <= N "Cannot have more segments than N-1."
+    @assert M <= N-1 "Cannot have more segments than N-1."
 
     Λ = Array{PiecewiseQuadratic{T}}(M, N)
 
@@ -570,9 +570,9 @@ i.e. so that `f[t[I]] .= Y`.
 `lazy` = true, means that the internal transition costs `ℓ[i,j]` will be calculated when needed.
 `tol` specifies the relative tolerance sent to `quadg` kused when calculating the integrals (continuous case).
 """
-function  fit_pwl_regularized(g::AbstractArray, ζ; lazy=true)
-    ℓ = lazy ? TransitionCostDiscrete{Float64}(g) :
-               compute_discrete_transition_costs(g)
+function  fit_pwl_regularized(g::AbstractArray, ζ; t=1:length(g), lazy=true)
+    ℓ = lazy ? TransitionCostDiscrete{Float64}(g, t) :
+               compute_discrete_transition_costs(g, t)
     # Discrete case, so cost at endpoint is quadratic
     cost_last = QuadraticPolynomial(1.0, -2*g[end], g[end]^2)
     fit_pwl_regularized_internal(ℓ, cost_last, ζ)

test/fit_pwl_constrained.jl

@@ -28,4 +28,21 @@ for problem_fcn in ["discontinuous1",
             @test I_vec[m] == I_sols[m]
         end
     end
+
+    # Iterate only over problems where solution exists
+    @testset "Data set: $problem_fcn, subsampled, constrained fit with m=$m" for m in 1:length(f_sols)
+        # Generate max(length(g)/5, m) random gridpoints in range 1:length(g)
+        randI = shuffle(1:length(g))[1:max(floor(Int,length(g)/5), m+1)]
+        # Merge with correct solution, and [1,length(g)] if solution doesn't exist
+        t_grid = sort(union(randI, I_sols[m], [1, length(g)]))
+        # Solve on subsampled grid,
+        I_vec, _, f_vec = fit_pwl_constrained(g, m, t = t_grid)
+        # Should be same if we had a solution
+        if !isempty(I_sols[m])
+            @test f_vec[m] ≈ f_sols[m]  atol=1e-10
+            @test t_grid[I_vec[m]] == I_sols[m]
+        else # Or higher cost if not
+            @test f_vec[m] >= f_sols[m] || isapprox(f_vec[m], f_sols[m], atol=1e-10)
+        end
+    end
 end

test/fit_pwl_regularized.jl

@@ -37,4 +37,26 @@ for problem_fcn in ["discontinuous1",
 
 
     end
+
+    # Iterate only over problems where solution exists
+    @testset "Data set: $problem_fcn, subsampled, regularization with ζ=$ζ" for ζ in ζ_vec
+        # Use costs in the solution file to find out how many segments the solution should contain
+        m_expected = indmin([f_sols[m] + m*ζ for m=1:length(f_sols)])
+
+        # Generate max(length(g)/5, m) random gridpoints in range 1:length(g)
+        randI = shuffle(1:length(g))[1:max(floor(Int,length(g)/5), m_expected+1)]
+        # Merge with correct solution, and [1,length(g)] if solution doesn't exist
+        t_grid = sort(union(randI, I_sols[m_expected], [1, length(g)]))
+        # Solve on subsampled grid
+        I_reg, _, f_reg = fit_pwl_regularized(g, ζ; t = t_grid)
+
+        # Should be same if we had a solution
+        if !isempty(I_sols[m_expected])
+            @test m_expected == length(I_reg) - 1
+            @test f_reg ≈ f_sols[m_expected]  atol=1e-10
+            @test t_grid[I_reg] == I_sols[m_expected]
+        else
+            @test f_reg >= f_sols[m_expected] || isapprox(f_reg, f_sols[m_expected], atol=1e-10)
+        end
+    end
 end

test/problems/discontinuous1.jl

@@ -33,7 +33,7 @@ I_sols = [
 [1, 10, 11, 15, 16, 20],
 [1, 5, 10, 11, 15, 16, 20],
 [1, 3, 8, 10, 11, 15, 16, 20],
-[],
+Int[],
 [1, 2, 4, 7, 9, 10, 11, 15, 16, 20]]
 
 f_sols = [

test/problems/super_exponential5.jl

@@ -29,17 +29,17 @@ I_sols = [
 [1, 3, 16, 18],
 [1, 3, 15, 17, 18],
 [1, 2, 4, 15, 17, 18],
-[],
+Int[],
 [1, 2, 3, 5, 14, 16, 17, 18],
 [1, 2, 3, 4, 6, 14, 16, 17, 18],
 [1, 2, 3, 4, 6, 13, 15, 16, 17, 18],
 [1, 2, 3, 4, 6, 12, 14, 15, 16, 17, 18],
 [1, 2, 3, 4, 5, 7, 12, 14, 15, 16, 17, 18],
-[],
+Int[],
 [1, 2, 3, 4, 5, 6, 7, 12, 13, 14, 15, 16, 17, 18],
-[],
+Int[],
 [1, 2, 3, 4, 5, 6, 7, 8, 11, 12, 13, 14, 15, 16, 17, 18],
-[],
+Int[],
 [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18]]
 
 f_sols = [