Add failing tests for ODEsolver when using AutoDiff - need to work ou…

…t how to fix

test/runtests.jl

@@ -550,4 +550,124 @@ using EllipseSampling
             end
         end
     end
+
+    # REAL LIKELIHOOD with ODE solver
+    begin
+        # function solvedmodel(θ, t)
+        #     return (θ[2] * θ[3]) ./ ((θ[2] - θ[3]) .* (exp.(-θ[1] .* t)) .+ θ[3])
+        # end
+
+        function DE!(dC, C, p, t)
+            λ,K=p
+            dC[1]= λ * C[1] * (1.0 - C[1]/K)
+        end
+
+        function odesolver(t, λ, K, C0)
+            p=(λ,K)
+            tspan=(0.0, maximum(t))
+            prob=ODEProblem(DE!, [C0], tspan, p)
+            sol=solve(prob, saveat=t)
+            return sol[1,:]
+        end
+
+        function model(θ, t)
+            y = odesolver(t, θ[1], θ[2], θ[3])
+            return y
+        end
+
+        function loglhood(θ, data)
+            y = model(θ, data.t)
+            e = sum(loglikelihood(data.dist, data.yobs .- y))
+            return e
+        end
+
+        λ, K, C0 = 0.01, 100.0, 10.0
+        t = 0:100:1000
+        σ = 10.0
+
+        λmin, λmax = 0.00, 0.05
+        Kmin, Kmax = 50.0, 150.0
+        C0min, C0max = 0.001, 50.0
+
+        θnames = [:λ, :K, :C0]
+        θG = [λ, K, C0]
+        lb = [λmin, Kmin, C0min]
+        ub = [λmax, Kmax, C0max]
+        par_magnitudes = [0.005, 10, 10]
+
+        ytrue = solvedmodel(θG, t)
+        Random.seed!(12348)
+        yobs = ytrue + σ * randn(length(t))
+        data = (yobs=yobs, σ=σ, t=t, dist=Normal(0, σ))
+
+        @testset "BoundaryIsAZeroRealLikelihood_ODEsolver" begin
+            N = 50
+
+            m = initialise_LikelihoodModel(loglhood, data, θnames, θG, lb, ub, par_magnitudes, show_progress=false,
+                find_zero_atol=0.0)
+            getMLE_ellipse_approximation!(m)
+
+            # UNIVARIATE
+            univariate_confidenceintervals!(m)
+
+            targetll = LikelihoodBasedProfileWiseAnalysis.get_target_loglikelihood(m, 0.95, LogLikelihood(), 1)
+            targetll_standardised = LikelihoodBasedProfileWiseAnalysis.get_target_loglikelihood(m, 0.95, EllipseApprox(), 1)
+
+            for i in 1:3
+                lls = [loglhood(m.uni_profiles_dict[i].interval_points.points[:, j], m.core.data) for j in 1:2]
+                @test isapprox(lls .- targetll, zeros(2), atol=1e-14)
+
+                @test isapprox(m.uni_profiles_dict[i].interval_points.ll .- targetll_standardised, zeros(2), atol=1e-14)
+            end
+
+            # BIVARIATE
+            for method in [IterativeBoundaryMethod(4, 2, 2), RadialRandomMethod(3), RadialMLEMethod(0.0), SimultaneousMethod(), Fix1AxisMethod()]
+                bivariate_confidenceprofiles!(m, N, method=method)
+            end
+
+            targetll = LikelihoodBasedProfileWiseAnalysis.get_target_loglikelihood(m, 0.95, LogLikelihood(), 2)
+
+            for i in 1:6
+                lls = [loglhood(m.biv_profiles_dict[i].confidence_boundary[:, j], m.core.data) for j in 1:N]
+                @test isapprox(lls .- targetll, zeros(N), atol=1e-12)
+            end
+        end
+
+        @testset "ValidDimensionalPoints_RealLikelihood_ODEsolver" begin
+            m = initialise_LikelihoodModel(loglhood, data, θnames, θG, lb, ub, par_magnitudes, show_progress=false)
+
+            # UNIVARIATE
+            dimensional_likelihood_samples!(m, 1, 100, sample_type=UniformGridSamples())
+            dimensional_likelihood_samples!(m, 1, 100, sample_type=UniformRandomSamples())
+            dimensional_likelihood_samples!(m, 1, 100, sample_type=LatinHypercubeSamples())
+
+            targetll_standardised = LikelihoodBasedProfileWiseAnalysis.get_target_loglikelihood(m, 0.95, EllipseApprox(), 1)
+
+            for i in 1:9
+                @test all(m.dim_samples_dict[i].ll .≥ targetll_standardised)
+            end
+
+            # BIVARIATE
+            dimensional_likelihood_samples!(m, 2, 10, sample_type=UniformGridSamples())
+            dimensional_likelihood_samples!(m, 2, 100, sample_type=UniformRandomSamples())
+            dimensional_likelihood_samples!(m, 2, 100, sample_type=LatinHypercubeSamples())
+
+            targetll_standardised = LikelihoodBasedProfileWiseAnalysis.get_target_loglikelihood(m, 0.95, EllipseApprox(), 2)
+
+            for i in 10:18
+                @test all(m.dim_samples_dict[i].ll .≥ targetll_standardised)
+            end
+
+            # FULL
+            dimensional_likelihood_samples!(m, 3, 10, sample_type=UniformGridSamples())
+            dimensional_likelihood_samples!(m, 3, 1000, sample_type=UniformRandomSamples())
+            dimensional_likelihood_samples!(m, 3, 1000, sample_type=LatinHypercubeSamples())
+
+            targetll_standardised = LikelihoodBasedProfileWiseAnalysis.get_target_loglikelihood(m, 0.95, EllipseApprox(), 3)
+
+            for i in 19:21
+                @test all(m.dim_samples_dict[i].ll .≥ targetll_standardised)
+            end
+        end
+    end
 end