Merge pull request #142 from SciML/wps

Make work-precision benchmarks more robust to failures

Author: ChrisRackauckas

Project.toml

@@ -1,7 +1,7 @@
 name = "DiffEqDevTools"
 uuid = "f3b72e0c-5b89-59e1-b016-84e28bfd966d"
 authors = ["Chris Rackauckas <[email protected]>"]
-version = "2.44.3"
+version = "2.44.4"
 
 [deps]
 DiffEqBase = "2b5f629d-d688-5b77-993f-72d75c75574e"
@@ -28,7 +28,7 @@ Distributed = "1.9"
 LinearAlgebra = "1.9"
 Logging = "1.9"
 NLsolve = "4.2"
-NonlinearSolve = "1, 2"
+NonlinearSolve = "3.13"
 ODEProblemLibrary = "0.1"
 OrdinaryDiffEq = "6"
 ParameterizedFunctions = "5"

src/benchmark.jl

@@ -217,69 +217,75 @@ function WorkPrecision(prob, alg, abstols, reltols, dts = nothing;
 
             stats[i] = sol.stats
 
-            if haskey(kwargs, :prob_choice)
-                cur_appxsol = appxsol[kwargs[:prob_choice]]
-            elseif prob isa AbstractArray
-                cur_appxsol = appxsol[1]
-            else
-                cur_appxsol = appxsol
-            end
-
-            if cur_appxsol !== nothing
-                errsol = appxtrue(sol, cur_appxsol)
-                errors[i] = Dict{Symbol, Float64}()
-                for err in keys(errsol.errors)
-                    errors[i][err] = mean(errsol.errors[err])
-                end
-            else
-                errors[i] = Dict{Symbol, Float64}()
-                for err in keys(sol.errors)
-                    errors[i][err] = mean(sol.errors[err])
+            if SciMLBase.successful_retcode(sol)
+                if haskey(kwargs, :prob_choice)
+                    cur_appxsol = appxsol[kwargs[:prob_choice]]
+                elseif prob isa AbstractArray
+                    cur_appxsol = appxsol[1]
+                else
+                    cur_appxsol = appxsol
                 end
-            end
-
-            benchmark_f = let dts = dts, _prob = _prob, alg = alg, sol = sol,
-                abstols = abstols, reltols = reltols, kwargs = kwargs
 
-                if dts === nothing
-                    if _prob isa DAEProblem
-                        () -> @elapsed solve(_prob, alg, sol.u, sol.t;
-                            abstol = abstols[i],
-                            reltol = reltols[i],
-                            timeseries_errors = false,
-                            dense_errors = false, kwargs...)
-                    else
-                        () -> @elapsed solve(_prob, alg, sol.u, sol.t, sol.k;
-                            abstol = abstols[i],
-                            reltol = reltols[i],
-                            timeseries_errors = false,
-                            dense_errors = false, kwargs...)
+                if cur_appxsol !== nothing
+                    errsol = appxtrue(sol, cur_appxsol)
+                    errors[i] = Dict{Symbol, Float64}()
+                    for err in keys(errsol.errors)
+                        errors[i][err] = mean(errsol.errors[err])
                     end
                 else
-                    if _prob isa DAEProblem
-                        () -> @elapsed solve(_prob, alg, sol.u, sol.t;
-                            abstol = abstols[i],
-                            reltol = reltols[i],
-                            dt = dts[i],
-                            timeseries_errors = false,
-                            dense_errors = false, kwargs...)
+                    errors[i] = Dict{Symbol, Float64}()
+                    for err in keys(sol.errors)
+                        errors[i][err] = mean(sol.errors[err])
+                    end
+                end
+
+                benchmark_f = let dts = dts, _prob = _prob, alg = alg, sol = sol,
+                    abstols = abstols, reltols = reltols, kwargs = kwargs
+
+                    if dts === nothing
+                        if _prob isa DAEProblem
+                            () -> @elapsed solve(_prob, alg, sol.u, sol.t;
+                                abstol = abstols[i],
+                                reltol = reltols[i],
+                                timeseries_errors = false,
+                                dense_errors = false, kwargs...)
+                        else
+                            () -> @elapsed solve(_prob, alg, sol.u, sol.t, sol.k;
+                                abstol = abstols[i],
+                                reltol = reltols[i],
+                                timeseries_errors = false,
+                                dense_errors = false, kwargs...)
+                        end
                     else
-                        () -> @elapsed solve(_prob, alg, sol.u, sol.t, sol.k;
-                            abstol = abstols[i],
-                            reltol = reltols[i],
-                            dt = dts[i],
-                            timeseries_errors = false,
-                            dense_errors = false, kwargs...)
+                        if _prob isa DAEProblem
+                            () -> @elapsed solve(_prob, alg, sol.u, sol.t;
+                                abstol = abstols[i],
+                                reltol = reltols[i],
+                                dt = dts[i],
+                                timeseries_errors = false,
+                                dense_errors = false, kwargs...)
+                        else
+                            () -> @elapsed solve(_prob, alg, sol.u, sol.t, sol.k;
+                                abstol = abstols[i],
+                                reltol = reltols[i],
+                                dt = dts[i],
+                                timeseries_errors = false,
+                                dense_errors = false, kwargs...)
+                        end
                     end
                 end
-            end
-            benchmark_f() # pre-compile
+                benchmark_f() # pre-compile
 
-            b_t = benchmark_f()
-            if b_t > seconds
-                times[i] = b_t
+                b_t = benchmark_f()
+                if b_t > seconds
+                    times[i] = b_t
+                else
+                    times[i] = mapreduce(i -> benchmark_f(), min, 2:numruns; init = b_t)
+                end
             else
-                times[i] = mapreduce(i -> benchmark_f(), min, 2:numruns; init = b_t)
+                # Unsuccessful retcode, give NaN time
+                errors[i] = Dict(:l∞ => NaN, :L2 => NaN, :final => NaN, :l2 => NaN, :L∞ => NaN)
+                times[i] = NaN
             end
         end
     end

src/plotrecipes.jl

@@ -110,6 +110,7 @@ end
         ys = [get_val_from_wp(wp, y) for wp in wp_set.wps]
         xguide --> key_to_label(x)
         yguide --> key_to_label(y)
+        legend --> :outerright
         label --> reshape(wp_set.names, 1, length(wp_set))
         return xs, ys
     elseif view == :dt_convergence

src/test_solution.jl

@@ -117,9 +117,9 @@ function appxtrue(sim::EnsembleSolution, appx_setup; kwargs...)
     for i in eachindex(sim)
         prob = sim[i].prob
         prob2 = SDEProblem(prob.f, prob.g, prob.u0, prob.tspan,
-            noise = NoiseWrapper(sim[i].W))
+            noise = NoiseWrapper(sim.u[i].W))
         true_sol = solve(prob2, appx_setup[:alg]; appx_setup...)
-        _new_sols[i] = appxtrue(sim[i], true_sol)
+        _new_sols[i] = appxtrue(sim.u[i], true_sol)
     end
     new_sols = convert(Vector{typeof(_new_sols[1])}, _new_sols)
     calculate_ensemble_errors(new_sols; converged = sim.converged,

test/benchmark_tests.jl

@@ -23,7 +23,7 @@ setups = [Dict(:alg => RK4()); Dict(:alg => Euler()); Dict(:alg => BS3());
 t1 = @elapsed sol = solve(prob, RK4(), dt = 1 / 2^(4))
 t2 = @elapsed sol2 = solve(prob, setups[1][:alg], dt = 1 / 2^(4))
 
-@test (sol2[end] == sol[end])
+@test (sol2.u[end] == sol.u[end])
 
 test_sol_2Dlinear = TestSolution(
     solve(prob_ode_2Dlinear, Vern7(), abstol = 1 / 10^14, reltol = 1 / 10^14))