[Nonlinear] improve test coverage (#2641)

Author: odow

.codecov.yml

@@ -1 +1,3 @@
 comment: false
+ignore:
+  - 'src/Nonlinear/univariate_expressions_generator.jl'

test/Nonlinear/Nonlinear.jl

@@ -880,6 +880,18 @@ function test_evaluate_subexpressions()
     return
 end
 
+function test_evaluate_manny_arguments()
+    model = MOI.Nonlinear.Model()
+    x = MOI.VariableIndex.(1:20)
+    v = Dict(xi => xi.value for xi in x)
+    expr = MOI.Nonlinear.add_expression(
+        model,
+        Expr(:call, :+, [:(sqrt($(x[i]))) for i in 1:20]...),
+    )
+    @test Nonlinear.evaluate(v, model, expr) ≈ sum(sqrt(i) for i in 1:20)
+    return
+end
+
 function test_NLPBlockData()
     model = Nonlinear.Model()
     x = MOI.VariableIndex(1)
@@ -1251,6 +1263,70 @@ function test_univariate_sign()
     end
 end
 
+function test_show_Model()
+    model = MOI.Nonlinear.Model()
+    @test sprint(show, model) ==
+          "A Nonlinear.Model with:\n 0 objectives\n 0 parameters\n 0 expressions\n 0 constraints"
+    p = MOI.Nonlinear.add_parameter(model, 2.0)
+    @test sprint(show, model) ==
+          "A Nonlinear.Model with:\n 0 objectives\n 1 parameter\n 0 expressions\n 0 constraints"
+    return
+end
+
+function test_set_objective_nothing()
+    model = MOI.Nonlinear.Model()
+    x = MOI.VariableIndex(1)
+    MOI.Nonlinear.set_objective(model, :(sin($x)))
+    @test sprint(show, model) ==
+          "A Nonlinear.Model with:\n 1 objective\n 0 parameters\n 0 expressions\n 0 constraints"
+    MOI.Nonlinear.set_objective(model, nothing)
+    @test sprint(show, model) ==
+          "A Nonlinear.Model with:\n 0 objectives\n 0 parameters\n 0 expressions\n 0 constraints"
+    return
+end
+
+function test_copy_evaluator()
+    model = Nonlinear.Model()
+    x = MOI.VariableIndex(1)
+    evaluator = Nonlinear.Evaluator(model, Nonlinear.ExprGraphOnly(), [x])
+    @test_throws(
+        ErrorException("Copying nonlinear problems not yet implemented"),
+        copy(evaluator),
+    )
+    return
+end
+
+function test_no_objective()
+    model = Nonlinear.Model()
+    x = MOI.VariableIndex(1)
+    evaluator = Nonlinear.Evaluator(model, Nonlinear.ExprGraphOnly(), [x])
+    @test_throws(
+        ErrorException(
+            "Unable to query objective_expr because no nonlinear objective was set",
+        ),
+        MOI.objective_expr(evaluator),
+    )
+    return
+end
+
+function test_convert_to_expr()
+    model = Nonlinear.Model()
+    x = MOI.VariableIndex(1)
+    expr = MOI.Nonlinear.add_expression(model, :(sin($x)))
+    evaluator = Nonlinear.Evaluator(model, Nonlinear.ExprGraphOnly(), [x])
+    @test MOI.Nonlinear.convert_to_expr(
+        evaluator,
+        model[expr];
+        moi_output_format = true,
+    ) == :(sin(x[$x]))
+    @test MOI.Nonlinear.convert_to_expr(
+        evaluator,
+        model[expr];
+        moi_output_format = false,
+    ) == :(sin($x))
+    return
+end
+
 end  # TestNonlinear
 
 TestNonlinear.runtests()

test/Nonlinear/ReverseAD.jl

@@ -1152,6 +1152,87 @@ function test_univariate_operator_with_no_second_order()
     return
 end
 
+function test_no_objective()
+    model = Nonlinear.Model()
+    x = MOI.VariableIndex(1)
+    evaluator = Nonlinear.Evaluator(model, Nonlinear.SparseReverseMode(), [x])
+    MOI.initialize(evaluator, [:Grad])
+    @test_throws(
+        ErrorException("No nonlinear objective."),
+        MOI.eval_objective(evaluator, [1.0]),
+    )
+    g = [0.0]
+    @test_throws(
+        ErrorException("No nonlinear objective."),
+        MOI.eval_objective_gradient(evaluator, g, [1.0]),
+    )
+    return
+end
+
+function test_x_power_1()
+    model = Nonlinear.Model()
+    x = MOI.VariableIndex(1)
+    MOI.Nonlinear.set_objective(model, :($x^1))
+    evaluator = Nonlinear.Evaluator(model, Nonlinear.SparseReverseMode(), [x])
+    MOI.initialize(evaluator, [:Grad, :Hess])
+    @test MOI.eval_objective(evaluator, [2.0]) ≈ 2.0
+    H = [NaN]
+    MOI.eval_hessian_lagrangian(evaluator, H, [2.0], 1.5, Float64[])
+    @test H == [0.0]
+    return
+end
+
+function test_variable_first_node_in_tape()
+    model = Nonlinear.Model()
+    x = MOI.VariableIndex(1)
+    expr = MOI.Nonlinear.add_expression(model, :($x))
+    MOI.Nonlinear.set_objective(model, :(sin($expr)))
+    evaluator = Nonlinear.Evaluator(model, Nonlinear.SparseReverseMode(), [x])
+    MOI.initialize(evaluator, [:Grad, :Jac, :Hess])
+    H = [NaN]
+    MOI.eval_hessian_lagrangian(evaluator, H, [2.0], 1.5, [])
+    @test H ≈ [-1.5 * sin(2.0)]
+    return
+end
+
+function test_subexpression_first_node_in_tape()
+    model = Nonlinear.Model()
+    x = MOI.VariableIndex(1)
+    expr = MOI.Nonlinear.add_expression(model, :($x))
+    expr2 = MOI.Nonlinear.add_expression(model, :($expr))
+    MOI.Nonlinear.set_objective(model, :(sin($expr2)))
+    evaluator = Nonlinear.Evaluator(model, Nonlinear.SparseReverseMode(), [x])
+    MOI.initialize(evaluator, [:Grad, :Jac, :Hess])
+    H = [NaN]
+    MOI.eval_hessian_lagrangian(evaluator, H, [2.0], 1.5, [])
+    @test H ≈ [-1.5 * sin(2.0)]
+    return
+end
+
+function test_parameter_in_hessian()
+    model = Nonlinear.Model()
+    x = MOI.VariableIndex(1)
+    p = MOI.Nonlinear.add_parameter(model, 3.0)
+    MOI.Nonlinear.set_objective(model, :(sin($x + $p)))
+    evaluator = Nonlinear.Evaluator(model, Nonlinear.SparseReverseMode(), [x])
+    MOI.initialize(evaluator, [:Grad, :Jac, :Hess])
+    H = [NaN]
+    MOI.eval_hessian_lagrangian(evaluator, H, [2.0], 1.5, [])
+    @test H ≈ [-1.5 * sin(2.0 + 3.0)]
+    return
+end
+
+function test_unsafe_vector_view()
+    x = Float64[]
+    GC.@preserve x begin
+        view = MOI.Nonlinear.ReverseAD._UnsafeVectorView(x, 3)
+        @test length(x) == 3
+        view[2] = 1.0
+        @test x[2] == 1.0
+    end
+    return
+end
+
 end  # module
 
 TestReverseAD.runtests()