[Nonlinear] improve test coverage of operators.jl (#2650)

Author: odow

src/Nonlinear/operators.jl

@@ -11,42 +11,31 @@ function _create_binary_switch(ids, exprs)
             push!(out.args, _create_binary_switch(ids[2:end], exprs[2:end]))
         end
         return out
-    else
-        mid = length(exprs) >>> 1
-        return Expr(
-            :if,
-            Expr(:call, :(<=), :id, ids[mid]),
-            _create_binary_switch(ids[1:mid], exprs[1:mid]),
-            _create_binary_switch(ids[mid+1:end], exprs[mid+1:end]),
-        )
     end
+    mid = length(exprs) >>> 1
+    return Expr(
+        :if,
+        Expr(:call, :(<=), :id, ids[mid]),
+        _create_binary_switch(ids[1:mid], exprs[1:mid]),
+        _create_binary_switch(ids[mid+1:end], exprs[mid+1:end]),
+    )
 end
 
 # We use a let block here for `expr` to create a local variable that does not
 # persist in the scope of the module. All we care about is the _eval_univariate
 # function that is eval'd as a result.
-let exprs = map(SYMBOLIC_UNIVARIATE_EXPRESSIONS) do arg
+let
+    exprs = map(SYMBOLIC_UNIVARIATE_EXPRESSIONS) do arg
         return :(return $(arg[1])(x), $(arg[2]))
     end
     @eval @inline function _eval_univariate(id, x::T) where {T}
         $(_create_binary_switch(1:length(exprs), exprs))
-        return error("Invalid operator_id")
-    end
-end
-
-# We use a let block here for `expr` to create a local variable that does not
-# persist in the scope of the module. All we care about is the function that is
-# eval'd as a result.
-let exprs = map(SYMBOLIC_UNIVARIATE_EXPRESSIONS) do arg
-        if arg === :(nothing)  # f''(x) isn't defined
-            :(error("Invalid operator_id"))
-        else
-            :(return $(arg[3]))
-        end
+        return error("Invalid id for univariate operator: $id")
     end
+    ∇²f_exprs = map(arg -> :(return $(arg[3])), SYMBOLIC_UNIVARIATE_EXPRESSIONS)
     @eval @inline function _eval_univariate_2nd_deriv(id, x::T) where {T}
-        $(_create_binary_switch(1:length(exprs), exprs))
-        return error("Invalid operator_id")
+        $(_create_binary_switch(1:length(∇²f_exprs), ∇²f_exprs))
+        return error("Invalid id for univariate operator: $id")
     end
 end
 
@@ -339,7 +328,7 @@ function _validate_register_assumptions(
             y = f(zeros(dimension)...)
         end
     catch
-        # We hit some other error, perhaps we called a function like log(0).
+        # We hit some other error, perhaps we called a function like log(-1).
         # Ignore for now, and hope that a useful error is shown to the user
         # during the solve.
     end
@@ -363,7 +352,7 @@ function _validate_register_assumptions(
                 _FORWARD_DIFF_METHOD_ERROR_HELPER,
             )
         end
-        # We hit some other error, perhaps we called a function like log(0).
+        # We hit some other error, perhaps we called a function like log(-1).
         # Ignore for now, and hope that a useful error is shown to the user
         # during the solve.
     end
@@ -747,7 +736,12 @@ function eval_univariate_hessian(
     x::T,
 ) where {T}
     if id <= registry.univariate_user_operator_start
-        return _eval_univariate_2nd_deriv(id, x)::T
+        ret = _eval_univariate_2nd_deriv(id, x)
+        if ret === nothing
+            op = registry.univariate_operators[id]
+            error("Hessian is not defined for operator $op")
+        end
+        return ret::T
     end
     offset = id - registry.univariate_user_operator_start
     operator = registry.registered_univariate_operators[offset]
@@ -910,13 +904,15 @@ _nan_to_zero(x) = isnan(x) ? 0.0 : x
         op::Symbol,
         H::AbstractMatrix,
         x::AbstractVector{T},
-    ) where {T}
+    )::Bool where {T}
 
 Evaluate the Hessian of operator `∇²op(x)`, where `op` is a multivariate
 function in `registry`.
 
 The Hessian is stored in the lower-triangular part of the matrix `H`.
 
+Returns a `Bool` indicating whether non-zeros were stored in the matrix.
+
 !!! note
     Implementations of the Hessian operators will not fill structural zeros.
     Therefore, before calling this function you should pre-populate the matrix

test/Nonlinear/Nonlinear.jl

@@ -374,6 +374,28 @@ function test_eval_univariate_function()
     return
 end
 
+function test_eval_univariate_missing_hessian()
+    r = Nonlinear.OperatorRegistry()
+    x = 2.0
+    @test Nonlinear.eval_univariate_function(r, :asec, x) ≈ asec(x)
+    @test Nonlinear.eval_univariate_gradient(r, :asec, x) ≈
+          1 / (abs(x) * sqrt(x^2 - 1))
+    @test_throws(
+        ErrorException("Hessian is not defined for operator asec"),
+        Nonlinear.eval_univariate_hessian(r, :asec, x),
+    )
+    return
+end
+
+function test_eval_univariate_hessian_bad_id()
+    r = Nonlinear.OperatorRegistry()
+    err = ErrorException("Invalid id for univariate operator: -1")
+    @test_throws err Nonlinear.eval_univariate_function(r, -1, 1.0)
+    @test_throws err Nonlinear.eval_univariate_gradient(r, -1, 1.0)
+    @test_throws err Nonlinear.eval_univariate_hessian(r, -1, 1.0)
+    return
+end
+
 function test_eval_univariate_gradient()
     r = Nonlinear.OperatorRegistry()
     for (op, x, y) in [
@@ -594,7 +616,29 @@ function test_eval_multivariate_gradient_mult()
     x = [1.1, 0.0, 2.2]
     g = zeros(3)
     Nonlinear.eval_multivariate_gradient(r, :*, g, x)
-    @test g == [0.0, 1.1 * 2.2, 0.0]
+    @test g ≈ [0.0, 1.1 * 2.2, 0.0]
+    x = [1.1, 3.3, 2.2]
+    Nonlinear.eval_multivariate_gradient(r, :*, g, x)
+    @test g ≈ [3.3 * 2.2, 1.1 * 2.2, 1.1 * 3.3]
+    return
+end
+
+function test_eval_multivariate_gradient_univariate_mult()
+    r = Nonlinear.OperatorRegistry()
+    x = [1.1]
+    g = zeros(1)
+    Nonlinear.eval_multivariate_gradient(r, :*, g, x)
+    @test g == [1.0]
+    return
+end
+
+function test_eval_multivariate_hessian_shortcut()
+    r = Nonlinear.OperatorRegistry()
+    x = [1.1]
+    H = LinearAlgebra.LowerTriangular(zeros(1, 1))
+    for op in (:+, :-, :ifelse)
+        @test !MOI.Nonlinear.eval_multivariate_hessian(r, op, H, x)
+    end
     return
 end
 
@@ -670,6 +714,18 @@ function test_eval_multivariate_function_registered()
     return
 end
 
+function test_eval_multivariate_function_registered_log()
+    r = Nonlinear.OperatorRegistry()
+    f(x...) = log(x[1] - 1)
+    Nonlinear.register_operator(r, :f, 2, f)
+    x = [1.1, 2.2]
+    @test Nonlinear.eval_multivariate_function(r, :f, x) ≈ f(x...)
+    x = [0.0, 0.0]
+    g = zeros(2)
+    @test_throws DomainError Nonlinear.eval_multivariate_gradient(r, :f, g, x)
+    return
+end
+
 function test_eval_multivariate_function_method_error()
     r = Nonlinear.OperatorRegistry()
     function f(x...)
@@ -1327,6 +1383,27 @@ function test_convert_to_expr()
     return
 end
 
+function test_create_binary_switch()
+    target = Expr(
+        :if,
+        Expr(:call, :(<=), :id, 2),
+        Expr(
+            :if,
+            Expr(:call, :(==), :id, 1),
+            :a,
+            Expr(:if, Expr(:call, :(==), :id, 2), :b),
+        ),
+        Expr(
+            :if,
+            Expr(:call, :(==), :id, 3),
+            :c,
+            Expr(:if, Expr(:call, :(==), :id, 4), :d),
+        ),
+    )
+    @test MOI.Nonlinear._create_binary_switch(1:4, [:a, :b, :c, :d]) == target
+    return
+end
+
 end  # TestNonlinear
 
 TestNonlinear.runtests()