Brianguenter/issue67

Project.toml

@@ -38,10 +38,10 @@ TestItems = "0.1"
 julia = "1.8"
 
 [extras]
+Memoize = "c03570c3-d221-55d1-a50c-7939bbd78826"
+Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
 TestItemRunner = "f8b46487-2199-4994-9208-9a1283c18c0a"
 TestItems = "1c621080-faea-4a02-84b6-bbd5e436b8fe"
-Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"
-Memoize = "c03570c3-d221-55d1-a50c-7939bbd78826"
 
 [targets]
-test = ["Test","TestItemRunner","TestItems","Memoize"]
+test = ["Test", "TestItemRunner", "TestItems", "Memoize"]

src/DerivativeGraph.jl

@@ -168,16 +168,21 @@ struct DerivativeGraph{T<:Integer}
     function DerivativeGraph(roots::AbstractVector, index_type::Type=Int64)
         postorder_number = IdDict{Node,index_type}()
 
-        (postorder_number, nodes, var_array) = postorder(roots)
+        new_roots = Vector{Node}(undef, length(roots))
+        for (i, root) in pairs(roots)
+            new_roots[i] = create_NoOp(root)
+        end
+
+        (postorder_number, nodes, var_array) = postorder(new_roots)
 
         expression_cache = IdDict()
 
-        edges = partial_edges(roots, postorder_number, expression_cache, length(var_array), length(roots))
+        edges = partial_edges(new_roots, postorder_number, expression_cache, length(var_array), length(new_roots))
 
         sort!(var_array, by=x -> postorder_number[x]) #sort by postorder number from lowest to highest
 
-        root_index_to_postorder_number = Vector{index_type}(undef, length(roots)) #roots are handled differently than variables because variable node can only occur once in list of variables but root node can occur multiple times in list of roots
-        for (i, x) in pairs(roots)
+        root_index_to_postorder_number = Vector{index_type}(undef, length(new_roots)) #new_roots are handled differently than variables because variable node can only occur once in list of variables but root node can occur multiple times in list of new_roots
+        for (i, x) in pairs(new_roots)
             root_index_to_postorder_number[i] = postorder_number[x]
         end
 
@@ -205,7 +210,7 @@ struct DerivativeGraph{T<:Integer}
         return new{index_type}(
             postorder_number,
             nodes,
-            roots,
+            new_roots,
             var_array,
             root_index_to_postorder_number,
             root_postorder_to_index,

src/ExpressionGraph.jl

@@ -386,6 +386,22 @@ derivative(::typeof(+), args::NTuple{N,Any}, ::Val{I}) where {I,N} = Node(1)
 
 function_variable_derivative(a::Node, index::Val{i}) where {i} = check_cache((Differential, children(a)[i]), EXPRESSION_CACHE)
 
+"""When constructing `DerivativeGraph` with repeated values in roots, e.g.,
+```julia
+@variables x
+f = sin(x)
+gr = DerivativeGraph([f,f,f])
+```
+all three of the f values reference the same element. To ensure that `partial_edges` creates an edge for each one of the roots we need a `NoOp` function. The derivative of `NoOp` is 1.0; the sole purpose of this node type is to ensure that the resulting derivative graph has a separate edge for each repeated root value. There are other ways this might be accomplished but this is the simplest, since it can be performed on the original `Node` graph before the recursive `partial_edges` traversal."""
+struct NoOp
+end
+
+function create_NoOp(child)
+    return Node(NoOp(), child)
+end
+
+derivative(NoOp, arg::Tuple{T}, ::Val{1}) where {T} = 1.0
+
 function derivative(a::Node, index::Val{1})
     # if is_variable(a)
     #     if arity(a) == 0