Create basic interface between computational graph to mindspore framework

src/backend/compiler.jl

@@ -1,4 +1,5 @@
 module Compilers
+using PyCall
 using ..ComputationalGraphs
 import ..ComputationalGraphs: id, name, set_name!, operator, subgraphs, subgraph_factors, factor
 
@@ -8,5 +9,6 @@ using ..RuntimeGeneratedFunctions
 RuntimeGeneratedFunctions.init(Compilers)
 
 include("static.jl")
+include("toMindspore.jl")
 
 end

src/backend/toMindspore.jl

@@ -0,0 +1,122 @@
+# ms = pyimport("mindspore")
+
+"""
+    function to_pystatic(operator::Type, subgraphs::AbstractVector{<:AbstractGraph}, subgraph_factors::AbstractVector)
+
+Returns the static representation of a computational graph node `g` with operator `operator`, subgraphs `subgraphs`, and subgraph factors `subgraph_factors` in python.
+"""
+function to_pystatic(operator::Type, subgraphs::AbstractVector{<:AbstractGraph}, subgraph_factors::AbstractVector)
+    error(
+        "Static representation for computational graph nodes with operator $(operator) not yet implemented! " *
+        "Please define a method `to_static(::Type{$(operator)}, subgraphs::$(typeof(subgraphs)), subgraph_factors::$(typeof(subgraph_factors)))`."
+    )
+end
+
+function to_pystatic(::Type{ComputationalGraphs.Sum}, subgraphs::Vector{Graph{F,W}}, subgraph_factors::Vector{F}) where {F,W}
+    if length(subgraphs) == 1
+        factor_str = subgraph_factors[1] == 1 ? "" : " * $(subgraph_factors[1])"
+        return "(g$(subgraphs[1].id)$factor_str)"
+    else
+        terms = ["g$(g.id)" * (gfactor == 1 ? "" : " * $gfactor") for (g, gfactor) in zip(subgraphs, subgraph_factors)]
+        return "(" * join(terms, " + ") * ")"
+    end
+end
+
+function to_pystatic(::Type{ComputationalGraphs.Prod}, subgraphs::Vector{Graph{F,W}}, subgraph_factors::Vector{F}) where {F,W}
+    if length(subgraphs) == 1
+        factor_str = subgraph_factors[1] == 1 ? "" : " * $(subgraph_factors[1])"
+        return "(g$(subgraphs[1].id)$factor_str)"
+    else
+        terms = ["g$(g.id)" * (gfactor == 1 ? "" : " * $gfactor") for (g, gfactor) in zip(subgraphs, subgraph_factors)]
+        return "(" * join(terms, " * ") * ")"
+        # return "(" * join(["g$(g.id)" for g in subgraphs], " * ") * ")"
+    end
+end
+
+function to_pystatic(::Type{ComputationalGraphs.Power{N}}, subgraphs::Vector{Graph{F,W}}, subgraph_factors::Vector{F}) where {N,F,W}
+    factor_str = subgraph_factors[1] == 1 ? "" : " * $(subgraph_factors[1])"
+    return "((g$(subgraphs[1].id))**$N$factor_str)"
+end
+
+function to_pystatic(::Type{ComputationalGraphs.Sum}, subgraphs::Vector{FeynmanGraph{F,W}}, subgraph_factors::Vector{F}) where {F,W}
+    if length(subgraphs) == 1
+        factor_str = subgraph_factors[1] == 1 ? "" : " * $(subgraph_factors[1])"
+        return "(g$(subgraphs[1].id)$factor_str)"
+    else
+        terms = ["g$(g.id)" * (gfactor == 1 ? "" : " * $gfactor") for (g, gfactor) in zip(subgraphs, subgraph_factors)]
+        return "(" * join(terms, " + ") * ")"
+    end
+end
+
+function to_pystatic(::Type{ComputationalGraphs.Prod}, subgraphs::Vector{FeynmanGraph{F,W}}, subgraph_factors::Vector{F}) where {F,W}
+    if length(subgraphs) == 1
+        factor_str = subgraph_factors[1] == 1 ? "" : " * $(subgraph_factors[1])"
+        return "(g$(subgraphs[1].id)$factor_str)"
+    else
+        terms = ["g$(g.id)" * (gfactor == 1 ? "" : " * $gfactor") for (g, gfactor) in zip(subgraphs, subgraph_factors)]
+        return "(" * join(terms, " * ") * ")"
+    end
+end
+
+function to_pystatic(::Type{ComputationalGraphs.Power{N}}, subgraphs::Vector{FeynmanGraph{F,W}}, subgraph_factors::Vector{F}) where {N,F,W}
+    factor_str = subgraph_factors[1] == 1 ? "" : " * $(subgraph_factors[1])"
+    return "((g$(subgraphs[1].id))**$N$factor_str)"
+end
+
+"""
+    function to_julia_str(graphs::AbstractVector{<:AbstractGraph})
+
+Compile a list of graphs into a string for a python static function and output a python script which support the static graph representation in mindspore framework.
+"""
+
+function to_python_str_ms(graphs::AbstractVector{<:AbstractGraph})
+    head = "import mindspore as ms\[email protected]\n"
+    head *= "def graphfunc():\n"
+    body = "    graph_list = []\n"
+    leafidx = 1
+    root = [id(g) for g in graphs]
+    inds_visitedleaf = Int[]
+    inds_visitednode = Int[]
+    for graph in graphs
+        for g in PostOrderDFS(graph) #leaf first search
+            g_id = id(g)
+            target = "g$(g_id)"
+            isroot = false
+            if g_id in root
+                isroot = true
+            end
+            if isempty(subgraphs(g)) #leaf
+                g_id in inds_visitedleaf && continue
+                factor_str = factor(g) == 1 ? "" : " * $(factor(g))"
+                body *= "    $target = ms.Tensor(1.0)$factor_str\n"
+                leafidx += 1
+                push!(inds_visitedleaf, g_id)
+            else
+                g_id in inds_visitednode && continue
+                factor_str = factor(g) == 1 ? "" : " * $(factor(g))"
+                body *= "    $target = $(to_pystatic(operator(g), subgraphs(g), subgraph_factors(g)))$factor_str\n"
+                push!(inds_visitednode, g_id)
+            end
+            if isroot
+                body *= "    graph_list.append($target)\n"
+            end
+        end
+    end
+    tail = "    return graph_list\n"
+    tail *= "output = graphfunc()"
+    expr = head * body * tail
+    # return head * body * tail
+    f = open("GraphFunc.py", "w")
+    write(f, expr)
+end
+
+# function to_mindspore_graph(graphs::AbstractVector{<:AbstractGraph})
+#     pyexpr = to_python_str_ms(graphs)
+#     py"""
+#         import mindspore as ms
+#         exec($pyexpr)
+#         ms_graph = jit(fn=graphfunc)
+#         out = ms_graph()
+#     """
+#     return py"out"
+# end