Add ONNX support for zeros_like, ones_like, and eye_like.

daniel-s-ingram · Dec 5, 2018 · 22e869e · 22e869e
1 parent 5fc6c2a
commit 22e869e
Show file tree

Hide file tree

Showing 7 changed files with 74 additions and 10 deletions.
diff --git a/Source/CNTKv2LibraryDll/API/CNTKLibrary.h b/Source/CNTKv2LibraryDll/API/CNTKLibrary.h
@@ -4161,7 +4161,12 @@ namespace CNTK
     /// Create an instance of the expand dims operation on specified tensor input operand, for the specified axis 
     ///
     CNTK_API FunctionPtr ExpandDims(const Variable& operand, const Axis& axis, const std::wstring& name = L"");
-
+
+    //
+    /// Create an instance of a constant-like operation. This produces a tensor with given constant value with the shape and dynamic axes specified by the operand.
+    ///
+    CNTK_API FunctionPtr ConstantLike(const Variable& operand, const double value, const std::wstring& name = L"");
+
     ///
     /// Create an instance of a zeros-like operation. This produces zeros with the shape and dynamic axes specified by the operand.
     ///

diff --git a/Source/CNTKv2LibraryDll/Function.cpp b/Source/CNTKv2LibraryDll/Function.cpp
@@ -1714,20 +1714,22 @@ namespace CNTK
         return AsBlock(std::move(result), { { operandPlaceholder, operand }}, L"ExpandDims", name);
     }
 
-    FunctionPtr ZerosLike(const Variable& operand, const std::wstring& name)
+    FunctionPtr ConstantLike(const Variable& operand, const double value, const std::wstring& name)
     {
         auto additionalProperties = Dictionary();
-        additionalProperties[PrimitiveFunctionAttribute::AttributeNameFillValue] = 0.0;
+        additionalProperties[PrimitiveFunctionAttribute::AttributeNameFillValue] = value;
 
         return UnaryOp(PrimitiveOpType::ConstantOp, operand, std::move(additionalProperties), name);
     }
 
-    FunctionPtr OnesLike(const Variable& operand, const std::wstring& name)
+    FunctionPtr ZerosLike(const Variable& operand, const std::wstring& name)
     {
-        auto additionalProperties = Dictionary();
-        additionalProperties[PrimitiveFunctionAttribute::AttributeNameFillValue] = 1.0;
+        return ConstantLike(operand, 0.0, name);
+    }
 
-        return UnaryOp(PrimitiveOpType::ConstantOp, operand, std::move(additionalProperties), name);
+    FunctionPtr OnesLike(const Variable& operand, const std::wstring& name)
+    {
+        return ConstantLike(operand, 1.0, name);
     }
 
     FunctionPtr CustomProxyOp(const std::vector<Variable>& operands, const std::wstring& customOp, const NDShape& outputShape, DataType outputType, const std::wstring& name)

diff --git a/Source/CNTKv2LibraryDll/proto/onnx/CNTKToONNX.cpp b/Source/CNTKv2LibraryDll/proto/onnx/CNTKToONNX.cpp
@@ -5743,6 +5743,19 @@ void CNTKToONNXHelper::CopyAttributes(const FunctionPtr& src, onnxruntime::Node*
             size_t k = src->Attributes()[L"numItems"].Value<size_t>();
             node->AddAttribute(attributesMap[L"numItems"], static_cast<int64_t>(k));
         }
+        else if (src->OpName() == L"EyeLikeOp")
+        {
+            bool isOutputSparse = src->Attributes().Contains(L"OutputSparse") ? (bool)src->Attributes()[L"OutputSparse"].Value<bool>() : false;
+            if(isOutputSparse)
+                LogicError("Node '%S': 'OutputSparse' is True. Sparse format export not supported.", src->AsString().c_str());
+        }
+        else if (src->OpName() == L"ConstantOp")
+        {
+            if(!src->Attributes().Contains(L"fillValue"))
+                LogicError("Node '%S': 'fillValue' not present. Cannot export op.", src->AsString().c_str());
+            auto fillValue = static_cast<float>(src->Attributes()[L"fillValue"].Value<double>());
+            node->AddAttribute("value", fillValue);
+        }
         else if (src->OpName() == L"Crop")
         {
             const NDShape& inputShape = src->Inputs()[0].Shape();
@@ -6155,8 +6168,6 @@ onnxruntime::Node* CNTKToONNXHelper::AddNode(const FunctionPtr& src, onnxruntime
         //
         if (src->OpName() == L"Times")
         {
-            if (src->Uid() == L"Times4771")
-                std::cout << "";
             size_t py_api_output_rank_argument = src->Attributes()[L"outputRank"].Value<size_t>();
             auto input1Shape = orderedInputs[0]->Shape();
             auto input2Shape = orderedInputs[1]->Shape();

diff --git a/Source/CNTKv2LibraryDll/proto/onnx/ONNXToCNTK.cpp b/Source/CNTKv2LibraryDll/proto/onnx/ONNXToCNTK.cpp
@@ -2929,6 +2929,21 @@ FunctionPtr ONNXToCNTKHelper::CreateFunction(const Node *node, const std::vector
         FunctionPtr cntkFunction = TopK(inputs[0], k, axis, ToFixedWStringFromMultiByte(node->Name()));
         return cntkFunction;
     }
+    else if (onnxOpName == "EyeLike")
+    {
+        // Only limited import support is provided.
+        FunctionPtr cntkFunction = EyeLike(inputs[0], false, ToFixedWStringFromMultiByte(node->Name()));
+        return cntkFunction;
+    }
+    else if (onnxOpName == "ConstantLike")
+    {
+        // Limited import support implemented. 'shape' attribute 
+        // node syntax not supported. Only syntax with input tensor
+        // for shape and 'value' attribute for value is supported.
+        float value = GetNamedAttributeAsFloat(node, "value", 0.0f);
+        FunctionPtr cntkFunction = ConstantLike(inputOperand0, static_cast<double>(value), ToFixedWStringFromMultiByte(node->Name()));
+        return cntkFunction;
+    }
     else if (onnxOpName == "Crop")
     {
         // inputShape: [W, H, C] x [N]

diff --git a/Source/CNTKv2LibraryDll/proto/onnx/Operators.cpp b/Source/CNTKv2LibraryDll/proto/onnx/Operators.cpp
@@ -477,6 +477,12 @@ namespace ONNX
         { L"OneHotOp", { {
             { L"OneHotOp", "OneHotEncoder"},
         } } },
+        { L"EyeLikeOp",{ {
+            { L"EyeLikeOp", "EyeLike" },
+        } } },
+        { L"ConstantOp",{ {
+            { L"ConstantOp", "ConstantLike" },
+        } } },
     };
 
     // given a cntkOpName and cntk attribute OpName which is saved in CNTK::Function's attribute,

diff --git a/bindings/python/cntk/tests/onnx_op_test.py b/bindings/python/cntk/tests/onnx_op_test.py
@@ -2062,4 +2062,28 @@ def test_Crop_Manual(tmpdir, dtype):
     y = C.constant(np.ones((1,2,1), dtype=np.float32))
     model = C.crop_manual(x, y, 1, 2, name='crop_manual')
     data = np.asarray(range(4*4), dtype=np.float32).reshape((1,4,4))
-    verify_one_input(model, data, tmpdir, "Crop_Manual_0")
+    verify_one_input(model, data, tmpdir, "Crop_Manual_0")
+
+# eye_like
+@pytest.mark.parametrize("dtype", DType_Config)
+def test_Eye_Like(tmpdir, dtype):
+    x = C.input_variable((4, 4), dynamic_axes=[], dtype=dtype, name='feature')
+    model = C.eye_like(x, sparse_output=False)
+    data = np.asarray(range(4*4), dtype=dtype).reshape((4,4))
+    verify_one_input(model, data, tmpdir, "Eye_Like_0")
+
+# zeros_like
+@pytest.mark.parametrize("dtype", DType_Config)
+def test_Zeros_Like(tmpdir, dtype):
+    x = C.input_variable((3, 4), dynamic_axes=[], dtype=dtype, name='feature')
+    model = C.zeros_like(x, name='zeros_like_op')
+    data = np.asarray(range(3*4), dtype=dtype).reshape((3,4))
+    verify_one_input(model, data, tmpdir, "Zeros_Like_0")
+
+# ones_like
+@pytest.mark.parametrize("dtype", DType_Config)
+def test_Ones_Like(tmpdir, dtype):
+    x = C.input_variable((3, 4), dynamic_axes=[], dtype=dtype, name='feature')
+    model = C.ones_like(x, name='ones_like_op')
+    data = np.asarray(range(3*4), dtype=dtype).reshape((3,4))
+    verify_one_input(model, data, tmpdir, "Ones_Like_0")
diff --git a/bindings/python/cntk/tests/onnx_verify_helper.py b/bindings/python/cntk/tests/onnx_verify_helper.py
@@ -30,6 +30,7 @@
     'MVN_1',
     'MVN_2',
     'MVN_3',
+    'Eye_Like_0',
 ]
 
 def parse_single_result_case(case_str):