fixed 02_Convolutional as well for new tensor TimesNode

Examples/Image/MNIST/Config/02_Convolution.ndl

@@ -51,8 +51,8 @@ DNN=[
     pool2 = MaxPooling(conv2_act, pool2W, pool2H, pool2hStride, pool2vStride, imageLayout=$imageLayout$)
 
     h1Dim = 128
-    # DNNSigmoidLayer and DNNLayer are defined in Macros.ndl
+    # DNNImageSigmoidLayer and DNNLayer are defined in Macros.ndl
-    h1 = DNNSigmoidLayer(512, h1Dim, pool2, 1)
+    h1 = DNNImageSigmoidLayer(4, 4, cMap2, h1Dim, pool2, 1)
     ol = DNNLayer(h1Dim, labelDim, h1, 1)
     
     ce = CrossEntropyWithSoftmax(labels, ol)

Examples/Image/MNIST/Config/Macros.ndl

@@ -6,6 +6,14 @@ DNNSigmoidLayer(inDim, outDim, x, parmScale) = [
     y = Sigmoid(z)
 ]
 
+DNNImageSigmoidLayer(inW, inH, inC, outDim, x, parmScale) = [
+    W = ImageParameter(outDim, inW, inH, inC, init="uniform", initValueScale=parmScale, imageLayout=$imageLayout$)
+    b = LearnableParameter(outDim, 1,         init="uniform", initValueScale=parmScale) 
+    t = Times(W, x)
+    z = Plus(t, b)
+    y = Sigmoid(z)
+]
+
 DNNLayer(inDim, outDim, x, parmScale) = [
     W = LearnableParameter(outDim, inDim, init="uniform", initValueScale=parmScale)
     b = LearnableParameter(outDim, 1,     init="uniform", initValueScale=parmScale)
@@ -26,7 +34,7 @@ DnnBNReLULayer(inDim, outDim, x, wScale, bValue, scValue, expAvg) = [
 
 ConvReLULayer(inp, outMap, inWCount, kW, kH, hStride, vStride, wScale, bValue) = [
     convW = LearnableParameter(outMap, inWCount, init="uniform", initValueScale=wScale)
-    convB = ImageParameter(1, 1, outMap,     init="fixedValue", value=bValue, imageLayout=$imageLayout$)
+    convB = ImageParameter(1, 1, outMap, init="fixedValue", value=bValue, imageLayout=$imageLayout$)
     conv = Convolution(convW, inp, kW, kH, outMap, hStride, vStride, zeroPadding=false, imageLayout=$imageLayout$)
     convPlusB = Plus(conv, convB);
     act = RectifiedLinear(convPlusB);

Source/CNTK/SynchronousExecutionEngine.cpp

@@ -114,7 +114,7 @@ void SynchronousNodeEvaluator<ElemType>::Evaluate(NDLNode<ElemType>* node, const
         else
         {
             if (parameter.size() < 3)
-                RuntimeError("%ls should have 3 parameters [imageWidth, imageHeight, imageChannels] plus other optional parameters (learningRateMultiplier=[1|0|float], init=[uniform|gaussian|fixedvalue], initValueScale=[1|float], value=[0|float]).", cnNodeType.c_str());
+                RuntimeError("%ls should have 3 or more parameters [imageWidth, imageHeight, imageChannels] plus other optional parameters (learningRateMultiplier=[1|0|float], init=[uniform|gaussian|fixedvalue], initValueScale=[1|float], value=[0|float]).", cnNodeType.c_str());
         }
 
         if (pass == ndlPassInitial)

Source/ComputationNetworkLib/LinearAlgebraNodes.h

@@ -298,6 +298,8 @@ public:
                     InvalidArgument("%ls %ls operation: Right [%s] operand must have zero dimensions.", NodeName().c_str(), OperationName().c_str(), dimsBstring.c_str());
                 else if (dimA == 0)
                     dimA = dimB; // infer dimension
+                else if (dimA != dimB)
+                    InvalidArgument("%ls %ls operation: Left [%s] and right [%s] operands' shapes are not compatible.", NodeName().c_str(), OperationName().c_str(), dimsAstring.c_str(), dimsBstring.c_str());
             }
 
             // swap back in case of TransposeTimes
@@ -305,7 +307,6 @@ public:
                 std::swap(dimsA[0], dimsA[1]);
 
             // update if LearnableParameter
-
             Input(0)->ValidateInferInputDimsFrom(TensorShape(dimsA));
 
             // and verify once again

Tests/EndToEndTests/Image/QuickE2E/cntk.cntk

@@ -98,7 +98,8 @@ train = [
 
         h1Dim = 128
         # DNNSigmoidLayer and DNNLayer are defined in Macros.ndl
-        h1 = DNNSigmoidLayer((cMap2 : 4 : 4), h1Dim, pool2, 1).out
+        h1 = DNNSigmoidLayer(if useCuDnn then (4 : 4 : cMap2/*cudnn: CHW*/) else (cMap2 : 4 : 4/*legacy: HWC*/), h1Dim, pool2, 1).out
+        # Note: 'CHW' and 'HWC' refer to row-major representations, while CNTK uses column-major, so must specify the values in reverse order
         ol = DNNLayer(h1Dim, labelDim, h1, 1).out
 
         ce = CrossEntropyWithSoftmax(labels, ol, tag="criterion")