Merge branch 'master' of https://git01.codeplex.com/cntk into dongyu/dev

Conflicts:
	.gitignore
	Documentation/CNTK-TechReport/lyx/CNTKBook_CNTK_Adv_Chapter.lyx
	Documentation/CNTK-TechReport/lyx/CNTKBook_CNTK_Programmer_Chapter.lyx

.gitignore

@@ -176,3 +176,4 @@ MachineLearning/CNTK/buildinfo.h
 MachineLearning/CNTK/buildinfo.h$$
 
 
+*.lyx#

CheckInSuites/MNIST/mnistCheckIn.config

@@ -61,7 +61,7 @@ mnistTrainSimpleNet=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_train.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]
@@ -81,7 +81,7 @@ mnistTrainSimpleNet=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_train.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]
@@ -101,7 +101,7 @@ mnistTrainSimpleNet=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_test.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]
@@ -147,7 +147,7 @@ mnistTrainNDLNet=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_train.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]
@@ -167,7 +167,7 @@ mnistTrainNDLNet=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_train.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]
@@ -187,7 +187,7 @@ mnistTrainNDLNet=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_test.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]
@@ -231,7 +231,7 @@ mnistAdaptSimpleNet=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_train.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]
@@ -251,7 +251,7 @@ mnistAdaptSimpleNet=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_train.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]
@@ -271,7 +271,7 @@ mnistAdaptSimpleNet=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_test.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]
@@ -302,7 +302,7 @@ mnistSimpleNetTest=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_test.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]
@@ -337,7 +337,7 @@ mnistSimpleNetCV=[
 	    dim=1
 		start=0
 		file=$DataFolder$\mnist_test.txt
-	    labelMappingFile=$ConfigFolder$\mnistlabels.txt
+	    labelMappingFile=$DataFolder$\mnistlabels.txt
 		labelDim=10
 		labelType=Category
 	  ]

Documentation/CNTK-TechReport/lyx/CNTKBook_CNTK_Adv_Chapter.lyx

@@ -692,14 +692,26 @@ EvalNodes=(ErrPredict)
 OutputNodes=(Plus2)
 \end_layout
 
+\begin_layout Plain Layout
+
+NodesReqMultiSeqHandling=(CE)
+\end_layout
+
 \end_inset
 
 After defining the network, it’s important to let CNTK know what the special
  nodes are in the network.
- For example, CNTK needs to know which input nodes are features and which
- are labels.
+ For example, CNTK needs to know which input nodes are features and labels.
  It also needs to know the default output nodes, evaluation nodes and training
  criteria nodes.
+ Note here the specification of the nodes that require special handling
+ (NodesReqMultiSeqHandling) when the network is evalauted or trained with
+ multiple sequences, e.g., when the network itself is an RNN or the model
+ is trained with the sequence-level criterion.
+ Since in these cases multiple sequences will be stitched together to improve
+ the speed and special handling is needed for, e.g., the criterion node, to
+ reset the RNN states at the right time and to mask out the samples when
+ the features/labels are missing.
  CNTK supports multiple inputs and outputs which can be represented by comma
  separated variable names surrounded by parentheses.
 \end_layout
@@ -1254,6 +1266,11 @@ CriteriaNodes=(CE)
 EvalNodes=(ErrPredict)
 \end_layout
 
+\begin_layout Plain Layout
+
+NodesReqMultiSeqHandling=(CE)
+\end_layout
+
 \end_inset
 
 we can tag these nodes as they are defined.
@@ -1284,6 +1301,11 @@ CE = SMBFF(L3, LDim, HDim, labels, tag=Criteria)
 Err=ErrorPrediction(labels, CE.F, tag=Eval)
 \end_layout
 
+\begin_layout Plain Layout
+
+CE = SMBFF(L3, LDim, HDim, labels, tag=MultiSeq)
+\end_layout
+
 \end_inset
 
 
@@ -2144,6 +2166,47 @@ startRow - the start row to get a slice
 numRows - the number of rows to get
 \end_layout
 
+\begin_layout Subsubsection
+RowStack
+\begin_inset Index idx
+status open
+
+\begin_layout Plain Layout
+RowStack
+\end_layout
+
+\end_inset
+
+
+\end_layout
+
+\begin_layout Standard
+Concatnate rows of input matrices to form a bigger matrix.
+ The resulting matrix is a sumof(rows) by m1.cols matrix.
+ It supports variable-length input.
+ The syntax is
+\end_layout
+
+\begin_layout Standard
+\begin_inset listings
+inline false
+status open
+
+\begin_layout Plain Layout
+
+RowStack(m1, m2, ...)
+\end_layout
+
+\end_inset
+
+
+\end_layout
+
+\begin_layout Itemize
+m1, m2 - input matrices.
+ can be any number of input matrices.
+\end_layout
+
 \begin_layout Subsubsection
 Scale
 \begin_inset Index idx
@@ -5521,6 +5584,21 @@ Output
 =true|false: Set the node as one of the output nodes.
 \end_layout
 
+\begin_layout Itemize
+MultiSeq
+\begin_inset Index idx
+status open
+
+\begin_layout Plain Layout
+MultiSeq
+\end_layout
+
+\end_inset
+
+=true|false: Set the node as one of the nodes that require special handling
+ when multiple sequences are used in the training or evaluation.
+\end_layout
+
 \begin_layout Subsubsection
 
 \series bold

Documentation/CNTK-TechReport/lyx/CNTKBook_CNTK_Programmer_Chapter.lyx

@@ -3675,6 +3675,58 @@ mType>& inputGradientValues, const Matrix<ElemType>& gradientValues)
 which contains the actual gradient computation code.
 \end_layout
 
+\begin_layout Subsubsection
+Customization of the Multi-Sequence Handling Code
+\begin_inset Index idx
+status open
+
+\begin_layout Plain Layout
+Customization of the Multi-Sequence Handling Code
+\end_layout
+
+\end_inset
+
+
+\end_layout
+
+\begin_layout Standard
+If your node will generate an output (function values) that has different
+ number of columns (recall that each column is a sample) than the input
+ (e.g., all the ciretrion nodes will generate a scalar value as the output),
+ you need to add the protected virtual function
+\end_layout
+
+\begin_layout Standard
+\begin_inset listings
+inline false
+status open
+
+\begin_layout Plain Layout
+
+    protected:         
+\end_layout
+
+\begin_layout Plain Layout
+
+         virtual bool UseCustomizedMultiSeqHandling() { return true; }
+\end_layout
+
+\end_inset
+
+
+\end_layout
+
+\begin_layout Standard
+This function indicates that you will use your customized code to handle
+ the condition in which multi-sequences are used in each minibatch (e.g.,
+ when training an RNN).
+ You need to add the MaskToZeroWhenLabelAndFeatureMissing calls at the approriat
+e places in your code to mask out both the function values and the gradient
+ values when a segment of the minibatch does not have features/labels.
+ An example of the customized handling code can be found inside the CrossEntropy
+WithSoftmaxNode.
+\end_layout
+
 \begin_layout Subsubsection
 The CNTKMath Library
 \begin_inset Index idx