Merge branch 'master' of https://github.com/Microsoft/CNTK into amita…

…ga/removeCopyCtorFromMatrix

.gitattributes

@@ -14,6 +14,7 @@ Readme text
 *.bat text
 *.py text
 *.pl text
+*.ps1 text
 *.sh text eol=lf
 build-and-test text eol=lf
 configure text eol=lf

Source/ActionsLib/EvalActions.cpp

@@ -57,6 +57,10 @@ static void DoEvalBase(const ConfigParameters& config, IDataReader& reader)
 
     int traceLevel = config(L"traceLevel", "0");
     size_t numMBsToShowResult = config(L"numMBsToShowResult", "100");
+    size_t maxSamplesInRAM = config(L"maxSamplesInRAM", (size_t)SIZE_MAX);
+    size_t numSubminiBatches = config(L"numSubminibatches", (size_t)1);
+    //TODO: switch to a global parallel setting for both training and evaluation.
+    bool useParallel = config(L"parallelTrain", false);
 
     ConfigArray evalNodeNames = config(L"evalNodeNames", "");
     vector<wstring> evalNodeNamesVector;
@@ -66,8 +70,8 @@ static void DoEvalBase(const ConfigParameters& config, IDataReader& reader)
     }
 
     auto net = ComputationNetwork::CreateFromFile<ElemType>(deviceId, modelPath);
-
-    SimpleEvaluator<ElemType> eval(net, numMBsToShowResult, traceLevel);
+    
+    SimpleEvaluator<ElemType> eval(net, useParallel, numMBsToShowResult, traceLevel, maxSamplesInRAM, numSubminiBatches);
     eval.Evaluate(&reader, evalNodeNamesVector, mbSize[0], epochSize);
 }
 
@@ -114,6 +118,10 @@ void DoCrossValidate(const ConfigParameters& config)
 
     int traceLevel = config(L"traceLevel", "0");
     size_t numMBsToShowResult = config(L"numMBsToShowResult", "100");
+    size_t maxSamplesInRAM = config(L"maxSamplesInRAM", (size_t)SIZE_MAX);
+    size_t numSubminiBatches = config(L"numSubminibatches", (size_t)1);
+    //TODO: switch to a global parallel setting for both training and evaluation.
+    bool useParallel = config(L"parallelTrain", false);
 
     ConfigArray evalNodeNames = config(L"evalNodeNames", "");
     vector<wstring> evalNodeNamesVector;
@@ -146,8 +154,8 @@ void DoCrossValidate(const ConfigParameters& config)
 
         cvModels.push_back(cvModelPath);
         auto net = ComputationNetwork::CreateFromFile<ElemType>(deviceId, cvModelPath);
-
-        SimpleEvaluator<ElemType> eval(net, numMBsToShowResult, traceLevel);
+        
+        SimpleEvaluator<ElemType> eval(net, useParallel, numMBsToShowResult, traceLevel, maxSamplesInRAM, numSubminiBatches);
 
         fprintf(stderr, "model %ls --> \n", cvModelPath.c_str());
         auto evalErrors = eval.Evaluate(&cvDataReader, evalNodeNamesVector, mbSize[0], epochSize);

Source/ActionsLib/OtherActions.cpp

@@ -476,49 +476,50 @@ template void DoWriteWordAndClassInfo<double>(const ConfigParameters& config);
 template <typename ElemType>
 void DoTopologyPlot(const ConfigParameters& config)
 {
-    wstring modelPath = config(L"modelPath");
-    wstring outdot = config(L"outputDotFile");  // filename for the dot language output, if not specified, %modelpath%.dot will be used
-    wstring outRending = config(L"outputFile"); // filename for the rendered topology plot
+    wstring modelPath     = config(L"modelPath");
+    wstring outputDotFile = config(L"outputDotFile"); // filename for the dot language output, if not specified, %modelpath%.dot will be used
+    wstring outputFile    = config(L"outputFile");    // filename for the rendered topology plot
     // this can be empty, in that case no rendering will be done
     // or if this is set, renderCmd must be set, so CNTK will call re
-    wstring RenderCmd = config(L"RenderCmd"); // if this option is set, then CNTK will call the render to convert the outdotFile to a graph
+    wstring renderCmd = config(L"renderCmd"); // if this option is set, then CNTK will call the render to convert the outdotFile to a graph
     // e.g. "d:\Tools\graphviz\bin\dot.exe -Tpng -x <IN> -o<OUT>"
     //              where <IN> and <OUT> are two special placeholders
 
-    // ========================================
-    // Sec. 1 option check
-    // ========================================
-    if (outdot.empty())
-    {
-        outdot = modelPath + L".dot";
-    }
+    // output dot file defaults to modelpath.dot
+    if (outputDotFile.empty())
+        outputDotFile = modelPath + L".dot";
 
-    wstring rescmd;
-    if (!outRending.empty()) // we need to render the plot
-    {
-        std::wregex inputPlaceHolder(L"(.+)(<IN>)(.*)");
-        std::wregex outputPlaceHolder(L"(.+)(<OUT>)(.*)");
+    ComputationNetwork net(CPUDEVICE);
+    net.Load<ElemType>(modelPath);
+
+    net.PlotNetworkTopology(outputDotFile);
+    fprintf(stderr, "Created network description in dot format: %ls\n", outputDotFile.c_str());
 
-        rescmd = regex_replace(RenderCmd, inputPlaceHolder, L"$1" + outdot + L"$3");
-        rescmd = regex_replace(rescmd, outputPlaceHolder, L"$1" + outRending + L"$3");
+    if (!outputFile.empty())
+    {
+        if (renderCmd.empty())
+            InvalidArgument("plot: If you specify an outputFile, you also need a renderCmd.");
+#if 0 // this part is problematic under early version of gcc  (< 4.9)
+        static const wregex  inputPlaceHolder(L"(.+)(<IN>)(.*)");
+        static const wregex outputPlaceHolder(L"(.+)(<OUT>)(.*)");
+
+        // patch in the pathnames
+        renderCmd = regex_replace(renderCmd, inputPlaceHolder,  L"$1" + outputDotFile + L"$3");
+        renderCmd = regex_replace(renderCmd, outputPlaceHolder, L"$1" + outputFile    + L"$3");
+#endif
+        msra::strfun::ReplaceAll(renderCmd, wstring(L"<IN>"), outputDotFile);
+        msra::strfun::ReplaceAll(renderCmd, wstring(L"<OUT>"), outputFile);
     }
 
-    ComputationNetwork net(-1);
-    net.Load<ElemType>(modelPath);
-    net.PlotNetworkTopology(outdot);
-    fprintf(stderr, "Output network description in dot language to %S\n", outdot.c_str());
 
-    if (!outRending.empty())
-    {
-        fprintf(stderr, "Executing a third-part tool for rendering dot:\n%S\n", rescmd.c_str());
+        fprintf(stderr, "Executing third-party tool for rendering dot:\n%ls\n", renderCmd.c_str());
 #ifdef __unix__
-        const auto rc = system(msra::strfun::utf8(rescmd).c_str());
-        rc /*ignoring the result--this gets flagged by gcc if we don't save the return value*/;
+        auto rc = system(msra::strfun::utf8(renderCmd).c_str());
+        rc; // ignoring the result--this gets flagged by gcc if we don't save the return value
 #else
-        _wsystem(rescmd.c_str());
+        _wsystem(renderCmd.c_str());
 #endif
-        fprintf(stderr, "Done\n");
-    }
+        fprintf(stderr, "Done.\n");
 }
 
 template void DoTopologyPlot<float>(const ConfigParameters& config);

Source/CNTK/BrainScript/ExperimentalNetworkBuilder.cpp

@@ -58,6 +58,7 @@ L"ParameterTensor(dims, learningRateMultiplier = 1.0, init = 'uniform'/*|fixedVa
     L"TransposeDimensions(input, dim1, dim2, tag='') = new ComputationNode [ operation = 'TransposeDimensions' ; inputs = input /*plus the function args*/ ]\n"
     L"Transpose(x) = TransposeDimensions(x, 1, 2)\n"
     L"Times(A, B, outputRank=1, tag='') = new ComputationNode [ operation = 'Times' ; inputs = ( A : B ) /*plus the function args*/ ]\n"
+    // TODO: Logistic should be generated with with BinaryStandardNode macro below.
     L"Logistic(label, probability, tag='') = new ComputationNode [ operation = 'Logistic' ; inputs = (label : probability) /*plus the function args*/ ]\n"
     L"WeightedLogistic(label, probability, instanceWeight, tag='') = new ComputationNode [ operation = 'Logistic' ; inputs = (label : probability : instanceWeight) /*plus the function args*/ ]\n"
     L"ReconcileMBLayout(dataInput, layoutInput, tag='') = new ComputationNode [ operation = 'ReconcileMBLayout' ; inputs = (dataInput : layoutInput) /*plus the function args*/ ]\n"

Source/CNTK/CNTK.cpp

@@ -405,7 +405,7 @@ int wmainWithBS(int argc, wchar_t* argv[]) // called from wmain which is a wrapp
     wstring startupMessage = msra::strfun::wstrprintf(L"running on %ls at %ls\n", msra::strfun::utf16(GetHostName()).c_str(), msra::strfun::utf16(TimeDateStamp()).c_str());
     startupMessage += msra::strfun::wstrprintf(L"command line: %ls", exePath.c_str());
     for (const auto& arg : args)
-        startupMessage += L" " + arg;
+        startupMessage += L"  " + arg;
 
     fprintf(stderr, "%ls\n", startupMessage.c_str());
 
@@ -580,9 +580,7 @@ int wmainOldCNTKConfig(int argc, wchar_t* argv[]) // called from wmain which is
     fprintf(stderr, "running on %s at %s\n", GetHostName().c_str(), timestamp.c_str());
     fprintf(stderr, "command line: \n");
     for (int i = 0; i < argc; i++)
-    {
-        fprintf(stderr, "%s ", WCharToString(argv[i]).c_str());
-    }
+        fprintf(stderr, "%*s%ls", i > 0 ? 2 : 0, "", argv[i]); // use 2 spaces for better visual separability
 
     // This simply merges all the different config parameters specified (eg, via config files or via command line directly),
     // and prints it.

Source/CNTK/ModelEditLanguage.cpp

@@ -596,7 +596,7 @@ void MELScript<ElemType>::CallFunction(const std::string& p_name, const ConfigPa
             case melPropBatchNormMode:
             {
                 bool evalMode = params[2];
-                netNdl->cn->SetBatchNormlizationNodesBelowEvalMode(evalMode, node);
+                netNdl->cn->SetBatchNormalizationNodesBelowEvalMode(evalMode, node);
                 break;
             }
             default:

Source/CNTK/prebuild.bat

@@ -11,18 +11,21 @@ echo #define _BUILDINFO_H >> buildinfo.h$$
 FOR /F %%i IN ('hostname') DO SET HOST=%%i
 :: assuming hostname always exists
 
-:: not sure whether git in path ?
-call git --version 2 > nul
-if not %ERRORLEVEL% == 9009 (
-    echo #define _GIT_EXIST >> buildinfo.h$$
-    FOR /F %%i IN ('git rev-parse --abbrev-ref HEAD') DO SET BRANCH=%%i
-    FOR /F %%i IN ('git rev-parse HEAD') DO SET COMMIT=%%i
-    set STATUS=
-    git diff --quiet --cached
-    if not errorlevel 1 git diff --quiet
-    if errorlevel 1 set STATUS= ^(modified^)
-    echo #define _BUILDBRANCH_  "!BRANCH!"      >> buildinfo.h$$
-    echo #define _BUILDSHA1_    "!COMMIT!!STATUS!">> buildinfo.h$$
+:: note: we'll only use git which is in path
+where -q git
+if not errorlevel 1 (
+    call git --version > NUL 2>&1
+    if not errorlevel 1 (
+        echo #define _GIT_EXIST >> buildinfo.h$$
+        FOR /F %%i IN ('call git rev-parse --abbrev-ref HEAD') DO SET BRANCH=%%i
+        FOR /F %%i IN ('call git rev-parse HEAD') DO SET COMMIT=%%i
+        set STATUS=
+        call git diff --quiet --cached
+        if not errorlevel 1 call git diff --quiet
+        if errorlevel 1 set STATUS= ^(modified^)
+        echo #define _BUILDBRANCH_  "!BRANCH!"      >> buildinfo.h$$
+        echo #define _BUILDSHA1_    "!COMMIT!!STATUS!">> buildinfo.h$$
+    )
 )
 
 :: For now, math lib is basically hardwired
@@ -75,4 +78,4 @@ echo #endif >> buildinfo.h$$
 
 ::: update file only if it changed (otherwise CNTK.cpp will get rebuilt each time)
 fc buildinfo.h$$ buildinfo.h > NUL 2>&1
-if ERRORLEVEL 1 move /Y buildinfo.h$$ buildinfo.h
+if errorlevel 1 move /Y buildinfo.h$$ buildinfo.h