Added a E2E Speech DNN parallel training test with no gradient quanti…

…zation

Tests/Speech/DNN/ParallelNoQuantization/run-test

@@ -0,0 +1,40 @@
+#!/bin/bash
+if [ "$TEST_DEVICE" == "cpu" ]; then
+  CNTK_DEVICE_ID=-1
+elif [ "$TEST_DEVICE" == "gpu" ]; then
+  CNTK_DEVICE_ID=0
+else
+  echo "Error: Unknown TEST_DEVICE specified!"
+  exit 3
+fi
+
+configFile=$TEST_DIR/../cntk.config
+RunDir=$TEST_RUN_DIR
+DataDir=$TEST_DATA_DIR
+
+if [ "$OS" == "Windows_NT" ]; then
+  # When running on cygwin translating /cygdrive/xxx paths to proper windows paths:
+  configFile=$(cygpath -aw $configFile)
+  RunDir=$(cygpath -aw $RunDir)
+  DataDir=$(cygpath -aw $DataDir)
+fi
+
+# Since we use the MS MPI program on Windows, the CNTK binary path argument
+# passed to mpiexec must be in the windows format
+CNTKBinaryPath=$TEST_CNTK_BINARY
+if [ "$OS" == "Windows_NT" ]; then
+  CNTKBinaryPath=$(cygpath -aw $CNTKBinaryPath)
+fi
+
+MPI_ARGS="-n 3"
+CNTK_ARGS="configFile=$configFile RunDir=$RunDir DataDir=$DataDir DeviceId=$CNTK_DEVICE_ID stderr=$RunDir/stderr"
+MODELS_DIR=$TEST_RUN_DIR/models
+[ -d $MODELS_DIR ] && rm -rf $MODELS_DIR
+mkdir -p $MODELS_DIR || exit $?
+echo === Running "$MPI_BINARY" $MPI_ARGS $CNTKBinaryPath $CNTK_ARGS
+"$MPI_BINARY" $MPI_ARGS $CNTKBinaryPath $CNTK_ARGS
+ExitCode=$?
+sed 's/^/MPI Rank 0: /' $TEST_RUN_DIR/stderr_speechTrain.logrank0
+sed 's/^/MPI Rank 1: /' $TEST_RUN_DIR/stderr_speechTrain.logrank1
+sed 's/^/MPI Rank 2: /' $TEST_RUN_DIR/stderr_speechTrain.logrank2
+exit $ExitCode

Tests/Speech/DNN/ParallelNoQuantization/testcases.yml

@@ -0,0 +1,40 @@
+dataDir: ../../Data
+tags:
+     # running on every BVT job in 'S' (Speech) leg in Debug-GPU and Release-CPU configurations:
+     - bvt-s  (flavor=='debug') ^ (device=='cpu')
+     # running unconditionally on every Nightly job in 'S' leg
+     - nightly-s
+
+testCases:
+  Must train epochs in exactly same order and parameters for each MPI Rank:
+    patterns:
+      - ^MPI Rank {{integer}}
+      - Starting Epoch {{integer}}
+      - learning rate per sample = {{float}}
+      - momentum = {{float}}
+
+  Epochs must be finished with expected results for each MPI Rank:
+    patterns:
+      - ^MPI Rank {{integer}}
+      - Finished Epoch[{{integer}} of {{integer}}]
+      - TrainLossPerSample = {{float,tolerance=0.01%}}
+      - EvalErrPerSample = {{float,tolerance=0.01%}}
+      - Ave LearnRatePerSample = {{float,tolerance=0%}}
+
+  Per-minibatch training results must match for each MPI Rank:
+    patterns:
+      - ^MPI Rank {{integer}}
+      - Epoch[{{integer}} of {{integer}}]-Minibatch[{{integer}}-{{integer}} of {{integer}}]
+      - SamplesSeen = {{integer}}
+      - TrainLossPerSample = {{float,tolerance=0.01%}}
+      - EvalErr[0]PerSample = {{float,tolerance=0.01%}}
+
+  DataParallelSGD training parameters must match for each MPI Rank:
+    patterns:
+      - ^MPI Rank {{integer}}
+      - Starting minibatch loop
+      - DataParallelSGD training
+      - MyRank = {{integer}}
+      - NumNodes = 3
+      - NumGradientBits = 32
+      - Distributed reading is ENABLED

Tests/Speech/DNN/cntk.config

@@ -0,0 +1,99 @@
+precision=float
+command=speechTrain
+deviceId=$DeviceId$
+
+parallelTrain=true
+
+speechTrain=[
+    action=train
+    modelPath=$RunDir$/models/cntkSpeech.dnn
+    deviceId=$DeviceId$
+    traceLevel=1
+    SimpleNetworkBuilder=[
+        layerSizes=363:512:512:132
+        trainingCriterion=CrossEntropyWithSoftmax
+        evalCriterion=ErrorPrediction
+        layerTypes=Sigmoid
+        initValueScale=1.0
+        applyMeanVarNorm=true
+        uniformInit=true
+        needPrior=true
+    ]
+
+    ExperimentalNetworkBuilder=[    // the same as above but with BS
+        layerSizes=363:512:512:132
+        trainingCriterion='CE'
+        evalCriterion='Err'
+
+        applyMeanVarNorm=true
+
+        L = Length(layerSizes)-1    // number of model layers
+        features = Input(layerSizes[0], 1, tag='feature') ; labels = Input(layerSizes[Length(layerSizes)-1], 1, tag='label')
+        featNorm = if applyMeanVarNorm
+                   then MeanVarNorm(features)
+                   else features
+        layers[layer:1..L-1] = if layer > 1
+                               then SBFF(layers[layer-1].Eh, layerSizes[layer], layerSizes[layer-1])
+                               else SBFF(featNorm, layerSizes[layer], layerSizes[layer-1])
+        outLayer = BFF(layers[L-1].Eh, layerSizes[L], layerSizes[L-1])
+        outZ = outLayer.z        // + PastValue(layerSizes[L], 1, outLayer.z)
+        CE = if trainingCriterion == 'CE'
+             then CrossEntropyWithSoftmax(labels, outZ, tag='criterion')
+             else Fail('unknown trainingCriterion ' + trainingCriterion)
+        Err = if evalCriterion == 'Err' then
+              ErrorPrediction(labels, outZ, tag='eval')
+              else Fail('unknown evalCriterion ' + evalCriterion)
+        logPrior = LogPrior(labels)
+        // TODO: how to add a tag to an infix operation?
+        ScaledLogLikelihood = Minus (outZ, logPrior, tag='output')
+    ]
+
+    SGD=[
+        epochSize=20480
+        minibatchSize=64:256:1024
+        learningRatesPerMB=1.0:0.5:0.1
+        numMBsToShowResult=10
+        momentumPerMB=0.9:0.656119
+        dropoutRate=0.0
+        maxEpochs=3
+        keepCheckPointFiles=true
+
+        ParallelTrain=[
+            parallelizationMethod=DataParallelSGD
+            distributedMBReading=true
+            DataParallelSGD=[
+              gradientBits=32
+            ]
+        ]
+
+        AutoAdjust=[
+            reduceLearnRateIfImproveLessThan=0
+            loadBestModel=true
+            increaseLearnRateIfImproveMoreThan=1000000000
+            learnRateDecreaseFactor=0.5
+            learnRateIncreaseFactor=1.382
+            autoAdjustLR=AdjustAfterEpoch
+        ]
+        clippingThresholdPerSample=1#INF
+    ]
+    reader=[
+      readerType=HTKMLFReader
+      readMethod=blockRandomize
+      miniBatchMode=Partial
+      randomize=Auto
+      verbosity=0
+      features=[
+          dim=363
+          type=Real
+          scpFile=glob_0000.scp
+      ]
+
+      labels=[
+          mlfFile=$DataDir$/glob_0000.mlf
+          labelMappingFile=$DataDir$/state.list
+
+          labelDim=132
+          labelType=Category
+      ]
+    ]
+]