Merge branch 'master' of https://git.codeplex.com/cntk
This commit is contained in:
Yongqiang Wang
Коммит
d8b03dfdf6
|
|
@ -1102,7 +1102,7 @@ namespace Microsoft { namespace MSR { namespace CNTK {
|
|||
pLeft->FunctionValues() = redU;
|
||||
pRight->FunctionValues() = redVT;
|
||||
|
||||
shared_ptr<ComputationNode<ElemType>> pTimes = AddNodeToNetAndAttachInputs(New<TimesNode<ElemType>>(m_deviceId, name + L"-SVD", true /*createOutputMatrix*/), pLeft, pRight);
|
||||
shared_ptr<ComputationNode<ElemType>> pTimes = AddNodeToNetAndAttachInputs(New<TimesNode<ElemType>>(m_deviceId, name + L"-SVD"), pLeft, pRight);
|
||||
|
||||
//========================================
|
||||
// Step 3. remove old node
|
||||
|
|
|
|||
|
|
@ -392,18 +392,10 @@ namespace Microsoft { namespace MSR { namespace CNTK {
|
|||
static const std::wstring TypeName() { return L"Times"; }
|
||||
public:
|
||||
|
||||
// TODO: The createOutputMatrix parameter here is temporarily added to allow creating the function values
|
||||
// matrix for the times node added during SVD decomposition. Since ValidateSubNetwork is called after addition
|
||||
// of the times node, the validation crashes if the function values matrix has not yet been allocated
|
||||
// This can be removed after the Validation has been fixed to not access the function values matrix at all
|
||||
DeclareConstructorFromConfigWithNumInputs(TimesNode);
|
||||
TimesNode(DEVICEID_TYPE deviceId, const wstring & name, bool createOutputMatrix = false) :
|
||||
TimesNode(DEVICEID_TYPE deviceId, const wstring & name) :
|
||||
Base(deviceId, name)
|
||||
{
|
||||
if (createOutputMatrix)
|
||||
{
|
||||
CreateMatrixIfNull(m_functionValues);
|
||||
}
|
||||
}
|
||||
|
||||
virtual void /*ComputationNode::*/ComputeInputPartial(const size_t inputIndex, const FrameRange & frameRange) override
|
||||
|
|
|
|||
|
|
@ -893,8 +893,8 @@ already there from last epoch
|
|||
Starting minibatch loop.
|
||||
randomordering: 21 retries for 100 elements (21.0%) to ensure window condition
|
||||
randomordering: recached sequence for seed 11: 6, 31, ...
|
||||
Epoch[12 of 12]-Minibatch[ 1- 10 of 10]: SamplesSeen = 100; TrainLossPerSample = 0.37213734; EvalErr[0]PerSample = 0.00000000; TotalTime = 0.65604s; TotalTimePerSample = 6.56038ms; SamplesPerSecond = 152
|
||||
Finished Epoch[12 of 12]: [Training Set] TrainLossPerSample = 0.37213734; EvalErrPerSample = 0; Ave LearnRatePerSample = 0.004999999888; EpochTime=0.656382
|
||||
Epoch[12 of 12]-Minibatch[ 1- 10 of 10]: SamplesSeen = 100; TrainLossPerSample = 0.37077690; EvalErr[0]PerSample = 0.00000000; TotalTime = 0.65604s; TotalTimePerSample = 6.56038ms; SamplesPerSecond = 152
|
||||
Finished Epoch[12 of 12]: [Training Set] TrainLossPerSample = 0.37077689; EvalErrPerSample = 0; Ave LearnRatePerSample = 0.004999999888; EpochTime=0.656382
|
||||
CNTKCommandTrainEnd: Train
|
||||
|
||||
|
||||
|
|
@ -2269,8 +2269,8 @@ reading from record 0 to 100 to be positioned properly for epoch
|
|||
Starting minibatch loop.
|
||||
randomordering: 21 retries for 100 elements (21.0%) to ensure window condition
|
||||
randomordering: recached sequence for seed 11: 6, 31, ...
|
||||
Epoch[12 of 12]-Minibatch[ 1- 10 of 10]: SamplesSeen = 100; TrainLossPerSample = 0.37792297; EvalErr[0]PerSample = 0.00000000; TotalTime = 1.34518s; TotalTimePerSample = 13.45185ms; SamplesPerSecond = 74
|
||||
Finished Epoch[12 of 12]: [Training Set] TrainLossPerSample = 0.37792295; EvalErrPerSample = 0; Ave LearnRatePerSample = 0.004999999888; EpochTime=1.371377
|
||||
Epoch[12 of 12]-Minibatch[ 1- 10 of 10]: SamplesSeen = 100; TrainLossPerSample = 0.37650299; EvalErr[0]PerSample = 0.00000000; TotalTime = 1.34518s; TotalTimePerSample = 13.45185ms; SamplesPerSecond = 74
|
||||
Finished Epoch[12 of 12]: [Training Set] TrainLossPerSample = 0.37650299; EvalErrPerSample = 0; Ave LearnRatePerSample = 0.004999999888; EpochTime=1.371377
|
||||
CNTKCommandTrainEnd: Train
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -864,8 +864,8 @@ already there from last epoch
|
|||
Starting minibatch loop.
|
||||
randomordering: 21 retries for 100 elements (21.0%) to ensure window condition
|
||||
randomordering: recached sequence for seed 11: 6, 31, ...
|
||||
Epoch[12 of 12]-Minibatch[ 1- 10 of 10]: SamplesSeen = 100; TrainLossPerSample = 0.37213734; EvalErr[0]PerSample = 0.00000000; TotalTime = 0.08724s; TotalTimePerSample = 0.87241ms; SamplesPerSecond = 1146
|
||||
Finished Epoch[12 of 12]: [Training Set] TrainLossPerSample = 0.37213734; EvalErrPerSample = 0; Ave LearnRatePerSample = 0.004999999888; EpochTime=0.087336
|
||||
Epoch[12 of 12]-Minibatch[ 1- 10 of 10]: SamplesSeen = 100; TrainLossPerSample = 0.37077690; EvalErr[0]PerSample = 0.00000000; TotalTime = 0.08724s; TotalTimePerSample = 0.87241ms; SamplesPerSecond = 1146
|
||||
Finished Epoch[12 of 12]: [Training Set] TrainLossPerSample = 0.37077689; EvalErrPerSample = 0; Ave LearnRatePerSample = 0.004999999888; EpochTime=0.087336
|
||||
CNTKCommandTrainEnd: Train
|
||||
|
||||
|
||||
|
|
@ -2182,8 +2182,8 @@ reading from record 0 to 100 to be positioned properly for epoch
|
|||
Starting minibatch loop.
|
||||
randomordering: 21 retries for 100 elements (21.0%) to ensure window condition
|
||||
randomordering: recached sequence for seed 11: 6, 31, ...
|
||||
Epoch[12 of 12]-Minibatch[ 1- 10 of 10]: SamplesSeen = 100; TrainLossPerSample = 0.37792297; EvalErr[0]PerSample = 0.00000000; TotalTime = 0.89367s; TotalTimePerSample = 8.93670ms; SamplesPerSecond = 111
|
||||
Finished Epoch[12 of 12]: [Training Set] TrainLossPerSample = 0.37792295; EvalErrPerSample = 0; Ave LearnRatePerSample = 0.004999999888; EpochTime=0.908817
|
||||
Epoch[12 of 12]-Minibatch[ 1- 10 of 10]: SamplesSeen = 100; TrainLossPerSample = 0.37650299; EvalErr[0]PerSample = 0.00000000; TotalTime = 0.89367s; TotalTimePerSample = 8.93670ms; SamplesPerSecond = 111
|
||||
Finished Epoch[12 of 12]: [Training Set] TrainLossPerSample = 0.37650299; EvalErrPerSample = 0; Ave LearnRatePerSample = 0.004999999888; EpochTime=0.908817
|
||||
CNTKCommandTrainEnd: Train
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -5,6 +5,7 @@ deviceId=$DeviceId$
|
|||
ndlMacros=$ConfigDir$/Macros.ndl
|
||||
|
||||
parallelTrain=false
|
||||
NumCPUThreads=8
|
||||
|
||||
Train=[
|
||||
action=train
|
||||
|
|
|
|||
|
|
@ -1,9 +1,9 @@
|
|||
dataDir: ../Data
|
||||
tags:
|
||||
# running on every BVT job in 'I' (Image) leg:
|
||||
- bvt-i os=='windows' or device=='gpu'
|
||||
- bvt-i device=='gpu'
|
||||
# running every Nightly job in 'I' leg
|
||||
- nightly-i os=='windows' or device=='gpu'
|
||||
- nightly-i device=='gpu'
|
||||
|
||||
testCases:
|
||||
CNTK Run must be completed:
|
||||
|
|
|
|||
Загрузка…
Ссылка в новой задаче