Skip to content

Image Classification API: Fix processing incomplete batch(<batchSize), images processed per epoch , enable EarlyStopping without Validation Set. Fixes #4274 and #4286 #4289

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ashbhandare merged 6 commits into from
Oct 4, 2019
Merged

Image Classification API: Fix processing incomplete batch(<batchSize), images processed per epoch , enable EarlyStopping without Validation Set. Fixes #4274 and #4286 #4289

Changes from 4 commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
b451c25
In ImageClassification, process incomplete batch where number of samp…
ashbhandare Oct 2, 2019
6f2915b
fixed batchIndex not reseting in train loop, enabled EarlyStopping wh…
ashbhandare Oct 3, 2019
995b227
Merge branch 'master' into Issue4274
ashbhandare Oct 3, 2019
123f31f
fixed changing shape of feature and label tensor for incomplete batch…
ashbhandare Oct 3, 2019
3b614eb
Merge branch 'master' into Issue4274
ashbhandare Oct 4, 2019
034ce8e
Improved featureBatchSizeInBytes calculation, improved exception mess…
ashbhandare Oct 4, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
72 changes: 72 additions & 0 deletions src/Microsoft.ML.Dnn/ImageClassificationTransform.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -186,6 +186,8 @@ private void CheckTrainingParameters(ImageClassificationEstimator.Options option

if (_session.graph.OperationByName(_labelTensor.name.Split(':')[0]) == null)
throw Host.ExceptParam(nameof(options.TensorFlowLabel), $"'{options.TensorFlowLabel}' does not exist in the model");
if (options.EarlyStoppingCriteria != null && options.ValidationSet == null && options.TestOnTrainSet == false)
throw Host.ExceptParam(nameof(options.EarlyStoppingCriteria), $"No Validation dataset provided and testing on Train disabled, Early Stopping not supported.");

@codemzs codemzs Oct 4, 2019

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

"No Validation dataset provided and testing on Train disabled, Early Stopping not supported. [](start = 79, length = 92)

Lets improve the exception message "Early stopping enabled but unable to find a validation set and/or train set testing disabled. Please disable early stopping or either provide a validation set or enable train set training."

}

private (Tensor, Tensor) AddJpegDecoding(int height, int width, int depth)
Expand Down Expand Up @@ -378,6 +380,7 @@ private void TrainAndEvaluateClassificationLayer(string trainBottleneckFilePath,
float crossentropy = 0;
for (int epoch = 0; epoch < epochs; epoch += 1)
{
batchIndex = 0;
metrics.Train.Accuracy = 0;
metrics.Train.CrossEntropy = 0;
metrics.Train.BatchProcessedCount = 0;
Expand Down Expand Up @@ -429,6 +432,42 @@ private void TrainAndEvaluateClassificationLayer(string trainBottleneckFilePath,
}
}

//Process last incomplete batch
if (batchIndex > 0)
{
featureTensorShape[0] = batchIndex;
featureBatchSizeInBytes = sizeof(float) * featureBatch.Length * batchIndex / batchSize;

@codemzs codemzs Oct 4, 2019 •
edited
Loading

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sizeof(float) * featureBatch.Length * batchIndex / batchSize; [](start = 49, length = 62)

sizeof(float) * featureLength * batchIndex

labelTensorShape[0] = batchIndex;
labelBatchSizeInBytes = sizeof(long) * batchIndex;
runner.AddInput(new Tensor(featureBatchPtr, featureTensorShape, TF_DataType.TF_FLOAT, featureBatchSizeInBytes), 0)
.AddInput(new Tensor(labelBatchPtr, labelTensorShape, TF_DataType.TF_INT64, labelBatchSizeInBytes), 1)
.Run();

Comment thread
ashbhandare marked this conversation as resolved.
metrics.Train.BatchProcessedCount += 1;

if (options.TestOnTrainSet && statisticsCallback != null)
{
var outputTensors = testEvalRunner
.AddInput(new Tensor(featureBatchPtr, featureTensorShape, TF_DataType.TF_FLOAT, featureBatchSizeInBytes), 0)
.AddInput(new Tensor(labelBatchPtr, labelTensorShape, TF_DataType.TF_INT64, labelBatchSizeInBytes), 1)
.Run();

outputTensors[0].ToScalar<float>(ref accuracy);
outputTensors[1].ToScalar<float>(ref crossentropy);
metrics.Train.Accuracy += accuracy;
metrics.Train.CrossEntropy += crossentropy;

outputTensors[0].Dispose();
outputTensors[1].Dispose();
}

batchIndex = 0;
featureTensorShape[0] = batchSize;
featureBatchSizeInBytes = sizeof(float) * featureBatch.Length;
labelTensorShape[0] = batchSize;
labelBatchSizeInBytes = sizeof(long) * batchSize;
}

if (options.TestOnTrainSet && statisticsCallback != null)
{
metrics.Train.Epoch = epoch;
Expand All @@ -440,7 +479,15 @@ private void TrainAndEvaluateClassificationLayer(string trainBottleneckFilePath,
}

if (validationSet == null)
{
//Early stopping check
if (options.EarlyStoppingCriteria != null)
{
if (options.EarlyStoppingCriteria.ShouldStop(metrics.Train))
break;
}
continue;
}

batchIndex = 0;
metrics.Train.BatchProcessedCount = 0;
Expand Down Expand Up @@ -478,6 +525,31 @@ private void TrainAndEvaluateClassificationLayer(string trainBottleneckFilePath,
}
}

//Process last incomplete batch
if(batchIndex > 0)
{
featureTensorShape[0] = batchIndex;
featureBatchSizeInBytes = sizeof(float) * featureBatch.Length * batchIndex / batchSize;

@codemzs codemzs Oct 4, 2019

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sizeof(float) * featureBatch.Length * batchIndex / batchSize; [](start = 50, length = 61)

sizeof(float) * featureLength * batchIndex

labelTensorShape[0] = batchIndex;
labelBatchSizeInBytes = sizeof(long) * batchIndex;
var outputTensors = validationEvalRunner
.AddInput(new Tensor(featureBatchPtr, featureTensorShape, TF_DataType.TF_FLOAT, featureBatchSizeInBytes), 0)
.AddInput(new Tensor(labelBatchPtr, labelTensorShape, TF_DataType.TF_INT64, labelBatchSizeInBytes), 1)
.Run();

outputTensors[0].ToScalar<float>(ref accuracy);
metrics.Train.Accuracy += accuracy;
metrics.Train.BatchProcessedCount += 1;
batchIndex = 0;

featureTensorShape[0] = batchSize;
featureBatchSizeInBytes = sizeof(float) * featureBatch.Length;
labelTensorShape[0] = batchSize;
labelBatchSizeInBytes = sizeof(long) * batchSize;

outputTensors[0].Dispose();
}
Comment thread
ashbhandare marked this conversation as resolved.

if (statisticsCallback != null)
{
metrics.Train.Epoch = epoch;
Expand Down