Skip to content

TensorFlow: Fixed shape issue where unknown shape will be induced from data. #2475

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
merged 5 commits into from
Feb 10, 2019
+135 −29
Merged

TensorFlow: Fixed shape issue where unknown shape will be induced from data. #2475

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
344d4a6
Fixed shape issue where unknown shape will be induced from data.
zeahmed Feb 8, 2019
9602427
Addressed reviewers' comments.
zeahmed Feb 9, 2019
7855afe
Resolved merged conflicts.
zeahmed Feb 9, 2019
9fb2b73
Fixed TestData nuget version in Dependencies.props
zeahmed Feb 9, 2019
c9389d1
Updated TensorFlow.TestData nuget version.
zeahmed Feb 10, 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
2 changes: 1 addition & 1 deletion build/Dependencies.props
View file
Original file line number Diff line number Diff line change
@@ -43,7 +43,7 @@
<PropertyGroup>
<BenchmarkDotNetVersion>0.11.3</BenchmarkDotNetVersion>
<MicrosoftMLTestModelsPackageVersion>0.0.3-test</MicrosoftMLTestModelsPackageVersion>
<MicrosoftMLTensorFlowTestModelsVersion>0.0.7-test</MicrosoftMLTensorFlowTestModelsVersion>
<MicrosoftMLTensorFlowTestModelsVersion>0.0.10-test</MicrosoftMLTensorFlowTestModelsVersion>
<MicrosoftMLOnnxTestModelsVersion>0.0.4-test</MicrosoftMLOnnxTestModelsVersion>
</PropertyGroup>

45 changes: 20 additions & 25 deletions src/Microsoft.ML.TensorFlow/TensorflowTransform.cs
View file
Original file line number Diff line number Diff line change
@@ -667,15 +667,6 @@ internal static (TFDataType[] tfInputTypes, TFShape[] tfInputShapes) GetInputInf
var tfInput = new TFOutput(session.Graph[inputs[i]]);
tfInputTypes[i] = tfInput.OutputType;
tfInputShapes[i] = session.Graph.GetTensorShape(tfInput);
if (tfInputShapes[i].NumDimensions != -1)
{
var newShape = new long[tfInputShapes[i].NumDimensions];
newShape[0] = tfInputShapes[i][0] == -1 ? BatchSize : tfInputShapes[i][0];

for (int j = 1; j < tfInputShapes[i].NumDimensions; j++)
newShape[j] = tfInputShapes[i][j];
tfInputShapes[i] = new TFShape(newShape);
}
}
return (tfInputTypes, tfInputShapes);
}
@@ -698,7 +689,14 @@ internal static (TFDataType[] tfOutputTypes, ColumnType[] outputTypes) GetOutput
{
var tfOutput = new TFOutput(session.Graph[outputs[i]]);
var shape = session.Graph.GetTensorShape(tfOutput);

// The transformer can only retreive the output as fixed length vector with shape of kind [-1, d1, d2, d3, ...]
// i.e. the first dimension (if unknown) is assumed to be batch dimension.
// If there are other dimension that are unknown the transformer will return a variable length vector.
// This is the work around in absence of reshape transformer.
int[] dims = shape.NumDimensions > 0 ? shape.ToIntArray().Skip(shape[0] == -1 ? 1 : 0).ToArray() : new[] { 0 };
for (int j = 0; j < dims.Length; j++)
dims[j] = dims[j] == -1 ? 0 : dims[j];
var type = TensorFlowUtils.Tf2MlNetType(tfOutput.OutputType);
outputTypes[i] = new VectorType(type, dims);
tfOutputTypes[i] = tfOutput.OutputType;
@@ -837,14 +835,22 @@ public Mapper(TensorFlowTransformer parent, Schema inputSchema) :
var originalShape = _parent.TFInputShapes[i];
var shape = originalShape.ToIntArray();

var colTypeDims = vecType.Dimensions.Prepend(1).Select(dim => (long)dim).ToArray();
var colTypeDims = vecType.Dimensions.Select(dim => (long)dim).ToArray();
if (shape == null)
_fullySpecifiedShapes[i] = new TFShape(colTypeDims);
else if (vecType.Dimensions.Length == 1)
else
{
// If the column is one dimension we make sure that the total size of the TF shape matches.
// Compute the total size of the known dimensions of the shape.
int valCount = shape.Where(x => x > 0).Aggregate((x, y) => x * y);
int valCount = 1;
int numOfUnkDim = 0;
foreach (var s in shape)
{
if (s > 0)
valCount *= s;
else
numOfUnkDim++;
}
// The column length should be divisible by this, so that the other dimensions can be integral.
int typeValueCount = type.GetValueCount();
if (typeValueCount % valCount != 0)
@@ -853,8 +859,8 @@ public Mapper(TensorFlowTransformer parent, Schema inputSchema) :
// If the shape is multi-dimensional, we should be able to create the length of the vector by plugging
// in a single value for the unknown shapes. For example, if the shape is [?,?,3], then there should exist a value
// d such that d*d*3 is equal to the length of the input column.
var d = originalShape.NumDimensions > 2 ? Math.Pow(typeValueCount / valCount, 1.0 / (originalShape.NumDimensions - 2)) : 1;
if (originalShape.NumDimensions > 2 && d - (int)d != 0)
var d = numOfUnkDim > 0 ? Math.Pow(typeValueCount / valCount, 1.0 / numOfUnkDim) : 0;
if (d - (int)d != 0)
throw Contracts.Except($"Input shape mismatch: Input '{_parent.Inputs[i]}' has shape {originalShape.ToString()}, but input data is of length {typeValueCount}.");

// Fill in the unknown dimensions.
@@ -863,17 +869,6 @@ public Mapper(TensorFlowTransformer parent, Schema inputSchema) :
l[ishape] = originalShape[ishape] == -1 ? (int)d : originalShape[ishape];
_fullySpecifiedShapes[i] = new TFShape(l);
}
else
{
if (shape.Select((dim, j) => dim != -1 && dim != colTypeDims[j]).Any(b => b))
throw Contracts.Except($"Input shape mismatch: Input '{_parent.Inputs[i]}' has shape {originalShape.ToString()}, but input data is {vecType.ToString()}.");

// Fill in the unknown dimensions.
var l = new long[originalShape.NumDimensions];
for (int ishape = 0; ishape < originalShape.NumDimensions; ishape++)
l[ishape] = originalShape[ishape] == -1 ? colTypeDims[ishape] : originalShape[ishape];
_fullySpecifiedShapes[i] = new TFShape(l);
}
}
}

117 changes: 114 additions & 3 deletions test/Microsoft.ML.Tests/ScenariosWithDirectInstantiation/TensorflowTests.cs
View file
Original file line number Diff line number Diff line change
@@ -74,6 +74,117 @@ public void TensorFlowTransformMatrixMultiplicationTest()
}
}

private class ShapeData
{
// Data will be passed as 1-D vector.
// Intended data shape [5], model shape [None]
[VectorType(5)]
public float[] OneDim;

// Data will be passed as flat vector.
// Intended data shape [2,2], model shape [2, None]
[VectorType(4)]
public float[] TwoDim;

// Data will be passed as 3-D vector.
// Intended data shape [1, 2, 2], model shape [1, None, 2]
[VectorType(1, 2, 2)]
public float[] ThreeDim;

// Data will be passed as flat vector.
// Intended data shape [1, 2, 2, 3], model shape [1, None, None, 3]
[VectorType(12)]
public float[] FourDim;

// Data will be passed as 4-D vector.
// Intended data shape [2, 2, 2, 2], model shape [2, 2, 2, 2]
[VectorType(2, 2, 2, 2)]
public float[] FourDimKnown;
}

private List<ShapeData> GetShapeData()
{
return new List<ShapeData>(new ShapeData[] {
new ShapeData() { OneDim = new[] { 0.1f, 0.2f, 0.3f, 0.4f, 0.5f },
TwoDim = new[] { 1.0f, 2.0f, 3.0f, 4.0f },
ThreeDim = new[] { 11.0f, 12.0f, 13.0f, 14.0f },
FourDim = new[]{ 21.0f, 22.0f, 23.0f, 24.0f, 25.0f, 26.0f,
27.0f, 28.0f, 29.0f, 30.0f, 31.0f, 32.0f },
FourDimKnown = new[]{ 41.0f , 42.0f, 43.0f, 44.0f, 45.0f, 46.0f, 47.0f, 48.0f,
49.0f , 50.0f, 51.0f, 52.0f, 53.0f, 54.0f, 55.0f, 56.0f}
},
new ShapeData() { OneDim = new[] { 100.1f, 100.2f, 100.3f, 100.4f, 100.5f },
TwoDim = new[] { 101.0f, 102.0f, 103.0f, 104.0f },
ThreeDim = new[] { 111.0f, 112.0f, 113.0f, 114.0f },
FourDim = new[]{ 121.0f, 122.0f, 123.0f, 124.0f, 125.0f, 126.0f,
127.0f, 128.0f, 129.0f, 130.0f, 131.0f, 132.0f},
FourDimKnown = new[]{ 141.0f , 142.0f, 143.0f, 144.0f, 145.0f, 146.0f, 147.0f, 148.0f,
149.0f , 150.0f, 151.0f, 152.0f, 153.0f, 154.0f, 155.0f, 156.0f }
}
});
}

[ConditionalFact(typeof(Environment), nameof(Environment.Is64BitProcess))] // TensorFlow is 64-bit only
public void TensorFlowTransformInputShapeTest()
{
var modelLocation = "model_shape_test";
var mlContext = new MLContext(seed: 1, conc: 1);
var data = GetShapeData();
// Pipeline
var loader = mlContext.Data.ReadFromEnumerable(data);
var inputs = new string[] { "OneDim", "TwoDim", "ThreeDim", "FourDim", "FourDimKnown" };
var outputs = new string[] { "o_OneDim", "o_TwoDim", "o_ThreeDim", "o_FourDim", "o_FourDimKnown" };

var trans = mlContext.Transforms.ScoreTensorFlowModel(modelLocation, outputs, inputs).Fit(loader).Transform(loader);

using (var cursor = trans.GetRowCursorForAllColumns())
{
int outColIndex = 5;
var oneDimgetter = cursor.GetGetter<VBuffer<float>>(outColIndex);
var twoDimgetter = cursor.GetGetter<VBuffer<float>>(outColIndex + 1);
var threeDimgetter = cursor.GetGetter<VBuffer<float>>(outColIndex + 2);
var fourDimgetter = cursor.GetGetter<VBuffer<float>>(outColIndex + 3);
var fourDimKnowngetter = cursor.GetGetter<VBuffer<float>>(outColIndex + 4);

VBuffer<float> oneDim = default;
VBuffer<float> twoDim = default;
VBuffer<float> threeDim = default;
VBuffer<float> fourDim = default;
VBuffer<float> fourDimKnown = default;
foreach (var sample in data)
{
Assert.True(cursor.MoveNext());

oneDimgetter(ref oneDim);
twoDimgetter(ref twoDim);
threeDimgetter(ref threeDim);
fourDimgetter(ref fourDim);
fourDimKnowngetter(ref fourDimKnown);

var oneDimValues = oneDim.GetValues();
Assert.Equal(sample.OneDim.Length, oneDimValues.Length);
Assert.True(oneDimValues.SequenceEqual(sample.OneDim));

var twoDimValues = twoDim.GetValues();
Assert.Equal(sample.TwoDim.Length, twoDimValues.Length);
Assert.True(twoDimValues.SequenceEqual(sample.TwoDim));

var threeDimValues = threeDim.GetValues();
Assert.Equal(sample.ThreeDim.Length, threeDimValues.Length);
Assert.True(threeDimValues.SequenceEqual(sample.ThreeDim));

var fourDimValues = fourDim.GetValues();
Assert.Equal(sample.FourDim.Length, fourDimValues.Length);
Assert.True(fourDimValues.SequenceEqual(sample.FourDim));

var fourDimKnownValues = fourDimKnown.GetValues();
Assert.Equal(sample.FourDimKnown.Length, fourDimKnownValues.Length);
Assert.True(fourDimKnownValues.SequenceEqual(sample.FourDimKnown));
}
Assert.False(cursor.MoveNext());
}
}

private class TypesData
{
[VectorType(2)]
@@ -142,7 +253,7 @@ public void TensorFlowTransformInputOutputTypesTest()

var loader = mlContext.Data.ReadFromEnumerable(data);

var inputs = new string[]{"f64", "f32", "i64", "i32", "i16", "i8", "u64", "u32", "u16", "u8","b"};
var inputs = new string[] { "f64", "f32", "i64", "i32", "i16", "i8", "u64", "u32", "u16", "u8", "b" };
var outputs = new string[] { "o_f64", "o_f32", "o_i64", "o_i32", "o_i16", "o_i8", "o_u64", "o_u32", "o_u16", "o_u8", "o_b" };
var trans = mlContext.Transforms.ScoreTensorFlowModel(model_location, outputs, inputs).Fit(loader).Transform(loader); ;

@@ -160,7 +271,7 @@ public void TensorFlowTransformInputOutputTypesTest()
var u8getter = cursor.GetGetter<VBuffer<byte>>(20);
var boolgetter = cursor.GetGetter<VBuffer<bool>>(21);


VBuffer<double> f64 = default;
VBuffer<float> f32 = default;
VBuffer<long> i64 = default;
@@ -449,7 +560,7 @@ public void TensorFlowTransformMNISTLRTrainingTest()
ReTrain = true
}))
.Append(mlContext.Transforms.Concatenate("Features", "Prediction"))
.Append(mlContext.Transforms.Conversion.MapValueToKey("KeyLabel","Label", maxNumKeys: 10))
.Append(mlContext.Transforms.Conversion.MapValueToKey("KeyLabel", "Label", maxNumKeys: 10))
.Append(mlContext.MulticlassClassification.Trainers.LightGbm("KeyLabel", "Features"));

var trainedModel = pipe.Fit(trainData);