scisharp
/
TensorFlow.NET
Not watched
1
0
Fork 0
Code
Releases 21 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
Browse Source

Merge branch 'master' into v0.11.x-numsharp

tags/v0.12
Oceania2018 6 years ago
parent
4ff53c93be 9dd61ab700
commit
787c772b06
13 changed files with 1016 additions and 13 deletions
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. +6
    -0
      README.md
  2. +8
    -2
      src/SciSharp.TensorFlow.Redist/README.md
  3. +1
    -1
      src/SciSharp.TensorFlow.Redist/Redist-CPU.nuspec
  4. +26
    -0
      src/SciSharp.TensorFlow.Redist/Redist-Windows-GPU.nuspec
  5. +1
    -1
      src/SciSharp.TensorFlow.Redist/SciSharp.TensorFlow.Redist-CPU.nupkgproj
  6. +187
    -0
      src/SciSharp.TensorFlow.Redist/SciSharp.TensorFlow.Redist-Windows-GPU.nupkgproj
  7. +1
    -0
      src/SciSharp.TensorFlow.Redist/libtensorflow-gpu-windows-x86_64-1.14.0.zip.sha
  8. +49
    -0
      src/TensorFlowHub/MnistDataSet.cs
  9. +17
    -4
      src/TensorFlowHub/MnistModelLoader.cs
  10. +26
    -2
      src/TensorFlowNET.Core/Tensors/Tensor.Operators.cs
  11. +15
    -2
      tensorflowlib/README.md
  12. +5
    -1
      test/TensorFlowNET.Examples/ImageProcessing/DigitRecognitionCNN.cs
  13. +674
    -0
      test/TensorFlowNET.UnitTest/OperationsTest.cs

+ 6
- 0
README.md View File

@@ -28,8 +28,14 @@ In comparison to other projects, like for instance TensorFlowSharp which only pr


Install TF.NET and TensorFlow binary through NuGet. Install TF.NET and TensorFlow binary through NuGet.
```sh ```sh
### install tensorflow C# binding
PM> Install-Package TensorFlow.NET PM> Install-Package TensorFlow.NET

### Install tensorflow binary
### For CPU version
PM> Install-Package SciSharp.TensorFlow.Redist PM> Install-Package SciSharp.TensorFlow.Redist
### For GPU version (CUDA and cuDNN are required)
PM> Install-Package SciSharp.TensorFlow.Redist-Windows-GPU
``` ```


Import TF.NET. Import TF.NET.


+ 8
- 2
src/SciSharp.TensorFlow.Redist/README.md View File

@@ -1,8 +1,14 @@
## SciSharp.TensorFlow.Redist ## ## SciSharp.TensorFlow.Redist ##




`SciSharp.TensorFlow.Redist` is a migration from [Microsoft.ML.TensorFlow.Redist](https://github.com/dotnet/machinelearning/tree/release/1.2/src/Redist/Microsoft.ML.TensorFlow.Redist). [ML.NET](https://github.com/dotnet/machinelearning) team will not maintain the package since [ML.NET](https://www.nuget.org/packages/Microsoft.ML) v1.4.0 going forward.
`SciSharp.TensorFlow.Redist` is a migration from [Microsoft.ML.TensorFlow.Redist](https://github.com/dotnet/machinelearning/tree/release/1.2/src/Redist/Microsoft.ML.TensorFlow.Redist). [ML.NET](https://github.com/dotnet/machinelearning) team will not maintain the package since [ML.NET](https://www.nuget.org/packages/Microsoft.ML) v1.3.0 going forward.


* CPU version for all platforms (Windows, Linux, OSX)
```powershell
PM> Install-Package SciSharp.TensorFlow.Redist
```

* GPU version for Windows
```powershell ```powershell
PM> Install-Package SciSharp.TensorFlow.Redist PM> Install-Package SciSharp.TensorFlow.Redist
``` ```
@@ -16,7 +22,7 @@ Related merged [commits](https://github.com/SciSharp/TensorFlow.NET/commit/854a5
On Windows, the tar command does not support extracting archives with symlinks. So when `dotnet pack` runs on Windows it will only package the Windows binaries. On Windows, the tar command does not support extracting archives with symlinks. So when `dotnet pack` runs on Windows it will only package the Windows binaries.


1. Run `dotnet pack` under `src/SciSharp.TensorFlow.Redist` directory in Linux. 1. Run `dotnet pack` under `src/SciSharp.TensorFlow.Redist` directory in Linux.
2. Run `nuget push SciSharp.TensorFlow.Redist.1.14.0.nupkg -k APIKEY -s https://api.nuget.org/v3/index.json`
2. Run `dotnet nuget push SciSharp.TensorFlow.Redist.1.14.0.nupkg -k APIKEY -s https://api.nuget.org/v3/index.json`







src/SciSharp.TensorFlow.Redist/Redist.nuspec → src/SciSharp.TensorFlow.Redist/Redist-CPU.nuspec View File

@@ -9,7 +9,7 @@
<license type="file">LICENSE.txt</license> <license type="file">LICENSE.txt</license>
<licenseUrl>https://aka.ms/deprecateLicenseUrl</licenseUrl> <licenseUrl>https://aka.ms/deprecateLicenseUrl</licenseUrl>
<projectUrl>https://www.tensorflow.org/</projectUrl> <projectUrl>https://www.tensorflow.org/</projectUrl>
<description>$packageId$ contains the TensorFlow C library version $version$ redistributed as a NuGet package.</description>
<description>$packageId$ contains the TensorFlow C library CPU version $version$ redistributed as a NuGet package.</description>
<releaseNotes>https://github.com/tensorflow/tensorflow/releases/tag/v$version$</releaseNotes> <releaseNotes>https://github.com/tensorflow/tensorflow/releases/tag/v$version$</releaseNotes>
<copyright>Copyright 2019 The TensorFlow Authors. All rights reserved.</copyright> <copyright>Copyright 2019 The TensorFlow Authors. All rights reserved.</copyright>
<tags>TensorFlow</tags> <tags>TensorFlow</tags>

+ 26
- 0
src/SciSharp.TensorFlow.Redist/Redist-Windows-GPU.nuspec View File

@@ -0,0 +1,26 @@
<?xml version="1.0" encoding="utf-8"?>
<package xmlns="http://schemas.microsoft.com/packaging/2012/06/nuspec.xsd">
<metadata>
<id>$packageId$</id>
<version>$version$</version>
<authors>The TensorFlow Authors</authors>
<owners>The TensorFlow Authors</owners>
<requireLicenseAcceptance>true</requireLicenseAcceptance>
<license type="file">LICENSE.txt</license>
<licenseUrl>https://aka.ms/deprecateLicenseUrl</licenseUrl>
<projectUrl>https://www.tensorflow.org/</projectUrl>
<description>$packageId$ contains the TensorFlow C library GPU version $version$ redistributed as a NuGet package.</description>
<releaseNotes>https://github.com/tensorflow/tensorflow/releases/tag/v$version$</releaseNotes>
<copyright>Copyright 2019 The TensorFlow Authors. All rights reserved.</copyright>
<tags>TensorFlow</tags>
<dependencies>
<group targetFramework=".NETStandard2.0" />
</dependencies>
</metadata>
<files>
<file src="CommonPackage.props" target="build\netstandard2.0\$packageId$.props" />
<file src="bin\packages\$packageId$\LICENSE.txt" target="LICENSE.txt" />
<file src="bin\packages\$packageId$\THIRD_PARTY_NOTICES.txt" target="THIRD_PARTY_NOTICES.txt" />
<file src="bin\packages\$packageId$\runtimes\**\*" target="runtimes" />
</files>
</package>

src/SciSharp.TensorFlow.Redist/SciSharp.TensorFlow.Redist.nupkgproj → src/SciSharp.TensorFlow.Redist/SciSharp.TensorFlow.Redist-CPU.nupkgproj View File

@@ -17,7 +17,7 @@
<NoBuild>true</NoBuild> <NoBuild>true</NoBuild>
<IncludeBuildOutput>false</IncludeBuildOutput> <IncludeBuildOutput>false</IncludeBuildOutput>


<NuspecFile>Redist.nuspec</NuspecFile>
<NuspecFile>Redist-CPU.nuspec</NuspecFile>
<NuspecProperties>packageId=$(PackageId);version=$(PackageVersion)</NuspecProperties> <NuspecProperties>packageId=$(PackageId);version=$(PackageVersion)</NuspecProperties>
<NuspecBasePath>$(ProjDir)</NuspecBasePath> <NuspecBasePath>$(ProjDir)</NuspecBasePath>



+ 187
- 0
src/SciSharp.TensorFlow.Redist/SciSharp.TensorFlow.Redist-Windows-GPU.nupkgproj View File

@@ -0,0 +1,187 @@
<Project Sdk="Microsoft.NET.Sdk">
<PropertyGroup>
<ProjDir>$(MSBuildThisFileDirectory)</ProjDir>
<BinDir>$(ProjDir)bin\</BinDir>
<ObjDir>$(ProjDir)obj\</ObjDir>

<TargetArchitecture Condition="'$(TargetArchitecture)' == ''">x64</TargetArchitecture>
<TargetFramework>netstandard2.0</TargetFramework>
<TensorFlowVersion>1.14.0</TensorFlowVersion>
<TensorFlowMajorVersion>1</TensorFlowMajorVersion>

<PackageAssetsPath>$(BinDir)packages\</PackageAssetsPath>
<PackageId>$(MSBuildProjectName)</PackageId>
<PackageVersion>$(TensorFlowVersion)</PackageVersion>

<NoBuild>true</NoBuild>
<IncludeBuildOutput>false</IncludeBuildOutput>

<NuspecFile>Redist-Windows-GPU.nuspec</NuspecFile>
<NuspecProperties>packageId=$(PackageId);version=$(PackageVersion)</NuspecProperties>
<NuspecBasePath>$(ProjDir)</NuspecBasePath>

<GenerateNuspecDependsOn>CopyFilesFromArchive</GenerateNuspecDependsOn>

<PackageRid Condition="'$(OS)' == 'Windows_NT'">win</PackageRid>
<PackageRid Condition="'$(OS)' != 'Windows_NT'">linux</PackageRid>
<PackageRid Condition="$([MSBuild]::IsOSPlatform('osx'))">osx</PackageRid>
<PackageRid>$(PackageRid)-$(TargetArchitecture)</PackageRid>

</PropertyGroup>

<PropertyGroup>
<IncludeMLNetNotices>false</IncludeMLNetNotices>
</PropertyGroup>
<ItemGroup>
<TensorFlowConfig Include="windows"
FileExtension=".zip"
FilesFromArchive="lib\tensorflow.dll;
include\tensorflow\c\LICENSE"
Runtime="win-x64"/>

<TensorFlowConfig Condition="'$(OS)' != 'Windows_NT'"
Include="linux"
FileExtension=".tar.gz"
FilesFromArchive="lib\libtensorflow.so;
lib\libtensorflow_framework.so.$(TensorFlowMajorVersion);
include\tensorflow\c\LICENSE"
Runtime="linux-x64" />

<TensorFlowConfig Condition="'$(OS)' != 'Windows_NT'"
Include="darwin" FileExtension=".tar.gz"
FilesFromArchive="lib\libtensorflow.dylib;
lib\libtensorflow_framework.$(TensorFlowMajorVersion).dylib;
include\tensorflow\c\LICENSE"
Runtime="osx-x64" />

<AdditionalDownloadFile Include="https://raw.githubusercontent.com/tensorflow/tensorflow/master/LICENSE"
DownloadFile="$(BinDir)LICENSE" />
</ItemGroup>

<Target Name="PrepareArchives">
<ItemGroup>
<!-- although we could extract all archives on all machines, mac requires a fixup which can only be run on mac
so we split these per-rid and join during the official build packaging. -->
<TensorFlowArchive
Include="@(TensorFlowConfig->'https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-gpu-%(Identity)-x86_64-$(TensorFlowVersion)%(FileExtension)')" />
<!-- set up metdata used by all targets -->
<TensorFlowArchive DownloadFile="$(BinDir)%(FileName)%(Extension)"
DownloadShaFile="$(BinDir)%(FileName)%(Extension).sha"
ExtractDirectory="$(BinDir)%(FileName)"
ExtractSemaphore="$(BinDir)%(FileName)\.extracted"
LocalShaFile="$(MSBuildProjectDirectory)\%(FileName)%(Extension).sha"/>
</ItemGroup>
<Message Importance="high" Text="%(TensorFlowConfig.Runtime)"/>
</Target>

<Target Name="DownloadArchives"
DependsOnTargets="PrepareArchives"
Inputs="$(MSBuildProjectFile)"
Outputs="@(TensorFlowArchive->'%(DownloadFile)');@(AdditionalDownloadFile->'%(DownloadFile)')">
<MakeDir Directories="$(BinDir)" />
<ItemGroup>
<_downloadFiles Include="@(TensorFlowArchive);@(AdditionalDownloadFile)" Url="%(Identity)" DestinationFile="%(DownloadFile)" />
</ItemGroup>
<Message Importance="High" Text="Downloading '%(_downloadFiles.Identity)' to '$(BinDir)'." />
<DownloadFile SourceUrl="%(_downloadFiles.Identity)" DestinationFolder="$(BinDir)">
<Output TaskParameter="DownloadedFile" ItemName="Content" />
</DownloadFile>
</Target>


<Target Name="ValidateAndExtractArchives"
DependsOnTargets="DownloadArchives"
Inputs="@(TensorFlowArchive->'%(DownloadFile)')"
Outputs="@(TensorFlowArchive->'%(ExtractSemaphore)')">

<GetFileHash Files="@(TensorFlowArchive->'%(DownloadFile)')" Algorithm="SHA512">
<Output
TaskParameter="Items"
ItemName="FilesWithHashes" />
</GetFileHash>

<WriteLinesToFile File="%(FilesWithHashes.Identity).sha" Lines="%(FilesWithHashes.FileHash)" Overwrite="true"/>

<!-- If specified we'll update the checked in SHAs with the downloaded ones. -->
<Copy Condition="'$(UpdateSHA)' == 'true'"
SourceFiles="@(TensorFlowArchive->'%(DownloadShaFile)')"
DestinationFiles="@(TensorFlowArchive->'%(LocalShaFile)')" />

<ItemGroup>
<TensorFlowArchive>
<DownloadSha>@(FilesWithHashes->'%(FileHash)')</DownloadSha>
<LocalSha>$([System.IO.File]::ReadAllText('%(LocalShaFile)').Replace("%0A", "").Replace("%0D", ""))</LocalSha>
</TensorFlowArchive>
</ItemGroup>

<Error Condition="!Exists('%(TensorFlowArchive.LocalShaFile)')" Text="SHA file '%(TensorFlowArchive.LocalShaFile)' does not exist. Build with /p:UpdateSHA=true to save it." />

<Message Importance="High" Text="@TensorFlowArchive->'%(TensorFlowArchive.DownloadFile) - %(TensorFlowArchive.LocalSha) - %(TensorFlowArchive.DownloadSha)"/>

<!-- Validate that the downloaded SHAs match the expected checked in SHAs -->
<Error Condition="'%(TensorFlowArchive.LocalSha)' != '%(TensorFlowArchive.DownloadSha)'" Text="Downloaded file '%(TensorFlowArchive.DownloadFile)' has unexpected SHA.%0A expected: %(TensorFlowArchive.LocalSha)%0A --actual: %(TensorFlowArchive.DownloadSha)%0ABuild with /p:UpdateSHA=true if you intentionally changed the URL and wish to update the SHAs, otherwise this could indicate an incomplete download or intercerpted URL and should be examined." />


<!-- The archives are valid, lets extract them, ensuring an empty directory -->
<RemoveDir Directories="@(TensorFlowArchive->'%(ExtractDirectory)')" />
<MakeDir Directories="@(TensorFlowArchive->'%(ExtractDirectory)')" />

<Message Importance="High" Text="Decompressing '%(TensorFlowArchive.DownloadFile)' to '%(TensorFlowArchive.ExtractDirectory)'." />

<Unzip Condition="'%(TensorFlowArchive.FileExtension)' == '.zip'"
SourceFiles="%(TensorFlowArchive.DownloadFile)"
DestinationFolder="%(TensorFlowArchive.ExtractDirectory)" />

<Exec Condition="'$(OS)' != 'Windows_NT' AND '%(TensorFlowArchive.FileExtension)' == '.tar.gz'"
WorkingDirectory="$(MSBuildThisFileDirectory)"
Command="tar -xzm --hard-dereference -f %(TensorFlowArchive.DownloadFile) -C %(TensorFlowArchive.ExtractDirectory)" />

<Exec Condition="'$(OS)' != 'Windows_NT'"
Command="chmod -R +w %(TensorFlowArchive.ExtractDirectory)" />

<Touch Files="@(TensorFlowArchive->'%(ExtractSemaphore)')" AlwaysCreate="true" />
</Target>

<!-- Select the files we want to copy out of each archive. -->
<Target Name="GetFilesFromArchive"
DependsOnTargets="ValidateAndExtractArchives" >
<ItemGroup>
<!-- batch rather than transform so that we can split FilesFromArchive metadata -->
<_fileFromArchive Include="%(TensorFlowArchive.FilesFromArchive)" ExtractDirectory="%(TensorFlowArchive.ExtractDirectory)" Runtime="%(TensorFlowArchive.Runtime)" />
<_fileFromArchive DestinationFile="%(FileName)%(Extension)"/>
<_fileFromArchive PackagePath="runtimes\%(_fileFromArchive.Runtime)\native\%(_fileFromArchive.DestinationFile)" />

<!-- LICENSE from the package is actually THIRD_PARTY_NOTICES-->
<_fileFromArchive Condition="'%(DestinationFile)' == 'LICENSE'" PackagePath="THIRD_PARTY_NOTICES.txt" Runtime="" />

<!-- copy to packaging location -->
<FilesFromArchive Include="@(_fileFromArchive->'%(ExtractDirectory)\%(Identity)')"
TargetPath="$(PackageAssetsPath)$(MSBuildProjectName)\%(PackagePath)" />
<!-- include LICENSE that was downloaded from GitHub -->
<FilesFromArchive Include="$(BinDir)\LICENSE"
TargetPath="$(PackageAssetsPath)$(MSBuildProjectName)\LICENSE.txt" />

<!-- copy to NativeAssets location, only for current RID, so that they may be used by tests -->
<!--<FilesFromArchive Condition="'$(PackageRID)' == '%(_fileFromArchive.Runtime)'"
Include="@(_fileFromArchive->'%(ExtractDirectory)\%(Identity)')"
TargetPath="$(NativeAssetsBuiltPath)\%(_fileFromArchive.DestinationFile)" />-->
</ItemGroup>
</Target>

<Target Name="CopyFilesFromArchive"
DependsOnTargets="GetFilesFromArchive">

<Message Importance="High" Text="@(FilesFromArchive) -> %(FilesFromArchive.TargetPath)" />
<Copy SourceFiles="@(FilesFromArchive)"
DestinationFiles="@(FilesFromArchive->'%(TargetPath)')" />

</Target>

<Target Name="Clean">
<Message Importance="High" Text="Deleting $(BinDir);$(ObjDir)" />
<RemoveDir Directories="$(BinDir);$(ObjDir)" />
</Target>
</Project>

+ 1
- 0
src/SciSharp.TensorFlow.Redist/libtensorflow-gpu-windows-x86_64-1.14.0.zip.sha View File

@@ -0,0 +1 @@
850A27858FA951DF77A78CD1BD78B54F6EE2532DD5A49F0579A7B02C795C62F0212F20177EAEA2BD77BD451A57FBBD1348362492F9E14BFE5CA5028C71711293

+ 49
- 0
src/TensorFlowHub/MnistDataSet.cs View File

@@ -27,5 +27,54 @@ namespace Tensorflow.Hub
labels.astype(dataType); labels.astype(dataType);
Labels = labels; Labels = labels;
} }

public (NDArray, NDArray) GetNextBatch(int batch_size, bool fake_data = false, bool shuffle = true)
{
var start = IndexInEpoch;
// Shuffle for the first epoch
if(EpochsCompleted == 0 && start == 0 && shuffle)
{
var perm0 = np.arange(NumOfExamples);
np.random.shuffle(perm0);
Data = Data[perm0];
Labels = Labels[perm0];
}

// Go to the next epoch
if (start + batch_size > NumOfExamples)
{
// Finished epoch
EpochsCompleted += 1;

// Get the rest examples in this epoch
var rest_num_examples = NumOfExamples - start;
//var images_rest_part = _images[np.arange(start, _num_examples)];
//var labels_rest_part = _labels[np.arange(start, _num_examples)];
// Shuffle the data
if (shuffle)
{
var perm = np.arange(NumOfExamples);
np.random.shuffle(perm);
Data = Data[perm];
Labels = Labels[perm];
}

start = 0;
IndexInEpoch = batch_size - rest_num_examples;
var end = IndexInEpoch;
var images_new_part = Data[np.arange(start, end)];
var labels_new_part = Labels[np.arange(start, end)];

/*return (np.concatenate(new float[][] { images_rest_part.Data<float>(), images_new_part.Data<float>() }, axis: 0),
np.concatenate(new float[][] { labels_rest_part.Data<float>(), labels_new_part.Data<float>() }, axis: 0));*/
return (images_new_part, labels_new_part);
}
else
{
IndexInEpoch += batch_size;
var end = IndexInEpoch;
return (Data[np.arange(start, end)], Labels[np.arange(start, end)]);
}
}
} }
} }

+ 17
- 4
src/TensorFlowHub/MnistModelLoader.cs View File

@@ -15,14 +15,27 @@ namespace Tensorflow.Hub
private const string TEST_IMAGES = "t10k-images-idx3-ubyte.gz"; private const string TEST_IMAGES = "t10k-images-idx3-ubyte.gz";
private const string TEST_LABELS = "t10k-labels-idx1-ubyte.gz"; private const string TEST_LABELS = "t10k-labels-idx1-ubyte.gz";


public static async Task<Datasets<MnistDataSet>> LoadAsync(string trainDir, bool oneHot = false)
public static async Task<Datasets<MnistDataSet>> LoadAsync(string trainDir, bool oneHot = false, int? trainSize = null, int? validationSize = null, int? testSize = null)
{ {
var loader = new MnistModelLoader(); var loader = new MnistModelLoader();
return await loader.LoadAsync(new ModelLoadSetting

var setting = new ModelLoadSetting
{ {
TrainDir = trainDir, TrainDir = trainDir,
OneHot = oneHot
});
OneHot = oneHot,
TrainSize = trainSize
};

if (trainSize.HasValue)
setting.TrainSize = trainSize.Value;

if (validationSize.HasValue)
setting.ValidationSize = validationSize.Value;

if (testSize.HasValue)
setting.TestSize = testSize.Value;

return await loader.LoadAsync(setting);
} }


public async Task<Datasets<MnistDataSet>> LoadAsync(ModelLoadSetting setting) public async Task<Datasets<MnistDataSet>> LoadAsync(ModelLoadSetting setting)


+ 26
- 2
src/TensorFlowNET.Core/Tensors/Tensor.Operators.cs View File

@@ -15,6 +15,7 @@
******************************************************************************/ ******************************************************************************/


using System; using System;
using System.Linq;
using static Tensorflow.Python; using static Tensorflow.Python;


namespace Tensorflow namespace Tensorflow
@@ -63,17 +64,37 @@ namespace Tensorflow
public static Tensor operator *(long constant, Tensor tensor) => BinaryOpWrapper("mul", constant, tensor); public static Tensor operator *(long constant, Tensor tensor) => BinaryOpWrapper("mul", constant, tensor);
public static Tensor operator *(ulong constant, Tensor tensor) => BinaryOpWrapper("mul", constant, tensor); public static Tensor operator *(ulong constant, Tensor tensor) => BinaryOpWrapper("mul", constant, tensor);


public static Tensor operator /(Tensor x, Tensor y) => BinaryOpWrapper("truediv", x, y);
public static Tensor operator /(Tensor x, float y) => BinaryOpWrapper("truediv", x, y);
private static readonly TF_DataType[] _intTfDataTypes = {
TF_DataType.TF_INT8, TF_DataType.TF_INT16, TF_DataType.TF_INT32, TF_DataType.TF_INT64,
TF_DataType.TF_QINT8, TF_DataType.TF_QINT16, TF_DataType.TF_QINT32,
TF_DataType.TF_UINT8, TF_DataType.TF_UINT16, TF_DataType.TF_UINT32, TF_DataType.TF_UINT64
};
public static Tensor operator /(double x, Tensor y) => BinaryOpWrapper("truediv", x, y);
public static Tensor operator /(float x, Tensor y) => BinaryOpWrapper("truediv", x, y); public static Tensor operator /(float x, Tensor y) => BinaryOpWrapper("truediv", x, y);
public static Tensor operator /(int x, Tensor y) => BinaryOpWrapper("floordiv", x, y);
public static Tensor operator /(Tensor x, Tensor y) =>
_intTfDataTypes.Contains(x._dtype)
? BinaryOpWrapper("floordiv", x, y)
: BinaryOpWrapper("truediv", x, y);
public static Tensor operator /(Tensor x, int y) => BinaryOpWrapper("floordiv", x, y);
public static Tensor operator /(Tensor x, float y) => BinaryOpWrapper("truediv", x, y);
public static Tensor operator /(Tensor x, double y) => BinaryOpWrapper("truediv", x, y); public static Tensor operator /(Tensor x, double y) => BinaryOpWrapper("truediv", x, y);


public static Tensor operator %(Tensor x, Tensor y) => BinaryOpWrapper("mod", x, y); public static Tensor operator %(Tensor x, Tensor y) => BinaryOpWrapper("mod", x, y);


public static Tensor operator >(double x, Tensor y) => gen_math_ops.greater(x, y);
public static Tensor operator >(float x, Tensor y) => gen_math_ops.greater(x, y);
public static Tensor operator >(int x, Tensor y) => gen_math_ops.greater(x, y);
public static Tensor operator >(Tensor x, Tensor y) => gen_math_ops.greater(x, y);
public static Tensor operator >(Tensor x, int y) => gen_math_ops.greater(x, y); public static Tensor operator >(Tensor x, int y) => gen_math_ops.greater(x, y);
public static Tensor operator >=(Tensor x, Tensor y) => gen_math_ops.greater_equal(x, y); public static Tensor operator >=(Tensor x, Tensor y) => gen_math_ops.greater_equal(x, y);
public static Tensor operator >(Tensor x, float y) => gen_math_ops.greater(x, y); public static Tensor operator >(Tensor x, float y) => gen_math_ops.greater(x, y);
public static Tensor operator >(Tensor x, double y) => gen_math_ops.greater(x, y); public static Tensor operator >(Tensor x, double y) => gen_math_ops.greater(x, y);

public static Tensor operator <(double x, Tensor y) => gen_math_ops.less(x, y);
public static Tensor operator <(float x, Tensor y) => gen_math_ops.less(x, y);
public static Tensor operator <(int x, Tensor y) => gen_math_ops.less(x, y);
public static Tensor operator <(Tensor x, Tensor y) => gen_math_ops.less(x, y);
public static Tensor operator <(Tensor x, int y) => gen_math_ops.less(x, y); public static Tensor operator <(Tensor x, int y) => gen_math_ops.less(x, y);
public static Tensor operator <=(Tensor x, Tensor y) => gen_math_ops.less_equal(x, y); public static Tensor operator <=(Tensor x, Tensor y) => gen_math_ops.less_equal(x, y);
public static Tensor operator <(Tensor x, float y) => gen_math_ops.less(x, y); public static Tensor operator <(Tensor x, float y) => gen_math_ops.less(x, y);
@@ -99,6 +120,9 @@ namespace Tensorflow
case "add": case "add":
result = gen_math_ops.add(x1, y1, name: scope); result = gen_math_ops.add(x1, y1, name: scope);
break; break;
case "floordiv":
result = gen_math_ops.floor_div(x1, y1, name: scope);
break;
case "truediv": case "truediv":
result = gen_math_ops.real_div(x1, y1, name: scope); result = gen_math_ops.real_div(x1, y1, name: scope);
break; break;


+ 15
- 2
tensorflowlib/README.md View File

@@ -16,6 +16,8 @@ Here are some pre-built TensorFlow binaries you can use for each platform:
- CPU-only: https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-cpu-windows-x86_64-1.14.0.zip - CPU-only: https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-cpu-windows-x86_64-1.14.0.zip
- GPU-enabled: https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-gpu-windows-x86_64-1.14.0.zip - GPU-enabled: https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-gpu-windows-x86_64-1.14.0.zip




### Run in Linux ### Run in Linux


`Install-Package TensorFlow.NET` `Install-Package TensorFlow.NET`
@@ -31,10 +33,21 @@ sudo apt install libgdiplus


More information about [System.Drawing on Linux](<https://www.hanselman.com/blog/HowDoYouUseSystemDrawingInNETCore.aspx>). More information about [System.Drawing on Linux](<https://www.hanselman.com/blog/HowDoYouUseSystemDrawingInNETCore.aspx>).




### Run in Mac OS ### Run in Mac OS


### GPU Tensorflow for windows
Before running verify you installed CUDA and cuDNN


### Tensorflow GPU for Windows

Before running verify you installed CUDA and cuDNN (TensorFlow v1.14 is compatible with CUDA v10.0 and cuDNN v7.4), and make sure the corresponding cuda version is compatible.

```powershell
PM> Install-Package SciSharp.TensorFlow.Redist-Windows-GPU
```




### Build from source for Windows ### Build from source for Windows




+ 5
- 1
test/TensorFlowNET.Examples/ImageProcessing/DigitRecognitionCNN.cs View File

@@ -16,6 +16,7 @@


using NumSharp; using NumSharp;
using System; using System;
using System.Diagnostics;
using Tensorflow; using Tensorflow;
using TensorFlowNET.Examples.Utility; using TensorFlowNET.Examples.Utility;
using static Tensorflow.Python; using static Tensorflow.Python;
@@ -144,6 +145,8 @@ namespace TensorFlowNET.Examples.ImageProcess
float loss_val = 100.0f; float loss_val = 100.0f;
float accuracy_val = 0f; float accuracy_val = 0f;


var sw = new Stopwatch();
sw.Start();
foreach (var epoch in range(epochs)) foreach (var epoch in range(epochs))
{ {
print($"Training epoch: {epoch + 1}"); print($"Training epoch: {epoch + 1}");
@@ -165,7 +168,8 @@ namespace TensorFlowNET.Examples.ImageProcess
var result = sess.run(new[] { loss, accuracy }, new FeedItem(x, x_batch), new FeedItem(y, y_batch)); var result = sess.run(new[] { loss, accuracy }, new FeedItem(x, x_batch), new FeedItem(y, y_batch));
loss_val = result[0]; loss_val = result[0];
accuracy_val = result[1]; accuracy_val = result[1];
print($"iter {iteration.ToString("000")}: Loss={loss_val.ToString("0.0000")}, Training Accuracy={accuracy_val.ToString("P")}");
print($"iter {iteration.ToString("000")}: Loss={loss_val.ToString("0.0000")}, Training Accuracy={accuracy_val.ToString("P")} {sw.ElapsedMilliseconds}ms");
sw.Restart();
} }
} }




+ 674
- 0
test/TensorFlowNET.UnitTest/OperationsTest.cs View File

@@ -467,5 +467,679 @@ namespace TensorFlowNET.UnitTest
} }
#endregion #endregion
} }

private IEnumerable<int> MultiplyArray(IReadOnlyCollection<int> first, IReadOnlyCollection<int> second)
{
if(first.Count != second.Count)
throw new ArgumentException("Arrays should be of equal size!");

var firstEnumerator = first.GetEnumerator();
var secondEnumerator = second.GetEnumerator();
var result = new List<int>();
while (firstEnumerator.MoveNext())
{
secondEnumerator.MoveNext();
result.Add(firstEnumerator.Current * secondEnumerator.Current);
}

firstEnumerator.Dispose();
secondEnumerator.Dispose();

return result;
}
private IEnumerable<float> MultiplyArray(IReadOnlyCollection<float> first, IReadOnlyCollection<float> second)
{
if(first.Count != second.Count)
throw new ArgumentException("Arrays should be of equal size!");

var firstEnumerator = first.GetEnumerator();
var secondEnumerator = second.GetEnumerator();
var result = new List<float>();
while (firstEnumerator.MoveNext())
{
secondEnumerator.MoveNext();
result.Add(firstEnumerator.Current * secondEnumerator.Current);
}

firstEnumerator.Dispose();
secondEnumerator.Dispose();

return result;
}
private IEnumerable<double> MultiplyArray(IReadOnlyCollection<double> first, IReadOnlyCollection<double> second)
{
if(first.Count != second.Count)
throw new ArgumentException("Arrays should be of equal size!");

var firstEnumerator = first.GetEnumerator();
var secondEnumerator = second.GetEnumerator();
var result = new List<double>();
while (firstEnumerator.MoveNext())
{
secondEnumerator.MoveNext();
result.Add(firstEnumerator.Current * secondEnumerator.Current);
}

firstEnumerator.Dispose();
secondEnumerator.Dispose();

return result;
}

[TestMethod]
public void mulOpTests()
{
const int rows = 2; // to avoid broadcasting effect
const int cols = 10;

#region intTest
const int firstIntVal = 2;
const int secondIntVal = 3;

var firstIntFeed = Enumerable.Repeat(firstIntVal, rows * cols).ToArray();
var secondIntFeed = Enumerable.Repeat(secondIntVal, rows * cols).ToArray();
var intResult = MultiplyArray(firstIntFeed, secondIntFeed).Sum();

var a = tf.placeholder(tf.int32, shape: new TensorShape(rows, cols));
var b = tf.placeholder(tf.int32, shape: new TensorShape(rows, cols));
var c = tf.reduce_sum(tf.reduce_sum(tf.multiply(a, b), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator *(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(a * b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator *(Tensor x, int y)
c = tf.reduce_sum(tf.reduce_sum(a * secondIntVal, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator *(int x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(firstIntVal * b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(b, new NDArray(secondIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}
#endregion

#region floatTest
const float firstFloatVal = 2.0f;
const float secondFloatVal = 3.0f;

var firstFloatFeed = Enumerable.Repeat(firstFloatVal, rows * cols).ToArray();
var secondFloatFeed = Enumerable.Repeat(secondFloatVal, rows * cols).ToArray();
var floatResult = MultiplyArray(firstFloatFeed, secondFloatFeed).Sum();

a = tf.placeholder(tf.float32, shape: new TensorShape(rows, cols));
b = tf.placeholder(tf.float32, shape: new TensorShape(rows, cols));
c = tf.reduce_sum(tf.reduce_sum(tf.multiply(a, b), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((float)o, floatResult);
}

// Testing `operator *(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(a * b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((float)o, floatResult);
}

// Testing `operator *(Tensor x, float y)
c = tf.reduce_sum(tf.reduce_sum(a * secondFloatVal, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((float)o, floatResult);
}

// Testing `operator *(float x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(firstFloatVal * b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(b, new NDArray(secondFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((float)o, floatResult);
}
#endregion

#region doubleTest
const double firstDoubleVal = 2.0;
const double secondDoubleVal = 3.0;

var firstDoubleFeed = Enumerable.Repeat(firstDoubleVal, rows * cols).ToArray();
var secondDoubleFeed = Enumerable.Repeat(secondDoubleVal, rows * cols).ToArray();
var doubleResult = MultiplyArray(firstDoubleFeed, secondDoubleFeed).Sum();

a = tf.placeholder(tf.float64, shape: new TensorShape(rows, cols));
b = tf.placeholder(tf.float64, shape: new TensorShape(rows, cols));
c = tf.reduce_sum(tf.reduce_sum(tf.multiply(a, b), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((double)o, doubleResult);
}

// Testing `operator *(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(a * b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((double)o, doubleResult);
}

// Testing `operator *(Tensor x, double y)
c = tf.reduce_sum(tf.reduce_sum(a * secondFloatVal, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((double)o, doubleResult);
}

// Testing `operator *(double x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(firstFloatVal * b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(b, new NDArray(secondDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((double) o, doubleResult);
}
#endregion
}

[TestMethod]
public void divOpTests()
{
const int rows = 2; // to avoid broadcasting effect
const int cols = 10;

#region intTest
const int firstIntVal = 6;
const int secondIntVal = 3;

var firstIntFeed = Enumerable.Repeat(firstIntVal, rows * cols).ToArray();
var secondIntFeed = Enumerable.Repeat(secondIntVal, rows * cols).ToArray();
var intResult = (int)(firstIntFeed.Sum() / (float)secondIntVal);

var a = tf.placeholder(tf.int32, shape: new TensorShape(rows, cols));
var b = tf.placeholder(tf.int32, shape: new TensorShape(rows, cols));
var c = tf.reduce_sum(tf.reduce_sum(gen_math_ops.floor_div(a, b), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator /(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(a / b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator /(Tensor x, int y)
c = tf.reduce_sum(tf.reduce_sum(a / secondIntVal, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator /(int x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(firstIntVal / b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(b, new NDArray(secondIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}
#endregion

#region floatTest
const float firstFloatVal = 6.0f;
const float secondFloatVal = 3.0f;

var firstFloatFeed = Enumerable.Repeat(firstFloatVal, rows * cols).ToArray();
var secondFloatFeed = Enumerable.Repeat(secondFloatVal, rows * cols).ToArray();
var floatResult = MultiplyArray(firstFloatFeed, secondFloatFeed.Select(x => 1/x).ToArray()).Sum();

a = tf.placeholder(tf.float32, shape: new TensorShape(rows, cols));
b = tf.placeholder(tf.float32, shape: new TensorShape(rows, cols));
c = tf.reduce_sum(tf.reduce_sum(tf.divide(a, b), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((float)o, floatResult);
}

// Testing `operator /(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(a / b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((float)o, floatResult);
}

// Testing `operator /(Tensor x, float y)
c = tf.reduce_sum(tf.reduce_sum(a / secondFloatVal, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((float)o, floatResult);
}

// Testing `operator /(float x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(firstFloatVal / b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(b, new NDArray(secondFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((float)o, floatResult);
}
#endregion

#region doubleTest
const double firstDoubleVal = 6.0;
const double secondDoubleVal = 3.0;

var firstDoubleFeed = Enumerable.Repeat(firstDoubleVal, rows * cols).ToArray();
var secondDoubleFeed = Enumerable.Repeat(secondDoubleVal, rows * cols).ToArray();
var doubleResult = MultiplyArray(firstDoubleFeed, secondDoubleFeed.Select(x => 1/x).ToArray()).Sum();

a = tf.placeholder(tf.float64, shape: new TensorShape(rows, cols));
b = tf.placeholder(tf.float64, shape: new TensorShape(rows, cols));
c = tf.reduce_sum(tf.reduce_sum(tf.divide(a, b), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((double)o, doubleResult);
}

// Testing `operator /(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(a / b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((double)o, doubleResult);
}

// Testing `operator /(Tensor x, double y)
c = tf.reduce_sum(tf.reduce_sum(a / secondFloatVal, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((double)o, doubleResult);
}

// Testing `operator /(double x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(firstFloatVal / b, 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(b, new NDArray(secondDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((double)o, doubleResult);
}
#endregion
}

[TestMethod]
public void greaterThanOpTests()
{
const int rows = 2; // to avoid broadcasting effect
const int cols = 10;

#region intTest
const int intThreshold = 10;

var firstIntFeed = Enumerable.Range(0, rows * cols).ToArray();
var secondIntFeed = Enumerable.Repeat(intThreshold, rows * cols).ToArray();
var intResult = firstIntFeed.Count(elem => elem > intThreshold);
var intResultTwo = firstIntFeed.Count(elem => elem < intThreshold);

var a = tf.placeholder(tf.int32, shape: new TensorShape(rows, cols));
var b = tf.placeholder(tf.int32, shape: new TensorShape(rows, cols));
var c = tf.reduce_sum(tf.reduce_sum(tf.cast(tf.greater(a, b), tf.int32), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator >(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a > b, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator >(Tensor x, int y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a > intThreshold, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator >(int x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(intThreshold > a, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResultTwo);
}
#endregion

#region floatTest
const float floatThreshold = 10.0f;

var firstFloatFeed = Enumerable.Range(0, rows * cols).Select(elem => (float)elem).ToArray();
var secondFloatFeed = Enumerable.Repeat(floatThreshold, rows * cols).ToArray();
var floatResult = firstFloatFeed.Count(elem => elem > floatThreshold);
var floatResultTwo = firstFloatFeed.Count(elem => elem < floatThreshold);

a = tf.placeholder(tf.float32, shape: new TensorShape(rows, cols));
b = tf.placeholder(tf.float32, shape: new TensorShape(rows, cols));
c = tf.reduce_sum(tf.reduce_sum(tf.cast(tf.greater(a, b), tf.int32), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, floatResult);
}

// Testing `operator >(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a > b, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, floatResult);
}

// Testing `operator >(Tensor x, float y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a > floatThreshold, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, floatResult);
}

// Testing `operator >(float x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(floatThreshold > a, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, floatResultTwo);
}
#endregion

#region doubleTest
const double doubleThreshold = 10.0;

var firstDoubleFeed = Enumerable.Repeat(0, rows * cols).Select(elem => (double)elem).ToArray();
var secondDoubleFeed = Enumerable.Repeat(doubleThreshold, rows * cols).ToArray();
var doubleResult = firstDoubleFeed.Count(elem => elem > doubleThreshold);
var doubleResultTwo = firstDoubleFeed.Count(elem => elem < doubleThreshold);

a = tf.placeholder(tf.float64, shape: new TensorShape(rows, cols));
b = tf.placeholder(tf.float64, shape: new TensorShape(rows, cols));
c = tf.reduce_sum(tf.reduce_sum(tf.cast(tf.greater(a, b), tf.int32), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, doubleResult);
}

// Testing `operator >(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a > b, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, doubleResult);
}

// Testing `operator >(Tensor x, double y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a > doubleThreshold, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, doubleResult);
}

// Testing `operator >(double x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(doubleThreshold > a, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, doubleResultTwo);
}
#endregion
}

[TestMethod]
public void lessThanOpTests()
{
const int rows = 2; // to avoid broadcasting effect
const int cols = 10;

#region intTest
const int intThreshold = 10;

var firstIntFeed = Enumerable.Range(0, rows * cols).ToArray();
var secondIntFeed = Enumerable.Repeat(intThreshold, rows * cols).ToArray();
var intResult = firstIntFeed.Count(elem => elem < intThreshold);
var intResultTwo = firstIntFeed.Count(elem => elem > intThreshold);

var a = tf.placeholder(tf.int32, shape: new TensorShape(rows, cols));
var b = tf.placeholder(tf.int32, shape: new TensorShape(rows, cols));
var c = tf.reduce_sum(tf.reduce_sum(tf.cast(tf.less(a, b), tf.int32), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator <(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a < b, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator <(Tensor x, int y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a < intThreshold, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResult);
}

// Testing `operator <(int x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(intThreshold < a, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstIntFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, intResultTwo);
}
#endregion

#region floatTest
const float floatThreshold = 10.0f;

var firstFloatFeed = Enumerable.Range(0, rows * cols).Select(elem => (float)elem).ToArray();
var secondFloatFeed = Enumerable.Repeat(floatThreshold, rows * cols).ToArray();
var floatResult = firstFloatFeed.Count(elem => elem < floatThreshold);
var floatResultTwo = firstFloatFeed.Count(elem => elem > floatThreshold);

a = tf.placeholder(tf.float32, shape: new TensorShape(rows, cols));
b = tf.placeholder(tf.float32, shape: new TensorShape(rows, cols));
c = tf.reduce_sum(tf.reduce_sum(tf.cast(tf.less(a, b), tf.int32), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, floatResult);
}

// Testing `operator <(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a < b, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, floatResult);
}

// Testing `operator <(Tensor x, float y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a < floatThreshold, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, floatResult);
}

// Testing `operator <(float x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(floatThreshold < a, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstFloatFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, floatResultTwo);
}
#endregion

#region doubleTest
const double doubleThreshold = 10.0;

var firstDoubleFeed = Enumerable.Repeat(0, rows * cols).Select(elem => (double)elem).ToArray();
var secondDoubleFeed = Enumerable.Repeat(doubleThreshold, rows * cols).ToArray();
var doubleResult = firstDoubleFeed.Count(elem => elem < doubleThreshold);
var doubleResultTwo = firstDoubleFeed.Count(elem => elem > doubleThreshold);

a = tf.placeholder(tf.float64, shape: new TensorShape(rows, cols));
b = tf.placeholder(tf.float64, shape: new TensorShape(rows, cols));
c = tf.reduce_sum(tf.reduce_sum(tf.cast(tf.less(a, b), tf.int32), 1));

using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, doubleResult);
}

// Testing `operator <(Tensor x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a < b, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))),
new FeedItem(b, new NDArray(secondDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, doubleResult);
}

// Testing `operator <(Tensor x, double y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(a < doubleThreshold, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, doubleResult);
}

// Testing `operator <(double x, Tensor y)
c = tf.reduce_sum(tf.reduce_sum(tf.cast(doubleThreshold < a, tf.int32), 1));
using (var sess = tf.Session())
{
var o = sess.run(c,
new FeedItem(a, new NDArray(firstDoubleFeed, new Shape(rows, cols))));
Assert.AreEqual((int)o, doubleResultTwo);
}
#endregion
}
} }
} }

Write Preview
Loading…
Cancel
Save