v1.6.0beta for UE 5.3

Changes

• Component API changes for expected server upgrade
• Matches https://github.com/getnamo/ml-remote-server/releases/tag/v0.4.0 which had most fixes

See https://github.com/getnamo/TensorFlow-Unreal-Examples/releases/tag/v1.6.0beta for working mnist tensorflow 2.11 example.

v1.5.0beta for UE 5.3

Beta Release, not all features may work correctly or be properly documented!

Changes

• Compile fixes for Unreal platform update to 5.3
• Socket.io plugin updated to 2.7.1
• MachineLearningRemote plugin updated to 0.9.0

Notes

• Uses MachineLearningRemote base. See https://github.com/getnamo/MachineLearningRemote-Unreal for API. Use a https://github.com/getnamo/ml-remote-server, example will auto connect to this server at port 8080.
• Early API, embedded python support is limited.
• Server setup notes found here: https://github.com/getnamo/ml-remote-server/blob/master/README.md

v1.4.0beta for UE 5.1

Beta Release, not all features may work correctly or be properly documented!

Changes

• Compile fixes for Unreal platform update to 5.1
• Socket.io plugin updated to 2.5.0
• MachineLearningRemote plugin updated to 0.8.0
• Fixed MLRemote server loop bugs

Notes

• Uses MachineLearningRemote base. Tensorflow no longer depends on UnrealEnginePython and instead depends on user installed python environment (or partial support for embedded python). Allows full flexibility of TF versions. See https://github.com/getnamo/machine-learning-remote-ue4 for API
• Early API, embedded python support is limited.
• Server setup notes found here: https://github.com/getnamo/ml-remote-server/blob/master/README.md

v1.3.0beta for UE 5.0

Beta Release, not all features may work correctly or be properly documented!

Changes

• Update to Unreal 5.0 production release
• Socket.io plugin updated to 2.0.1
• MachineLearningRemote plugin updated to 0.7.0
• Plugin repo name changes to support future unreal versions

Notes

• Uses MachineLearningRemote base. Tensorflow no longer depends on UnrealEnginePython and instead depends on user installed python environment (or partial support for embedded python). Allows full flexibility of TF versions. See https://github.com/getnamo/machine-learning-remote-ue4 for API
• Early API, embedded python support is limited.
• Server setup notes found here: https://github.com/getnamo/ml-remote-server/blob/master/README.md

v1.0.0alpha2 for UE 4.25

Early Release, not all features may work correctly!

Changes

• Refactor to use machine-learning-remote-ue4 base. Tensorflow no longer depends on UnrealEnginePython and instead depends on user installed python environment (or partial support for embedded python). Allows full flexibility of TF versions. See https://github.com/getnamo/machine-learning-remote-ue4 for API
• UE 4.25 support
• Early API, embedded python support is limited.
• Server setup notes found here: https://github.com/getnamo/ml-remote-server/blob/master/README.md

v0.14.0 for UE4.23

Changes

• UnrealEnginePython plugin dependency updated to 1.10.0
• Socketio-client-ue4 plugin dependency update to 1.1.0

Note on GPU release
Tensorflow v1.12.0 GPU depends on CUDA v9.0 and cuDNN v7 see https://www.tensorflow.org/install/gpu for detailed instructions on installing those pre-requisites. If you can't satisfy those requirements, use the cpu version.

v0.13.0 for UE4.22

Changes

• UnrealEnginePython plugin dependency updated to 1.9.0
• Socketio-client-ue4 plugin dependency update to 1.0.0
• TensorFlow default updated to 1.12.0. Requires h5py to be 2.7.0.

Note on GPU release
Tensorflow v1.12.0 GPU depends on CUDA v9.0 and cuDNN v7 see https://www.tensorflow.org/install/gpu for detailed instructions on installing those pre-requisites. If you can't satisfy those requirements, use the cpu version.

v0.11.0 for UE4.21

Changes

• UnrealEnginePython plugin dependency updated to 1.8.0
• Socketio-client-ue4 plugin dependency update to 0.10.0
• Compile fixes for 4.21

Note on GPU release
Tensorflow v1.10.0 GPU depends on CUDA v9.0 and cuDNN v7 see https://www.tensorflow.org/install/gpu for detailed instructions on installing those pre-requisites. If you can't satisfy those requirements, use the cpu version.

v0.10.1 for UE4.20

Changes

• Packaging will no longer attempt to copy site-packages. Either run your packaged build once to auto-pull dependencies or copy manually after packaging is complete. This cannot be automated due to staging directory having nearly always a too long path name to copy correctly (>260 chars).
• runJsonInput is now multi-threaded if tensorflow component is set to use multithreading (this is default behavior).
• run_on_gt is now a separate function
• upythread.run_on_bt changed to accept arguments as second argument, and a reflow gamethread callback as third argument. This allows for passing functions to a background thread with arguments.
• breaking api bumps plugin to 0.10.1
• UnrealEnginePython updated to 1.7.4 for threading changes.
• python36.dll will auto copy to packaged build project binaries directory so it will start on a fresh system. (0.10.1)

Note on GPU release
Tensorflow v1.10.0 GPU depends on CUDA v9.0 and cuDNN v7 see https://www.tensorflow.org/install/gpu for detailed instructions on installing those pre-requisites. If you can't satisfy those requirements, use the cpu version.

v0.9.3 for UE4.20

Changes

• Compile fixes for packaging.
• UnrealEnginePython and Tensorflow Plugins will now auto-copy their plugin scripts to packaged builds
• Embedded python will-auto copy to packaged build. site-packages may cause path too long error, delete {Project Root}/Plugins/UnrealEnginePython/Binaries/Win64/Lib/site-packages folder before packaging. Can manually copy over once packaged.
• Your project content/scripts will also auto copy (UnrealEnginePython v1.7.1 update)
• Tensorflow upgraded to 1.10
• Socket.io plugin updated to 0.9.3

Intermediate release, use https://github.com/getnamo/tensorflow-ue4/releases/tag/0.10.0 instead.