Swift For tensorflow Workshop

Tensor flow installation

• Download and install the xCode11 realese for Swift for Tensorflow from https://github.com/tensorflow/swift/blob/master/Installation.md

• Install xcode11 beta on your machine.

• Make a MacOS command line tool project

• Open your project in xCode 11 and set the toolchain to use the one you just downloaded.

• In your projects build settings

  • Add /Library/Developer/Toolchains/swift-tensorflow-RELEASE-0.4.xctoolchain/usr/lib/swift/macosx to your Runpath Search Paths
  • Change the optimization level in the Swift Compiler - Code generation Section to -O

• Add libtensorflow.so and libtensorflow_framework.so to Link Binary With Libraries in Build Phases, these libraries are located in /Library/Developer/Toolchains/swift-tensorflow-RELEASE-0.4.xctoolchain/usr/lib/swift/macosx

• In File > Project Settings change build system to a legacy build system.

• Try import TensortFlow and import Python and make sure you can compile and run (if you cant cry a little and throw a tantram and mayber xcode will have pity)

• Check which version of python your tensorflow is using by typing

  import Python
  print(Python.version)
  

  Make sure all python libraries you want to used are installed for that version of python (just pip/pip3 install)

  Usefull Links

  • Swift for tensorflow github
  • Swift for tensorflow python interoperability tutorial
  • Swift for tensorflow model training tutorial
  • Github of my more complex project for detecting sign language letters from images