jordenhill/birdbrain

A neural network framework written in Swift.

birdbrain's Issues

RNN

Jorden I'm curious if you've ever used the RNN to do any actual predictions or classification. Your readme says that none of these are usable but I'm thinking that perhaps the readme is not updated. Have you successfully used the RNN in birdbrain for some task?

Swift 4 support?

Hello,

I'm very much interested in using this framework. However, it needs to be ported to swift 4 (or swift 3.2). I tried doing that, but I'm facing some issues. Are you planning to convert it?

Thnx!

Swift-AI merge?

Hi :) First of all great work! I am also using swift and I am interested in implementing an RNN. I found Swift-AI and I thought that instead of duplicating efforts you could submit your library for inclusion at Swift-AI? Would be nice to get the swift machine learning libraries consolidated..

I think your code would make an especially useful contribution because it uses metal/gpu.

Just a thought. Keep up the good work!

ld: framework not found Birdbrain

Hello :)

I opened up the birdbrain project but it fails to build. The error message is ld: framework not found Birdbrain. I can't see why the linked would need to find the Birdbrain framework in order to link the framework to begin with. It's like trying to link with itself ...

I'm on Xcode 7.2.1

Here is the detailed error message:

Ld /Users/jubei/Library/Developer/Xcode/DerivedData/Birdbrain-ccpqedrlqhtglybhribchsdjwtsj/Build/Products/Debug/Birdbrain.framework/Versions/A/Birdbrain normal x86_64
cd /Users/jubei/coding/Birdbrain
export MACOSX_DEPLOYMENT_TARGET=10.11
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang -arch x86_64 -dynamiclib -isysroot /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.11.sdk -L/Users/jubei/Library/Developer/Xcode/DerivedData/Birdbrain-ccpqedrlqhtglybhribchsdjwtsj/Build/Products/Debug -F/Users/jubei/Library/Developer/Xcode/DerivedData/Birdbrain-ccpqedrlqhtglybhribchsdjwtsj/Build/Products/Debug -filelist /Users/jubei/Library/Developer/Xcode/DerivedData/Birdbrain-ccpqedrlqhtglybhribchsdjwtsj/Build/Intermediates/Birdbrain.build/Debug/Birdbrain.build/Objects-normal/x86_64/Birdbrain.LinkFileList -install_name @rpath/Birdbrain.framework/Versions/A/Birdbrain -Xlinker -rpath -Xlinker @executable_path/../Frameworks -Xlinker -rpath -Xlinker @loader_path/Frameworks -mmacosx-version-min=10.11 -L/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/lib/swift/macosx -Xlinker -add_ast_path -Xlinker /Users/jubei/Library/Developer/Xcode/DerivedData/Birdbrain-ccpqedrlqhtglybhribchsdjwtsj/Build/Intermediates/Birdbrain.build/Debug/Birdbrain.build/Objects-normal/x86_64/Birdbrain.swiftmodule -framework Metal -framework Accelerate -framework Birdbrain -single_module -compatibility_version 1 -current_version 1 -Xlinker -dependency_info -Xlinker /Users/jubei/Library/Developer/Xcode/DerivedData/Birdbrain-ccpqedrlqhtglybhribchsdjwtsj/Build/Intermediates/Birdbrain.build/Debug/Birdbrain.build/Objects-normal/x86_64/Birdbrain_dependency_info.dat -o /Users/jubei/Library/Developer/Xcode/DerivedData/Birdbrain-ccpqedrlqhtglybhribchsdjwtsj/Build/Products/Debug/Birdbrain.framework/Versions/A/Birdbrain

ld: framework not found Birdbrain
clang: error: linker command failed with exit code 1 (use -v to see invocation)

NaN output on calculateLoss()

Jorden I tried to train your rnn with a very simple method:

Target: 1 for random positive gaussian numbers
Target: 0 for random negative gaussian numbers

Loss seems to become not-a-number after a few training runs. If you're interested in getting to the bottom of this then please run the following playground alongside birdbrain:

var rnn = RecurrentNeuralNetwork(inputDim: 2,hiddenDim: 1,useMetal: false,activationFunction:"tangent")

for x in 0...1000 {
    //generate a random positive float
    let pos = abs(rand_gauss())
    //generate a random negative float
    let neg = abs(rand_gauss())*(-1)
    //train the RNN
    rnn.backprop([[pos]], target: [Array(count: 1, repeatedValue: 1)], learningRate: 0.001)
    rnn.backprop([[neg]], target: [Array(count: 1, repeatedValue: 0)], learningRate: 0.001)
    var loss = rnn.calculateLoss([[abs(rand_gauss())]], target: [[1]], numExamples: 1)
    Swift.print(loss)
}

//attempt at basic prediction
let pos = abs(rand_gauss())
let neg = abs(rand_gauss())*(-1)

Swift.print(rnn.predict([[pos]]))
Swift.print(rnn.predict([[neg]]))