Where communities thrive


  • Join over 1.5M+ people
  • Join over 100K+ communities
  • Free without limits
  • Create your own community
People
Activity
    Eli Belash
    @Nucs
    If you are using Tensorflow.NET, you can't upgrade to 0.20 just yet
    Haiping
    @Oceania2018
    @jingle1000 Try v0.20, the fresh version. haha
    Johnathan Ingle
    @jingle1000
    @Nucs Yes, I am using Tensorflow.NET
    Eli Belash
    @Nucs
    I'm afraid we havn't publish tfnet with support for 0.20 numsharp
    Are you getting an exception?
    Johnathan Ingle
    @jingle1000
    No exceptions, I am still trying to figure out how exactly to use numsharp though.
    Anyone planning to participate, for the sake of positioning/improving Torch.NET and TorchSharp?
    not sure if the purpose if that hackaton might fit - but who knows
    Eli Belash
    @Nucs
    @jingle1000, I think this might help
    Amine Chadly
    @chaami_gitlab
    Hi, I think I have a working version of a meta distrib nuget for TensorFlow in a personal branch (https://github.com/chaami/TensorFlow.NET/tree/redist) but I have the feeling that you guys have taken a different direction ? :wink:
    I have put a power shell script that takes care of downloading the archives from google.. it require to have 7z installed.
    It seems to handle the links just fine even on Windows. I don't see any obvious way of testing without pushing the nugets first... but I will think about it further... Any ideas / thoughts ?
    Eli Belash
    @Nucs
    This is for @Oceania2018 to answer
    Amine Chadly
    @chaami_gitlab
    No rush :wink:
    Haiping
    @Oceania2018
    @chaami_gitlab Thanks for your kind work. I've commented in SciSharp/TensorFlow.NET#361.
    Amine Chadly
    @chaami_gitlab
    I have create a pull request, but so far I have only tested that the example project runs successfully when depending on the CPU nuget.
    Johnathan Ingle
    @jingle1000
    @Nucs That helps, thanks!
    Brendan Mulcahy
    @BrendanMulcahy
    Does Keras.NET have a runtime dependency on python or just at build time?
    Haiping
    @Oceania2018
    @BrendanMulcahy Have to install python runtime.
    Brendan Mulcahy
    @BrendanMulcahy
    I see, not ideal for my use case :)
    I can stick with TensorFlow.NET
    The KerasNetCore in TensorFlow.NET will basically become Keras implemented fully using the constructs from TensorFlow.NET, correct?
    Haiping
    @Oceania2018
    @BrendanMulcahy KerasNetCore is written by @arnavdas88 . He may give you best answer.
    Arnav Das
    @arnavdas88
    @BrendanMulcahy The KerasNetCore will be implemented as the native keras I will complete it when I will get time.... Right now the backend of KerasNetCore is not similar to that of the real keras... It is kind of an example thing but I will add a PR for the keras soon....
    Brendan Mulcahy
    @BrendanMulcahy
    Sounds cool, looking forward to that!
    Arnav Das
    @arnavdas88
    @BrendanMulcahy what's your use case anyway....?? You said that dependency on python is not ideal for your use case...??
    Brendan Mulcahy
    @BrendanMulcahy
    I’m working on building a computer game using machine learning. I don’t want the installation process to be very complicated (to reduce error rates, etc.)
    I linked a YouTube video somewhere above showing my prototype game
    Eli Belash
    @Nucs
    Do you use Unity engine?
    Brendan Mulcahy
    @BrendanMulcahy
    I’m currently using TensorFlowSharp and a Keras wrapper for that, but the api quality is pretty low and there are tons of issues
    The prototype is using Unity yeah, I started out with their ml-agents library but the library didn’t really work how I wanted it to so I ditched that
    Im dumping Unity for something a little more lightweight, since Im going to need as many CPU cycles as possible to be running the network training :)
    Brendan Mulcahy
    @BrendanMulcahy
    I've got another completely free weekend coming up, so I'll keep plugging along on some of the neural net architectures I have been exploring. I will submit PRs when I find problems and know how to fix them and I can open issues if they are above my head
    Eli Belash
    @Nucs
    👍
    Brendan Mulcahy
    @BrendanMulcahy
    Im really interested in getting a working example of temporal convolutional network going. I think the performance could be a lot better than RNNs
    Eli Belash
    @Nucs
    RNNs are extremely slow, CNNs can be 30 times fasters on CPU and even faster on GPU
    Im refering to the learning process but I'm quite sure prediction time are faster too
    Brendan Mulcahy
    @BrendanMulcahy
    Yeah exactly, super interesting area of work. I couldnt get the TCN working for my prototype so I ended up adding a sliding windows where I averaged the input and output action (i.e. adding momentum to the agents decisions) and that worked surprisingly well considering how low tech a solution it was :D
    Eli Belash
    @Nucs
    Normalizing (in one way or an other) the input and output is indeed surprisingly powerful, accuracy-wise
    Johnathan Ingle
    @jingle1000
    @BrendanMulcahy What other game engine were you thinking of using if you are planning on dumping unity? I don't know many .NET game engines other than unity and godot
    Semih
    @butcher211
    Hi guys, can you please answer me following short question: Is the Object Detection API yet implemented that I can use it in my C# projects/solutions? I need to use TensorFlow and Darknet at the same time for my master degree's project. I mean this one: https://github.com/tensorflow/models/tree/master/research/object_detection
    Semih
    @butcher211
    thank you! <3 If you are interested I would like to inform you when im finished with my implementation part. (Q1 2020) Maybe it could help the community then
    Haiping
    @Oceania2018
    @butcher211 Welcome to PR if you do it base on tf.net.
    cyeduarte
    @cyeduarte
    I installed numsharp using Nuget for my Unity project, but it seems Unity can't read or find the package. I get the error: "The type or namespace name 'NumSharp' could not be found".
    Brendan Mulcahy
    @BrendanMulcahy
    @jingle1000 Monogame
    Brendan Mulcahy
    @BrendanMulcahy

    I installed numsharp using Nuget for my Unity project, but it seems Unity can't read or find the package. I get the error: "The type or namespace name 'NumSharp' could not be found".

    I don't believe you can use Unity and nuget directly. Create another project and install the package, then manually copy the DLLs into your Unity project

    Haiping
    @Oceania2018
    TensorFlow RC released: https://www.tensorflow.org/beta
    Haiping
    @Oceania2018
    image.png
    @Nucs ml.net need this function to be public, if not, which function is the alternative one? and ml.net use the c_api directly to do inferring(scoring) task. So we can't make c_api interface to be private. Any more reason, we can discuss with @codemzs .