Cntk: Creating Custom Model in C# API

Created on 17 Apr 2017  路  10Comments  路  Source: microsoft/CNTK

Hi All,
I am a newbie in CNTK. Can we create a custom model in C# API? Is it possible to save the model in database?
Thanks in advance.

feature request

Most helpful comment

I find it ironic that a Microsoft person create C# bindings for Tensorflow and not for CNTK.
https://github.com/migueldeicaza/TensorFlowSharp
Some timeline from MS will be nice.

All 10 comments

The C# API currently supports only model evaluation/inference. For training, creating or saving a model, you need to use Python or C++.

Thanks,

Thanks a lot

I was sorry to see model building under C# did not make the cut in the release candidate. Is there any timeline for this functionality going forward in a future release.

I find it ironic that a Microsoft person create C# bindings for Tensorflow and not for CNTK.
https://github.com/migueldeicaza/TensorFlowSharp
Some timeline from MS will be nice.

@larcai I was also particularly surprised by this. I expected some kind of deep integration between CNTK and C#/F# from Microsoft.

@larcai Microsoft future for AI must be, in addition to Azure cloud, the wide presence (cross platform) of mobile AI is equally critical.

I see both CNTK for .NET and tensorflowSharp (some challenge right now) to play an important role in that.

=> With mobile cellular PC : WinOnArm, it would be great to start seeing the performance of CNTK on ARM (Native)

@larcai Just saw your PR . Do you have working example for C# doing what python is doing for those tutorials illustrated through the jupyter notebooks?

that was some time ago
I'll see what I can do with the latest v2.0 release and make an new PR

We are working on C#/.NET support. Will close this thread for now. Please follow our iteration plan:
https://github.com/Microsoft/CNTK/wiki/Iteration-Plans
to track progress.

Could be a useful project to use for developing model and training them: https://github.com/deepakkumar1984/SiaNet

A c# wrapper on CNTK backend for easy stacking of neural network.

Was this page helpful?
0 / 5 - 0 ratings