.NET UDFs and have them work in the tensorflow graph

[Oceania2018]

That would be cool, if it would be possible to write .NET UDFs and have them work in the tensorflow graph.

Basically, when this is in place all kinds of fancy ops can be built in .Net including existing .Net functionality.
https://github.com/eaplatanios/tensorflow_scala/blob/master/modules/jni/src/main/native/ops/jvm_callback_op.cc
we'd need one for .NET.
Technically this is for the JVM and we'd need one for the CLR.

Tensorflow is like a distributed computation engine that could be used in many scenarios, not only in deep learning.

This is offical docs for Custom Operation:
https://www.tensorflow.org/guide/extend/op
https://github.com/tensorflow/custom-op

[Oceania2018]

I remember TF.NET is callbacked for deallocating memory when Tensor is passed into C++. That’ll be good start.

Deallocator deallocator = (IntPtr values, IntPtr len, ref bool closure) =>
{
    Marshal.FreeHGlobal(values);
    closure = true;
};

https://github.com/SciSharp/TensorFlow.NET/blob/master/src/TensorFlowNET.Core/Tensors/Tensor.Creation.cs