AutoGraphSharp

This is the AutoGraph implementation for C# code.

To test this code, clone the repo and open the example project. AutoGraphSharp generates the Tensorflow Graph creation code at compile time.

In a future release, a development NuGet package will be publihed.

Example

This function:

    [AutoGraph]
    public static int Function(int a, int b)
    {
        var c = (a + b *3 )/a;

        if (a > b)
            c = 1;
        else
            c = 2;

        return c;
    }

will be automatically translated to this function:

    [AutoGraph]
    public static int Function(int a, int b, TensorFlow.TFSession session)
    {
        var runner = session.GetRunner();
        var _a = new AutoTFOutput(session.Graph.Placeholder(TFTensor.TensorTypeFromType(a.GetType())), session);
        runner.AddInput(_a, new TFTensor(a));
        var _b = new AutoTFOutput(session.Graph.Placeholder(TFTensor.TensorTypeFromType(b.GetType())), session);
        runner.AddInput(_b, new TFTensor(b));
        return (int)runner.Run(_Function(_a, _b, session)).GetValue();
    }

    [AutoGraph]
    public static AutoTFOutput _Function(AutoTFOutput a, AutoTFOutput b, TensorFlow.TFSession session)
    {
        var c = (a + b * 3) / a;
        var predicate1 = a > b;
        Func<Tuple<AutoTFOutput>> ifTrue1 = () =>
        {
            var _c = c;
            _c = new AutoTFOutput(session.Graph.Const(1), session);
            return Tuple.Create(_c);
        };
        Func<Tuple<AutoTFOutput>> ifFalse1 = () =>
        {
            var _c = c;
            _c = new AutoTFOutput(session.Graph.Const(2), session);
            return Tuple.Create(_c);
        };
        var res = new AutoCond<Tuple<AutoTFOutput>>(predicate1, ifTrue1, ifFalse1, session);
        res.Deconstruct(out c);
        return c;
    }

Note: This is an incomplete work in progress. The first goal if to implement a generic if statement Graph translation that will work for any C# if permutation. Once there, the rest of the C# language features will follow (while loops, switch cases etc).