Can't compute model output with NNSharp

[Nadjib23]

Hi, i wanted to try NNSharp and i made a model with keras to classify iris flowers (classic) so the model takes 4 inputs and it gives an output of (0 or 1 or 2), it's a simple example juste to see how it works, here is my code :

       `double[] input = { 5.1, 3.5, 1.4, 0.2 };
        // Read the previously created json.
        var reader = new ReaderKerasModel("C:/Users/ben-n_000/Desktop/m.json");
        SequentialModel model = reader.GetSequentialExecutor();

        DataArray data = new DataArray(input.Length);
        for(int i=0;i< input.Length; i++)
        {
            data[i] = input[i];
        }
        Data2D k = (Data2D)data;

        IData output = model.ExecuteNetwork(k);

        Data2D ou = (Data2D)output;

        //output.
        Console.WriteLine(ou. !!!!);`

Here i can't find a methode to display the output and i can't find the right documentation !!!

[adamtiger]

Hi! Thank you for your question!

Currently there is no function for creating a string representation for Data2D. So you should write the elements for the screen one by one. But I do not recommend this. Instead create another Data2D tensor with the expected results and compare it to the output of your network. To understand the shape of the output just use the NNSharp docs. There, under Keras you can find the implemented layers and it will tell you what is the output. So just look the documentation for the last layer of your network.

Let me know if your test is successful or not!

[Nadjib23]

Hi thank you for you quick answer, i find a way to get a string representation of the different layers of the keras model, that show the inputs and outputs for all the layers including the last layer (output)

    `// String representation of the keras model 

       SequentialModelData d = (SequentialModelData)model.GetSummary();

        Console.WriteLine(d.GetStringRepresentation());
      `

since i can see the model architecture, and know the output dimension (1 output for my case), i can print out the output (at index 0 in the channels parameter) because i have only one output

    ` IData output = model.ExecuteNetwork(k);

     Data2D ou = (Data2D)output;

    //output : 3rd parameter 0 is the channels parameter

    Console.WriteLine(ou[0, 0, 0, 0]);

`

[adamtiger]

Yes, that is good solution as well.