[ANSWERED] Error on getting output from Softmax layer.

[Poaz]

Hello,

I'm trying to get the output of the final layer from a model trained with Keras using tensorflow as backend. I'm using the tensorflowsharp with Unity.

I'm trying to access the last layer to get the final output, but i get this exception:

TFException: Shape [?,10] is not fully defined
	 [[Node: dense_one_input = Placeholder[dtype=DT_FLOAT, shape=[?,10], _device="/job:localhost/replica:0/task:0/cpu:0"]()]]
TensorFlow.TFStatus.CheckMaybeRaise (TensorFlow.TFStatus incomingStatus, System.Boolean last) (at <6ed6db22f8874deba74ffe3e566039be>:0)
TensorFlow.TFSession.Run (TensorFlow.TFOutput[] inputs, TensorFlow.TFTensor[] inputValues, TensorFlow.TFOutput[] outputs, TensorFlow.TFOperation[] targetOpers, TensorFlow.TFBuffer runMetadata, TensorFlow.TFBuffer runOptions, TensorFlow.TFStatus status) (at <6ed6db22f8874deba74ffe3e566039be>:0)
TensorFlow.TFSession+Runner.Run (TensorFlow.TFStatus status) (at <6ed6db22f8874deba74ffe3e566039be>:0)
TensorflowBrain+<Predictor>c__Iterator0.MoveNext () (at Assets/Scripts/TensorflowBrain.cs:39)
UnityEngine.SetupCoroutine.InvokeMoveNext (System.Collections.IEnumerator enumerator, System.IntPtr returnValueAddress) (at C:/buildslave/unity/build/Runtime/Export/Coroutines.cs:17)
UnityEngine.MonoBehaviour:StartCoroutine(IEnumerator)
TensorflowBrain:Start() (at Assets/Scripts/TensorflowBrain.cs:17)

Keras Model:

model.add(Dense(40, input_dim=10, name="dense_one"))
model.add(Activation('relu'))
model.add(Dropout(dropout))
model.add(Dense(40, name="dense_two"))
model.add(Activation('relu'))
model.add(Dropout(dropout))
model.add(Dense(7, name="final", activation='softmax'))

model.compile(loss='sparse_categorical_crossentropy',
              optimizer=adam,
              metrics=['accuracy'])

C# Code:

using (var graph = new TFGraph())
        {
            tensor = new[] { 140.285470581055f, 41.11237678527829f, 5.836468505859f, 0.0f, 0.0f, 0.0f, 2.463f, 0.0f, 64.28699999999999f, 0.7287418126212261f };
            if (tensor != null)
            {
                
                graph.Import(File.ReadAllBytes("Assets/TensorModel/" + modelFile));
                    var session = new TFSession(graph);
                    var runner = session.GetRunner();
                    runner.AddInput(graph["dense_one/kernel"][0], tensor);                    
                    runner.Fetch(graph["final/Softmax"][0]);

                    var output = runner.Run();

                    TFTensor results = output[0];

                    var re = (float[])results.GetValue(jagged: false);

                    for(int i = 0; i < re.Length; i++)
                    {
                        Debug.Log(re[i]);
                    }
            }
        }

I hope that you can help me. Thanks in advance.

[Poaz]

Anyone?

[alex-zu]

>tensor = new[] { 140.285470581055f, 41.11237678527829f, 5.836468505859f, 0.0f, 0.0f, 0.0f, 2.463f, 0.0f, 64.28699999999999f, 0.7287418126212261f };
>...
>runner.AddInput(graph["dense_one/kernel"][0], tensor); 

Dense layer expect 2d-array, see:

Input shape

nD tensor with shape: (batch_size, ..., input_dim). The most common situation would be a 2D input with shape (batch_size, input_dim).

In your case :
runner.AddInput(graph["dense_one/kernel"][0], new[] {tensor});

And result also will 2d array:
var re = (float[,])results.GetValue(jagged: false);
or
var re = (float[][])results.GetValue(jagged: true);