self._make_predict_function()
  File "/usr/local/lib/python3.4/dist-packages/keras/engine/training.py", line 679, in _make_predict_function
    **self._function_kwargs)
  File "/usr/local/lib/python3.4/dist-packages/keras/backend/tensorflow_backend.py", line 615, in function
    return Function(inputs, outputs, updates=updates)
  File "/usr/local/lib/python3.4/dist-packages/keras/backend/tensorflow_backend.py", line 589, in __init__
    with tf.control_dependencies(self.outputs):
  File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/ops.py", line 3192, in control_dependencies
    return get_default_graph().control_dependencies(control_inputs)
  File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/ops.py", line 2993, in control_dependencies
    c = self.as_graph_element(c)
  File "/usr/local/lib/python3.4/dist-packages/tensorflow/python/framework/ops.py", line 2291, in as_graph_element
    raise ValueError("Tensor %s is not an element of this graph." % obj)
ValueError: Tensor Tensor("Sigmoid_2:0", shape=(?, 17), dtype=float32) is not an element of this graph.

    seq1=Input(dtype='int32',shape=(400,),name='input_text')
    seq2=Input(dtype='int32',shape=(20,),name='input_titles')

    embeddeding=Embedding(max_features,embedding_dims,dropout=0.3)

    encoding_1=embeddeding(seq1)
    encoding_2=embeddeding(seq2)

    filter_lengths = [1,3,6]


    def max_1d(X):
        return K.max(X, axis=1)
    convs1=[]
    convs2=[]
    for fl in filter_lengths:

        conv1=Convolution1D(nb_filter=nb_filter,
                        filter_length=fl,
                        border_mode='valid',
                        activation='relu',
                        subsample_length=1)(encoding_1)
        conv1=Lambda(max_1d, output_shape=(nb_filter,))(conv1)
        convs1.append(conv1)

        conv2=Convolution1D(nb_filter=nb_filter,
                        filter_length=fl,
                        border_mode='valid',
                        activation='relu',
                        subsample_length=1)(encoding_2)
        conv2=Lambda(max_1d, output_shape=(nb_filter,))(conv2)
        convs2.append(conv2)

    m=merge([*convs1,*convs2],mode='concat')
    m=Highway(activation='relu')(m)
    m=Highway(activation='relu')(m)
    m=Dropout(0.5)(m)
    hovedkategori_loss=Dense(labsHovedKat.shape[1],activation='sigmoid',name='hovedkategori')(m)

    m1=merge([hovedkategori_loss,m],mode='concat')
    underkategori_loss=Dense(labsUnderKat.shape[1],activation='sigmoid',name='underkategori')(m1)

    model=Model(input=[seq1,seq2],output=[hovedkategori_loss,underkategori_loss])
    model.compile(optimizer='adam',loss='binary_crossentropy',metrics={'hovedkategori':'accuracy','underkategori':'accuracy'})