python - 对象没有属性 'inbound_nodes'

Question

我正在尝试定义一个模型并对其进行编译，但由于某种原因我无法编译或定义该模型...

def fws():
    filter_size = 8
    pooling_size = 6
    stride_step = 2
    J = 80
    splits = 33
    total_frames_with_deltas = 45
    pool_splits = ((splits - pooling_size)+1)/2
    print "pool_splits" + str(pool_splits)
    print "Printing shapes"

    list_of_input = [Input(shape=(8,3)) for i in range(splits*total_frames_with_deltas)]
    output_convolution = []

    for steps in range(total_frames_with_deltas):
        conv = Conv1D(filters = J, kernel_size = 8)
        column = 0
        skip = 45
        conv_output = []
        for _ in range(splits):
            conv_output.append(conv(list_of_input[(column*skip)+steps]))
            column = column + 1
        output_convolution.append((conv_output))

    print len(output_convolution)
    print len(output_convolution[0])

    out = 0
    output_conv = []

    for row in range(splits):
        for column in range(total_frames_with_deltas):
            #print row
            #print column
            out = out + output_convolution[column][row]
        output_conv.append(out)

    output_con = Concatenate()(output_conv)
    output_con = Reshape((splits,-1))(output_con)

    pooled = MaxPooling1D(pool_size = pooling_size, strides = stride_step)(output_con)
    print pooled.shape
    #reshape = Reshape((3,-1))(pooled)

    #fc
    dense1 = Dense(units = 1000, activation = 'relu',    name = "dense_1")(pooled)
    dense2 = Dense(units = 1000, activation = 'relu',    name = "dense_2")(dense1)
    dense3 = Dense(units = 50 , activation = 'softmax', name = "dense_3")(dense2)
    raw_input("Model definition ok!")

    model = Model(inputs = list_of_input , outputs = dense3)
    raw_input("Model definition with input/output")

    model.compile(loss="categorical_crossentropy", optimizer='sgd' , metrics = [metrics.categorical_accuracy])

这是完整的错误信息:

  File "keras_cnn_phoneme_original_fit_generator.py", line 231, in <module>
    fws()
  File "keras_cnn_phoneme_original_fit_generator.py", line 212, in fws
    model = Model(inputs = list_of_input , outputs = dense3)
  File "/usr/local/lib/python2.7/dist-packages/keras/legacy/interfaces.py", line 88, in wrapper
    return func(*args, **kwargs)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 1676, in __init__
    build_map_of_graph(x, finished_nodes, nodes_in_progress)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 1666, in build_map_of_graph
    layer, node_index, tensor_index)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 1666, in build_map_of_graph
    layer, node_index, tensor_index)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 1666, in build_map_of_graph
    layer, node_index, tensor_index)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 1666, in build_map_of_graph
    layer, node_index, tensor_index)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 1666, in build_map_of_graph
    layer, node_index, tensor_index)
  File "/usr/local/lib/python2.7/dist-packages/keras/engine/topology.py", line 1664, in build_map_of_graph
    next_node = layer.inbound_nodes[node_index]
AttributeError: 'NoneType' object has no attribute 'inbound_nodes'

错误似乎是在定义网络的输入和输出时发生的。我不确定为什么。卷积和池化都是为了处理输入而设计的。所以这个错误对我来说毫无意义？

Answer 1

有点晚了，但我刚刚遇到了类似的问题，我想其他人可能也有类似的问题。我认为你弄错的是:

out = out + output_convolution[column][row]

尝试将其更改为:

out = add([out, output_convolution[column][row]]))

add 在 keras.layers.merge 中。与 tensorflow 不同，keras 似乎无法将 a+b 解释为图中的节点，因此它会刹车。

另外，为了将来引用，我尝试做的是减去两个张量 (a - b)，如下所示:

subt = add([a, -b])

这引发了同样的异常。我的做法是将 b 定义为 -b，虽然不花哨，但确实有效。