如何在 theano/lasagne 中以树的形式遍历模型层？

Question

我需要按树级别遍历模型的层，以便获得祖先（输入）和前驱（输出）层以及同一级别中的相邻层。

这是一个模型作为示例：

import lasagne
def toy_model():
    
    l_input = lasagne.layers.InputLayer(shape=(None, inp_len, n_inputs))

    l_dim_a = lasagne.layers.DimshuffleLayer(l_input, (0, 2, 1))

    l_conv_a = lasagne.layers.Conv1DLayer(
        incoming=l_dim_a, num_filters=16, pad='same',
        filter_size=3, stride=1, nonlinearity=lasagne.nonlinearities.rectify)
    l_conv_a_b = lasagne.layers.batch_norm(l_conv_a)

    l_conv_b = lasagne.layers.Conv1DLayer(
        incoming=l_dim_a, num_filters=16, pad='same',
        filter_size=3, stride=1, nonlinearity=lasagne.nonlinearities.rectify)
    l_conv_b_b = lasagne.layers.batch_norm(l_conv_b)

    l_conv_c = lasagne.layers.Conv1DLayer(
        incoming=l_dim_a, num_filters=16, pad='same',
        filter_size=3, stride=1, nonlinearity=lasagne.nonlinearities.rectify)
    l_conv_c_b = lasagne.layers.batch_norm(l_conv_c)

    l_c_a = lasagne.layers.ConcatLayer([l_conv_a_b, l_conv_b_b, l_conv_c_b], axis=1)
    
    l_dim_b = lasagne.layers.DimshuffleLayer(l_conv_c, (0, 2, 1))
    
    l_c_b = lasagne.layers.ConcatLayer([l_input, l_dim_b], axis=2)

    l_reshape = lasagne.layers.ReshapeLayer(l_c_b, (batch_size* inp_len, n_inputs + (3*3) ))
   
    l_FC = lasagne.layers.DenseLayer(l_reshape, num_units=200, nonlinearity=lasagne.nonlinearities.rectify)


    l_prop = lasagne.layers.DenseLayer(l_FC, num_units=n_classes, nonlinearity=lasagne.nonlinearities.softmax)

    l_output = lasagne.layers.ReshapeLayer(l_prop, (batch_size, inp_len, n_classes))

    return l_input, l_output

提供一个示例（如果有）将会有所帮助。