CNTK Python API：加载模型后访问图层

Question

加载模型后我无法访问图层。

我创建的模型如下：

def create_model(vocab_dim, hidden_dim):

    input_seq_axis1 = Axis('inputAxis1')

    input_sequence_before = sequence.input_variable(shape=vocab_dim, sequence_axis=input_seq_axis1, is_sparse = use_sparse)

    input_sequence_after = sequence.input_variable(shape=vocab_dim, sequence_axis=input_seq_axis1, is_sparse = use_sparse)

    e=Sequential([

        C.layers.Embedding(hidden_dim),

        Stabilizer()

        ],name='Embedding')

    a = Sequential([

        e,  

        C.layers.Recurrence(C.layers.LSTM(hidden_dim//2),name='ForwardRecurrence'),

        ],name='ForwardLayer')

    b = Sequential([

        e,  

        C.layers.Recurrence(C.layers.LSTM(hidden_dim//2),go_backwards=True),

       ],name='BackwardLayer')

    latent_vector = C.splice(a(input_sequence_before), b(input_sequence_after))

    bias = C.layers.Parameter(shape = (vocab_dim, 1), init = 0, name='Bias')

    weights = C.layers.Parameter(shape = (vocab_dim, hidden_dim), init = C.initializer.glorot_uniform(), name='Weights')

    z = C.times_transpose(weights, latent_vector,name='Transpose') + bias

    z = C.reshape(z, shape = (vocab_dim))

    return z

然后我加载模型：

def load_my_model(vocab_dim, hidden_dim):

    z=load_model("models/lm_epoch0.dnn")

    input_sequence_before = z.arguments[0]

    input_sequence_after = z.arguments[1]

    a=z.ForwardLayer

    b=z.BackwardLayer

    latent_vector = C.splice(a(input_sequence_before), b(input_sequence_after))

我收到一个错误：TypeError("argument ForwardRecurrence 的类型 SequenceOver[inputAxis1][Tensor[100]] 与传递的变量的类型 SequenceOver[inputAxis1][SparseTensor[50000]] 不兼容",)

看起来名称引用的层 (z.ForwardLayer) 表示来自层立即输入的函数。如何计算“latent_vector”（我需要这个变量来创建交叉熵和损失函数以继续训练）？

Answer 1

根据错误，与 ForwardLayer 的预期 (100) 相比，您的输入 seq 的尺寸太大 (5000)。当您通过 选择节点 ForwardLayer 时z.ForwardLayer，您只能选择那个非常特定的节点/层，而不是与其连接的计算图的层/节点/其余部分。你应该这样做a = C.combine([z.ForwardLayer.owner])，你应该没事。