python - Keras 二维输入到二维输出

Question

首先，我阅读了this和 this与我的名字相似的问题，但仍然没有答案。

我想建立一个用于序列预测的前馈网络。 (我意识到 RNN 更适合这个任务，但我有我的理由)。序列的长度为 128，每个元素是一个包含 2 个条目的向量，因此每个批处理的形状应为 (batch_size, 128, 2) 并且目标是序列中的下一步，因此目标张量的形状应为 (batch_size, 1, 2)。

网络架构是这样的:

    model = Sequential()
    model.add(Dense(50, batch_input_shape=(None, 128, 2), kernel_initializer="he_normal" ,activation="relu"))
    model.add(Dense(20, kernel_initializer="he_normal", activation="relu"))
    model.add(Dense(5, kernel_initializer="he_normal", activation="relu"))
    model.add(Dense(2))

但尝试训练时出现以下错误:

ValueError: Error when checking target: expected dense_4 to have shape (128, 2) but got array with shape (1, 2)

我尝试过以下变体:

model.add(Dense(50, input_shape=(128, 2), kernel_initializer="he_normal" ,activation="relu"))

但得到同样的错误。

Answer 1

如果您查看 model.summary() 输出，您会发现问题所在:

Layer (type)                 Output Shape              Param #   
=================================================================
dense_13 (Dense)             (None, 128, 50)           150       
_________________________________________________________________
dense_14 (Dense)             (None, 128, 20)           1020      
_________________________________________________________________
dense_15 (Dense)             (None, 128, 5)            105       
_________________________________________________________________
dense_16 (Dense)             (None, 128, 2)            12        
=================================================================
Total params: 1,287
Trainable params: 1,287
Non-trainable params: 0
_________________________________________________________________

如您所见，模型的输出是 (None, 128,2) 而不是 (None, 1, 2)(或 (无，2)) 如您所料。所以，你可能知道也可能不知道 Dense layer is applied on the last axis of its input array结果，正如您在上面看到的，时间轴和维度一直保留到最后。

如何解决？您提到您不想使用 RNN 层，因此您有两个选择:您需要在模型中的某处使用 Flatten 层，或者您也可以使用一些 Conv1D + Pooling1D 层，甚至是全局池化层。例如(这些只是为了演示，你可以做不同的):

使用Flatten层

model = models.Sequential()
model.add(Dense(50, batch_input_shape=(None, 128, 2), kernel_initializer="he_normal" ,activation="relu"))
model.add(Dense(20, kernel_initializer="he_normal", activation="relu"))
model.add(Dense(5, kernel_initializer="he_normal", activation="relu"))
model.add(Flatten())
model.add(Dense(2))

model.summary()

模型总结:

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_17 (Dense)             (None, 128, 50)           150       
_________________________________________________________________
dense_18 (Dense)             (None, 128, 20)           1020      
_________________________________________________________________
dense_19 (Dense)             (None, 128, 5)            105       
_________________________________________________________________
flatten_1 (Flatten)          (None, 640)               0         
_________________________________________________________________
dense_20 (Dense)             (None, 2)                 1282      
=================================================================
Total params: 2,557
Trainable params: 2,557
Non-trainable params: 0
_________________________________________________________________

使用 GlobalAveragePooling1D 层

model = models.Sequential()
model.add(Dense(50, batch_input_shape=(None, 128, 2), kernel_initializer="he_normal" ,activation="relu"))
model.add(Dense(20, kernel_initializer="he_normal", activation="relu"))
model.add(GlobalAveragePooling1D())
model.add(Dense(5, kernel_initializer="he_normal", activation="relu"))
model.add(Dense(2))

model.summary()

模型总结:

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
dense_21 (Dense)             (None, 128, 50)           150       
_________________________________________________________________
dense_22 (Dense)             (None, 128, 20)           1020      
_________________________________________________________________
global_average_pooling1d_2 ( (None, 20)                0         
_________________________________________________________________
dense_23 (Dense)             (None, 5)                 105       
_________________________________________________________________
dense_24 (Dense)             (None, 2)                 12        
=================================================================
Total params: 1,287
Trainable params: 1,287
Non-trainable params: 0
_________________________________________________________________

请注意，在上述两种情况下，您都需要将标签(即目标)数组 reshape 为 (n_samples, 2)(或者您可能想要使用 Reshape 层最后)。