Keras 2D输入到2D输出

Question

首先，我读过与我相似的this和this问题，但仍然没有答案。

我想建立一个前馈网络进行序列预测。(我意识到RNN更适合这项任务，但我有自己的理由)。序列长度为128，每个元素是一个有2个条目的向量，因此每一批应该是(batch_size, 128, 2)形状，目标是序列中的下一步，因此目标张量应该是形状(batch_size, 1, 2)。

网络体系结构如下所示：

    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))

但是，试图训练，我得到了以下错误：

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

我试过一些变体，比如：

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

但也有同样的错误。

Answer 1

如果您查看model.summary()输出，您将看到问题所在：

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)，而不是您预期的(None, 1, 2) (或(None, 2))。因此，您可能知道也可能不知道Dense layer is applied on the last axis of its input array，因此，正如您在上面看到的，时间轴和维度一直保存到最后。

如何解决这个问题？您提到不想使用RNN层，因此您有两个选项:要么在模型中的某个地方使用Flatten层，要么可以使用一些Conv1D + Pooling1D层，甚至一个GlobalPooling层。例如(这些只是为了演示，您可能会采取不同的做法)：

使用Flatten layer的

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()

示范摘要：

_________________________________________________________________
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 layer的

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()

​模型摘要：

_________________________________________________________________
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)数组重组为(n_samples, 2) (或者您可能希望在最后使用Reshape层)。