我想用Keras创建一个包含两个LSTM层的模型。然而，以下代码却产生了错误：
from keras.models import Sequentialfrom keras.layers import LSTM, Dropout, Activationfrom keras.callbacks import ModelCheckpointfrom keras.utils import to_categoricalmodel = Sequential()model.add(LSTM(5, activation="softmax"))model.add(LSTM(5, activation="softmax"))model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['categorical_accuracy'])# 这些值是要预测的。directions = [-2, -1, 0, 1, 2]# 样本数据。我们有三个时间步，每个时间步一个特征，以及一个结果值。data = [[[[1], [2], [3]], -1], [[[3], [2], [1]], 2], [[[4], [5], [7]], 1], [[[1], [-1], [10]], -2]]X = []y_ = []# 现在我们从上面的数据中取10000个样本。for i in np.random.choice(len(data), 10000): X.append(data[i][0]) y_.append(data[i][1])X = np.array(X)y_ = np.array(y_)y = to_categorical(y_ + 2, num_classes=5)model.fit(X, y, epochs=3, validation_data=(X, y))print(model.summary())loss, acc = model.evaluate(X, y)print("Loss: {:.2f}".format(loss))print("Accuracy: {:.2f}%".format(acc*100))我得到的错误如下：
ValueError: Input 0 is incompatible with layer lstm_10: expected ndim=3, found ndim=2完整的错误追溯如下：
---------------------------------------------------------------------------ValueError Traceback (most recent call last)<ipython-input-35-58fa9218c3f3> in <module> 31 model.fit(X, y, 32 epochs=3,---> 33 validation_data=(X, y)) 34 print(model.summary()) 35 C:\Anaconda3\lib\site-packages\keras\engine\training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs) 950 sample_weight=sample_weight, 951 class_weight=class_weight,--> 952 batch_size=batch_size) 953 # Prepare validation data. 954 do_validation = FalseC:\Anaconda3\lib\site-packages\keras\engine\training.py in _standardize_user_data(self, x, y, sample_weight, class_weight, check_array_lengths, batch_size) 675 # to match the value shapes. 676 if not self.inputs:--> 677 self._set_inputs(x) 678 679 if y is not None:C:\Anaconda3\lib\site-packages\keras\engine\training.py in _set_inputs(self, inputs, outputs, training) 587 assert len(inputs) == 1 588 inputs = inputs[0]--> 589 self.build(input_shape=(None,) + inputs.shape[1:]) 590 return 591 C:\Anaconda3\lib\site-packages\keras\engine\sequential.py in build(self, input_shape) 219 self.inputs = [x] 220 for layer in self._layers:--> 221 x = layer(x) 222 self.outputs = [x] 223 self._build_input_shape = input_shapeC:\Anaconda3\lib\site-packages\keras\layers\recurrent.py in __call__(self, inputs, initial_state, constants, **kwargs) 530 531 if initial_state is None and constants is None:--> 532 return super(RNN, self).__call__(inputs, **kwargs) 533 534 # If any of `initial_state` or `constants` are specified and are KerasC:\Anaconda3\lib\site-packages\keras\engine\base_layer.py in __call__(self, inputs, **kwargs) 412 # Raise exceptions in case the input is not compatible 413 # with the input_spec specified in the layer constructor.--> 414 self.assert_input_compatibility(inputs) 415 416 # Collect input shapes to build layer.C:\Anaconda3\lib\site-packages\keras\engine\base_layer.py in assert_input_compatibility(self, inputs) 309 self.name + ': expected ndim=' + 310 str(spec.ndim) + ', found ndim=' +--> 311 str(K.ndim(x))) 312 if spec.max_ndim is not None: 313 ndim = K.ndim(x)ValueError: Input 0 is incompatible with layer lstm_10: expected ndim=3, found ndim=2看起来第一个LSTM层的输出维度（应该是dim=2）与第二个LSTM层所需的输入维度（批次、时间步、特征的dim=3）不匹配。
让我困惑的是，像我这样添加LSTM层的方式在这里似乎是有效的：例如： https://adventuresinmachinelearning.com/keras-lstm-tutorial/
当我移除第二个LSTM层时，模型是可以工作的。
回答：
默认情况下，LSTM只在序列的最后一个元素后返回其最终输出。如果你想将两个LSTM层连接起来，那么你需要将第一个LSTM层在序列的每个元素后的输出传递给第二个LSTM层。例如：
model = Sequential()model.add(LSTM(5, return_sequences=True))model.add(LSTM(5, activation="softmax"))有关return_sequence如何工作的详细信息，请参阅文档 https://keras.io/layers/recurrent/
