python - 检查输入 : expected lstm_28_input to have shape (5739, 8) 时出错，但得到形状为 (1, 8) 的数组

Question

我遇到了 keras 尺寸错误

输入形状是这样的

print(train_X.shape, train_y.shape, test_X.shape, test_y.shape)

结果

(5739, 1, 8) (5739,) (1435, 1, 8) (1435,)

型号如下

batch_size=128
epochs=20
from keras_self_attention import SeqSelfAttention
from keras.layers import Flatten
model = keras.models.Sequential()
model.add(keras.layers.LSTM(epochs, input_shape=(train_X.shape[0], train_X.shape[2]), return_sequences=True))
model.add(Flatten())
model.add(keras.layers.Dense(units=1))
model.compile(loss='mse', optimizer='adam')
model.summary()

结果

_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
lstm_33 (LSTM)               (None, 5739, 20)          2320      
_________________________________________________________________
seq_self_attention_35 (SeqSe (None, 5739, 20)          1345      
_________________________________________________________________
flatten_8 (Flatten)          (None, 114780)            0         
_________________________________________________________________
dense_33 (Dense)             (None, 1)                 114781    
=================================================================
Total params: 118,446
Trainable params: 118,446
Non-trainable params: 0
_________________________________________________________________

但是我在拟合步骤中出错了

history = model.fit(train_X, train_y, epochs=epochs, batch_size=batch_size, validation_data=(test_X, test_y), verbose=2, shuffle=False)

错误

ValueError: Error when checking input: expected lstm_33_input to have shape (5739, 8) but got array with shape (1, 8)

但是我打印的输入形状是(5739,8)，但我无法理解(1,8)来自哪里。以及如何修复它。

input_shape=(train_X.shape[0], train_X.shape[2])
print(input_shape)
(5739, 8)

是 test_X、test_Y 或 train 中输入形状的问题吗？我应该如何解决它？

Answer 1

Keras 中的 LSTM 层需要形状为 (n_timesteps, n_features) 的批量数据。您正在使用错误的尺寸构建图层。

首先，将训练数据 reshape 为 n_data_points、n_timesteps、n_features 形状:

train_X_ = np.swapaxes(train_X, 1, 2)
train_X_.shape # now of shape (5739, 8, 1)

然后指定您的模型的正确尺寸:

model = keras.models.Sequential()
# input shape for the LSTM layer will be (8,1). No need to specify the batch shape.
model.add(keras.layers.LSTM(20, input_shape=(train_X_.shape[1], train_X_.shape[2]), return_sequences=True)) 
model.add(keras.layers.Flatten())
model.add(keras.layers.Dense(1))

model.compile(optimizer='adam', loss='mse')

这将正常工作:

model.fit(train_X_, train_y)