ValueError: Input 0 of layer sequential_33 is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: [64, 100]

Question

我正在关注此 guide 以学习构建一个简单的 RNN。与指南不同，我只希望我的模型按升序预测下一个整数（例如 x = [1,2,3] y = [2,3,4]）但是在尝试训练我的模型时，我收到此错误消息：

ValueError: Input 0 of layer sequential_33 is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: [64, 100]

就像指南中一样，我的数据集的形状是：

<BatchDataset shapes: ((64, 100), (64, 100)), types: (tf.int64, tf.int64)>

与指南略有不同，我的模型定义为

BATCH_SIZE = 64
n_neurons  = 101
model = Sequential()
# shape [batch_size, timesteps, features]
model.add(Input(batch_input_shape = (BATCH_SIZE,100,1)))
model.add(LSTM(n_neurons ,return_sequences=True, stateful=True))
model.add(Dense(1))
model.compile(loss='mean_squared_error', optimizer='adam', metrics=['accuracy'])
print(model.summary())

摘要蜂鸣：

Model: "sequential_37"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
lstm_40 (LSTM)               (64, 100, 101)            41612     
_________________________________________________________________
dense_28 (Dense)             (64, 100, 1)              102       
=================================================================
Total params: 41,714
Trainable params: 41,714
Non-trainable params: 0
_________________________________________________________________
None

你能帮我理解为什么会出现这个错误，以及如何解决它吗？

我已确保数据集与指南中的维度相同，并为输入层提供了“batch_input_shape= (BATCH_SIZE,100,1)”，因为我了解到LSTM 至少需要具有形状 [batch_size、时间步长、特征] 的 3D 数据。所以我很困惑我仍然不正确。

如有任何帮助，我们将不胜感激！

Answer 1

您应该向模型提供形状 (64, 100, 1) 而不是 (64, 100)。只需为您的数据添加一个维度

ValueError: Input 0 of layer sequential_33 is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: [64, 100]

ValueError: Input 0 of layer sequential_33 is incompatible with the layer: expected ndim=3, found ndim=2. Full shape received: [64, 100]

python

lstm

keras

tensorflow