ValueError: Input 0 of layer lstm is incompatible with the layer: expected ndim=3, found ndim=4. Full shape received: (None, 32, 24, 7)

Question

我还在想办法解决这个错误... 因为我必须将第一层输入形状固定为 input_shape=(BATCH_SIZE, N_PAST, N_FEATURES) 我收到关于 LSTM 和 GRU

的任何错误的错误

    model = tf.keras.models.Sequential([
        tf.keras.layers.LSTM(64, return_sequences=True, input_shape=(BATCH_SIZE,N_PAST, N_FEATURES)),
        tf.keras.layers.Dense(N_FEATURES)
    ])

    model.summary()

    optimizer =  tf.keras.optimizers.SGD(lr=1e-8, momentum=0.9)
    model.compile(
        loss="mse",
        optimizer=optimizer,
        metrics=["mae"]
    )
    model.fit(
        train_set, validation_data=valid_set,validation_steps=100, epochs=100
    )

Answer 1

永远不需要为模型提供 batch_size 维度的固定值，tensorflow 将根据给定的数据形状动态处理此问题。

所以在模型的构建中：

tf.keras.layers.Dense(7, input_shape=(N_PAST, N_FEATURES), activation='relu')

执行 summary() 时，该层的输入形状应为 (None、N_PAST、N_FEATURES)

Answer 2

您不需要添加 BATCH_SIZE:

input_shape=(N_PAST, N_FEATURES)

ValueError: Input 0 of layer lstm is incompatible with the layer: expected ndim=3, found ndim=4. Full shape received: (None, 32, 24, 7)

ValueError: Input 0 of layer lstm is incompatible with the layer: expected ndim=3, found ndim=4. Full shape received: (None, 32, 24, 7)

neural-network

conv-neural-network

lstm

keras

tensorflow