ValueError: Data cardinality is ambiguous

Question

我正在尝试使用从 DataFrame 获取的数据训练 LSTM 网络。

代码如下：

x_lstm=x.to_numpy().reshape(1,x.shape[0],x.shape[1])

model = keras.models.Sequential([
    keras.layers.LSTM(x.shape[1], return_sequences=True, input_shape=(x_lstm.shape[1],x_lstm.shape[2])),
    keras.layers.LSTM(NORMAL_LAYER_SIZE, return_sequences=True),
    keras.layers.LSTM(NORMAL_LAYER_SIZE),
    keras.layers.Dense(y.shape[1])
])

optimizer=keras.optimizers.Adadelta()

model.compile(loss="mse", optimizer=optimizer)
for i in range(150):
    history = model.fit(x_lstm, y)
    save_model(model,'tmp.rnn')

失败

ValueError: Data cardinality is ambiguous:
  x sizes: 1
  y sizes: 99
Please provide data which shares the same first dimension.

当我将模型更改为

model = keras.models.Sequential([
    keras.layers.LSTM(x.shape[1], return_sequences=True, input_shape=x_lstm.shape),
    keras.layers.LSTM(NORMAL_LAYER_SIZE, return_sequences=True),
    keras.layers.LSTM(NORMAL_LAYER_SIZE),
    keras.layers.Dense(y.shape[1])
])

失败并出现以下错误：

Input 0 of layer lstm_9 is incompatible with the layer: expected ndim=3, found ndim=4. Full shape received: [None, 1, 99, 1200]

如何让它工作？

x 的形状为 (99, 1200)（99 个项目，每个项目有 1200 个特征，这只是一个更大的数据集的样本），y 的形状为 (99, 1)

Answer 1

正如Error所暗示的那样，X和y的First Dimension是不同的。 First Dimension表示Batch Size，应该是一样的。

请确保 Y 也有 shape, (1, something).

我可以使用下面显示的代码重现您的错误：

from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, LSTM
import tensorflow as tf
import numpy as np


# define sequences
sequences = [
    [1, 2, 3, 4],
       [1, 2, 3],
             [1]
    ]

# pad sequence
padded = pad_sequences(sequences)
X = np.expand_dims(padded, axis = 0)
print(X.shape) # (1, 3, 4)

y = np.array([1,0,1])
#y = y.reshape(1,-1)
print(y.shape) # (3,)

model = Sequential()
model.add(LSTM(4, return_sequences=False, input_shape=(None, X.shape[2])))
model.add(Dense(1, activation='sigmoid'))

model.compile (
    loss='mean_squared_error',
    optimizer=tf.keras.optimizers.Adam(0.001))

model.fit(x = X, y = y)

如果我们遵守 Print 声明，

Shape of X is  (1, 3, 4)
Shape of y is (3,)

可以通过取消注释行 y = y.reshape(1,-1) 来修复此错误，这使得 First Dimension (Batch_Size) 等于 (1) 对于 X 和 y.

现在，工作代码如下所示，连同输出：

from tensorflow.keras.preprocessing.sequence import pad_sequences
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, LSTM
import tensorflow as tf
import numpy as np


# define sequences
sequences = [
    [1, 2, 3, 4],
       [1, 2, 3],
             [1]
    ]

# pad sequence
padded = pad_sequences(sequences)
X = np.expand_dims(padded, axis = 0)
print('Shape of X is ', X.shape) # (1, 3, 4)

y = np.array([1,0,1])
y = y.reshape(1,-1)
print('Shape of y is', y.shape) # (1, 3)

model = Sequential()
model.add(LSTM(4, return_sequences=False, input_shape=(None, X.shape[2])))
model.add(Dense(1, activation='sigmoid'))

model.compile (
    loss='mean_squared_error',
    optimizer=tf.keras.optimizers.Adam(0.001))

model.fit(x = X, y = y)

以上代码的输出是：

Shape of X is  (1, 3, 4)
Shape of y is (1, 3)
1/1 [==============================] - 0s 1ms/step - loss: 0.2588
<tensorflow.python.keras.callbacks.History at 0x7f5b0d78f4a8>

希望这对您有所帮助。快乐学习！