Question

我有一个实现这样的自动编码器的keras代码：

ENCODING_DIM = 5

# input placeholder
input_img = tf.keras.layers.Input(shape=(320,))

# this is the encoded representation of the input
encoded = tf.keras.layers.Dense(35, activation='relu')(input_img)
encoded = tf.keras.layers.Dense(20, activation='relu')(encoded)
encoded = tf.keras.layers.Dense(ENCODING_DIM, activation='relu')(encoded)

decoded = tf.keras.layers.Dense(20, activation='relu')(encoded)
decoded = tf.keras.layers.Dense(35, activation='relu')(decoded)
decoded = tf.keras.layers.Dense(320, activation='sigmoid')(decoded)

autoencoder = tf.keras.models.Model(input_img, decoded)

encoder = tf.keras.models.Model(input_img, encoded)
encoded_input = tf.keras.layers.Input(shape=(ENCODING_DIM,))

decoder_layer = autoencoder.layers[-1]
#decoded_input = tf.keras.models.Model(encoded_input,decoder_layer(encoded_input))

autoencoder.compile(optimizer='nadam', loss='binary_crossentropy')
from keras.callbacks import ModelCheckpoint

它运行完美。

现在我希望输入尺寸可变（例如，第一个向量[320x1]，第二个[280x1]等）

现在我尝试这样做：

ENCODING_DIM = 5

# input placeholder
input_img = tf.keras.layers.Input(shape=(None,))

# this is the encoded representation of the input
encoded = tf.keras.layers.Dense(35, activation='relu')(input_img)
encoded = tf.keras.layers.Dense(20, activation='relu')(encoded)
encoded = tf.keras.layers.Dense(ENCODING_DIM, activation='relu')(encoded)

decoded = tf.keras.layers.Dense(20, activation='relu')(encoded)
decoded = tf.keras.layers.Dense(35, activation='relu')(decoded)
decoded = tf.keras.layers.Dense(320, activation='sigmoid')(decoded)

autoencoder = tf.keras.models.Model(input_img, decoded)

encoder = tf.keras.models.Model(input_img, encoded)
encoded_input = tf.keras.layers.Input(shape=(ENCODING_DIM,))

decoder_layer = autoencoder.layers[-1]
#decoded_input = tf.keras.models.Model(encoded_input,decoder_layer(encoded_input))

autoencoder.compile(optimizer='nadam', loss='binary_crossentropy')
from keras.callbacks import ModelCheckpoint

但它会返回如下错误：

ValueError                                Traceback (most recent call last)
<ipython-input-24-7764c4707491> in <module>()
     14 
     15 # this is the encoded representation of the input
---> 16 encoded = tf.keras.layers.Dense(35, activation='relu')(input_img)
     17 encoded = tf.keras.layers.Dense(20, activation='relu')(encoded)
     18 encoded = tf.keras.layers.Dense(ENCODING_DIM, activation='relu')(encoded)

2 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/keras/layers/core.py in build(self, input_shape)
    935     input_shape = tensor_shape.TensorShape(input_shape)
    936     if tensor_shape.dimension_value(input_shape[-1]) is None:
--> 937       raise ValueError('The last dimension of the inputs to `Dense` '
    938                        'should be defined. Found `None`.')
    939     last_dim = tensor_shape.dimension_value(input_shape[-1])

ValueError: The last dimension of the inputs to `Dense` should be defined. Found `None`.

如何实现具有不同输入尺寸的自动编码器？

Answer 1

在您的情况下，

密集层将创建35个神经元，每个神经元将连接到每个输入要素（320个中的一个）。例如，它将初始化大小为35x320的权重矩阵。当输入大小未知时，至少在涉及密集层时，无法初始化这种矩阵。您必须将输入填充到最大可能的输入长度（320？），才能在定义模型时应用模型。

具有可变输入的Autoeconders Keras

1 个答案: