嵌入式linux CNN,CNN嵌入式层的实现

嗨，我一直在尝试遵循以下两个示例：

但是，对于示例1，我一直遇到以下错误

Input 0 of layer conv2d_1 is incompatible with the layer: expected ndim=4, found ndim=3. Full shape received: [None, 100, 20]

以及以下错误示例2。

ValueError: Layer conv2d_1 expects 1 inputs, but it received 100 input tensors. Inputs received: [, , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , , ]

我已经更改了代码，而不是使用word_list = tf.unstack(word_vectors, axis=1)，我已更改为以下实现来解决错误，但是我不知道实现是否正确。

word_vectors = tf.contrib.layers.embed_sequence(

x, vocab_size=n_words, embed_dim=EMBEDDING_SIZE)

word_list = tf.reshape( word_vectors, [-1, MAX_DOCUMENT_LENGTH, EMBEDDING_SIZE, 1])

conv1, pool1 = cnn(word_list, FILTER_SHAPE1, 'CNN_Layer1')

函数cnn和其他参数已定义

MAX_DOCUMENT_LENGTH = 100

N_FILTERS = 10

FILTER_SHAPE1 = [20, 20]

POOLING_WINDOW = 4

POOLING_STRIDE = 2

MAX_LABEL = 15

EMBEDDING_SIZE = 20

def cnn(x, filter_shape, name):

with tf.variable_scope(name):

conv1 = tf.layers.conv2d(

x,

filters=N_FILTERS,

kernel_size=filter_shape,

padding='VALID',

activation=tf.nn.relu)

pool1 = tf.layers.max_pooling2d(

conv1,

pool_size=POOLING_WINDOW,

strides=POOLING_STRIDE,

padding='SAME')

return conv1, pool1