tf.nn.embedding_lookup函数的用法主要是选取一个张量里面索引对应的元素。tf.nn.embedding_lookup(tensor, id):tensor就是输入张量,id就是张量对应的索引,其他的参数不介绍。id可以是一个数字,或者一维数组,也可以是矩阵
例如:import tensorflow as tf;
import numpy as np;
c = np.random.random([10,1])
b = tf.nn.embedding_lookup(c,[[1, 3, 6],
[2, 4, 7]])
with tf.Session() as sess:
sess.run(tf.initialize_all_variables())
print(sess.run(b))
print (c)
c:[[[0.01653263]
[0.23159859]
[0.47069613]]
[[0.53892802]
[0.84475463]
[0.26659862]]]
b:[[0.1707615 ]
[0.01653263]
[0.53892802]
[0.23159859]
[0.84475463]
[0.71732015]
[0.47069613]
[0.26659862]
[0.59343494]
[0.83624844]]