embedding_lookup()的用法

这个函数真的很常用，尤其word2vec

tf.nn.embedding_lookup()就是根据input_ids中的id，寻找embeddings中的第id行。比如input_ids=[1,3,5]，则找出embeddings中第1，3，5行，组成一个tensor返回。

实例 1

import tensorflow as tf
import numpy as np

input_ids = tf.placeholder(tf.int32, shape=[None], name="input_ids")
embedding = tf.Variable(np.identity(5, dtype=np.int32))
input_embedding = tf.nn.embedding_lookup(embedding, input_ids)

sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())
print("embedding=\n", embedding.eval())
print("input_embedding=\n", sess.run(input_embedding, feed_dict={input_ids: [1, 2, 3, 0, 3, 2, 1]}))

结果

embedding=
 [[1 0 0 0 0]
 [0 1 0 0 0]
 [0 0 1 0 0]
 [0 0 0 1 0]
 [0 0 0 0 1]]
input_embedding=
 [[0 1 0 0 0]
 [0 0 1 0 0]
 [0 0 0 1 0]
 [1 0 0 0 0]
 [0 0 0 1 0]
 [0 0 1 0 0]
 [0 1 0 0 0]]
[Finished in 3.8s]

实例2

import tensorflow as tf
import numpy as np

input_ids = tf.placeholder(dtype=tf.int32, shape=[3, 2])
embedding = tf.Variable(np.identity(5, dtype=np.int32))
input_embedding = tf.nn.embedding_lookup(embedding, input_ids)
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())

print("embedding=\n", embedding.eval())
print("input_embedding=\n", sess.run(input_embedding, feed_dict={input_ids: [[1, 2], [2, 1], [3, 3]]}))

结果

embedding=
 [[1 0 0 0 0]
 [0 1 0 0 0]
 [0 0 1 0 0]
 [0 0 0 1 0]
 [0 0 0 0 1]]
input_embedding=
 [[[0 1 0 0 0]
  [0 0 1 0 0]]

 [[0 0 1 0 0]
  [0 1 0 0 0]]

 [[0 0 0 1 0]
  [0 0 0 1 0]]]
[Finished in 4.0s]

来自：https://blog.csdn.net/u013041398/article/details/60955847
https://blog.csdn.net/laolu1573/article/details/77170407