tf.segment_sum和tf.unsorted_segment_sum理解实例

本文来自 guotong1988 的CSDN 博客 ，全文地址请点击：https://blog.csdn.net/guotong1988/article/details/77622790

 import tensorflow as tf
 c = tf.constant([[1,2,3,4], [-1,-2,-3,-4], [5,6,7,8]])
 result = tf.segment_sum(c, tf.constant([0, 0, 1]))#第二个参数长度必须为3
 result_ = tf.segment_sum(c, tf.constant([0, 1, 1]))
 result__ = tf.segment_sum(c, tf.constant([0, 1, 2]))
 result2 = tf.unsorted_segment_sum(c, tf.constant([2, 1, 1]),3)#第二个参数长度必须为3
 result3 = tf.unsorted_segment_sum(c, tf.constant([1, 0, 1]),2)
 #result4 = tf.unsorted_segment_sum(c, tf.constant([2, 0, 1]),2) #错误，segment_ids[0] = 2 is out of range [0, 2)
 result4 = tf.unsorted_segment_sum(c, tf.constant([2, 0, 1]),3)
 result5 = tf.unsorted_segment_sum(c, tf.constant([3, 1, 0]),5)
 sess = tf.Session()
 print("result")
 print(sess.run(result))
 print("result_")
 print(sess.run(result_))
 print("result__")
 print(sess.run(result__))
 print("result2")
 print(sess.run(result2))
 print("result3")
 print(sess.run(result3))
 print("result4")
 print(sess.run(result4))
 print("result5")
 print(sess.run(result5))

运行结果：

result
[[0 0 0 0]
 [5 6 7 8]]
result_
[[1 2 3 4]
 [4 4 4 4]]
result__
[[ 1  2  3  4]
 [-1 -2 -3 -4]
 [ 5  6  7  8]]
result2
[[0 0 0 0]
 [4 4 4 4]
 [1 2 3 4]]
result3
[[-1 -2 -3 -4]
 [ 6  8 10 12]]
result4
[[-1 -2 -3 -4]
 [ 5  6  7  8]
 [ 1  2  3  4]]
result5
[[ 5  6  7  8]
 [-1 -2 -3 -4]
 [ 0  0  0  0]
 [ 1  2  3  4]
 [ 0  0  0  0]]

将索引值相同的进行求和，其余的按顺序计算。