[人工智能]tensorflow函数分析：gather和batch_gather

gather函数

gather(tensor_a,tensor_b)，根据tensor_b调整保留第2个纬度的顺序，如果输入2维或多维tensor_b会返回多次调整结果，值调整第2维不调整其他纬度，例：

  1 import tensorflow as tf
  2 tensor_a = tf.Variable([[[1,2],[3,4],[5,6]],[[4,5],[6,7],[8,9]]])
  3 tensor_b = tf.Variable([[1,0],[0,1]],dtype=tf.int32)
  4 tensor_c = tf.Variable([0,0],dtype=tf.int32)
  5 with tf.Session() as sess:
  6     sess.run(tf.global_variables_initializer())
  7     print('*************')
  8     print(sess.run(tf.gather(tensor_a,tensor_b)))
  9     print('*************')
 10     print(sess.run(tf.gather(tensor_a,tensor_c)))

*************
[[[[4 5]
   [6 7]
   [8 9]]

  [[1 2]
   [3 4]
   [5 6]]]


 [[[1 2]
   [3 4]
   [5 6]]

  [[4 5]
   [6 7]
   [8 9]]]]
*************
[[[1 2]
  [3 4]
  [5 6]]

 [[1 2]
  [3 4]
  [5 6]]]

batch_gather函数

batch_gather(tensor_a,tensor_b)调整多个纬度，从第2个纬度开始，tensor_b有几个纬度就可以调整几个（当然tensor_a纬度必须足够多），例子：

  1 import tensorflow as tf
  2 tensor_a = tf.Variable([[[1,2],[3,4],[5,6]],[[4,5],[6,7],[8,9]]])
  3 tensor_b = tf.Variable([[1,0,2],[2,1,0]],dtype=tf.int32)
  4 tensor_c = tf.Variable([0,0],dtype=tf.int32)
  5 with tf.Session() as sess:
  6     sess.run(tf.global_variables_initializer())
  7     print('*************')
  8     print(sess.run(tf.batch_gather(tensor_a,tensor_b)))
  9     print('*************')
 10     print(sess.run(tf.gather(tensor_a,tensor_c)))

*************
[[[3 4]
  [1 2]
  [5 6]]

 [[8 9]
  [6 7]
  [4 5]]]
*************
[[[1 2]
  [3 4]
  [5 6]]

 [[1 2]
  [3 4]
  [5 6]]]