Shengxin experiment record (part2)--tf.reduce_sum() usage introduction

学习笔记,仅供参考,有错必纠
参考自：tf.reduce_sum()用法介绍

前言

I came across this function while reproducing…emmm… Speaking of which, I really know very little about deep learning.

方法定义

tf.reduce_sum()The role of is to calculate the sum of the elements in the tensor in a certain way.

tf.reduce_sum(
    input_tensor, 
    axis=None, 
    keepdims=None,
    name=None)

参数解释：

input_tensoris the tensor to be processed.
axisSpecifies which dimension to sum by,All elements are added by default.
keepdimsIndicates that the dimensions of the original tensor are not maintained,默认为False,若为TrueThe original tensor dimensions are maintained, .
nameUsed to define the operation name.

示例

有形如(2,3,4)的三维张量：

import tensorflow as tf
a = tf.constant([[[1, 2, 3, 4], 
                  [5, 6, 7, 8], 
                  [9, 10, 9, 8]], 
                 [[7, 6, 5, 4], 
                  [3, 2, 1, 0], 
                  [1, 2, 3, 4]]])
print(a.shape)
# (2, 3, 4)

分别设置axis为0,1,2The result of performing the addition is as follows：

b = tf.reduce_sum(a, axis = 0)
c = tf.reduce_sum(a, axis = 1)
d = tf.reduce_sum(a, axis = 2)

sess = tf.InteractiveSession()

b
b.eval()

<tf.Tensor 'Sum_18:0' shape=(3, 4) dtype=int32>
array([[ 8,  8,  8,  8],
       [ 8,  8,  8,  8],
       [10, 12, 12, 12]])

c
c.eval()

<tf.Tensor 'Sum_19:0' shape=(2, 4) dtype=int32>
array([[15, 18, 19, 20],
       [11, 10,  9,  8]])

d
d.eval()

<tf.Tensor 'Sum_20:0' shape=(2, 3) dtype=int32>
array([[10, 26, 36],
       [22,  6, 10]])

可以看到,The original tensor is three-dimensional,指定axis进行一次tf.reduce_sum()Then it becomes a two-dimensional tensor. 如果设置keepdims=True,则结果如下,It can be seen that the tensor after calculating the sum is still three-dimensional,其中指定axisThe number of elements in that dimension is 1,The other dimensions remain unchanged.

b = tf.reduce_sum(a, axis = 0, keepdims=True)
c = tf.reduce_sum(a, axis = 1, keepdims=True)
d = tf.reduce_sum(a, axis = 2, keepdims=True)

b
b.eval()

<tf.Tensor 'Sum_24:0' shape=(1, 3, 4) dtype=int32>
array([[[ 8,  8,  8,  8],
        [ 8,  8,  8,  8],
        [10, 12, 12, 12]]])

c
c.eval()

<tf.Tensor 'Sum_25:0' shape=(2, 1, 4) dtype=int32>
array([[[15, 18, 19, 20]],

       [[11, 10,  9,  8]]])

d
d.eval()

<tf.Tensor 'Sum_26:0' shape=(2, 3, 1) dtype=int32>
array([[[10],
        [26],
        [36]],

       [[22],
        [ 6],
        [10]]])

sess.close()