以上是reduce_sum函数的定义
reduce_sum(
input_tensor,
axis=None,
keep_dims=False,
name=None,
reduction_indices=None
)
说明:
reduce_sum() 就是求和,由于求和的对象是tensor,所以是沿着tensor的某些维度求和。函数名中加了reduce是表示求和后会降维,当然可以通过设置参数来保证不降维,但是默认就是要降维的。
参数解释:
1)input_tensor:输入的张量。
2)axis:沿着哪个维度求和。
对于二维的input_tensor张量,0表示按列求和,1表示按行求和,[0, 1]表示先按列求和再按行求和。
3)keep_dims:默认值为Flase,表示默认要降维。若设为True,则不降维。
4)name:名字。
5)reduction_indices:默认值是None,即把input_tensor降到 0维,也就是一个数。
对于2维input_tensor,reduction_indices=0时,按列;reduction_indices=1时,按行。
注意,reduction_indices与axis不能同时设置。
例子1:
import tensorflow as tf
import numpy as np
x = np.asarray([[1,1,1],[1,1,1]])
x_p = tf.placeholder(tf.int32, [2,3])
y = tf.reduce_sum(x_p, 0)
with tf.Session() as sess:
output = sess.run(y, feed_dict={x_p:x})
print(output)
y = tf.reduce_sum(x_p, reduction_indices=0)
with tf.Session() as sess:
output = sess.run(y, feed_dict={x_p:x})
print(output)
y = tf.reduce_sum(x_p, 1)
with tf.Session() as sess:
output = sess.run(y, feed_dict={x_p:x})
print(output)
y = tf.reduce_sum(x_p, reduction_indices=1)
with tf.Session() as sess:
output = sess.run(y, feed_dict={x_p:x})
print(output)
y = tf.reduce_sum(x_p, 1, keep_dims = True)
with tf.Session() as sess:
output = sess.run(y, feed_dict={x_p:x})
print(output)
y = tf.reduce_sum(x_p, [0,1])
with tf.Session() as sess:
output = sess.run(y, feed_dict={x_p:x})
print(output)
结果:
[2 2 2]
[2 2 2]
[3 3]
[3 3]
[[3]
[3]]
6
例子2:
import tensorflow as tf
import numpy as np
x = np.asarray([[[1,2,3],[4,5,6]],[[7,8,9],[10,11,12]]])
x_p = tf.placeholder(tf.int32,[2,2,3])
y = tf.reduce_sum(x_p,0)
with tf.Session() as sess:
y = sess.run(y,feed_dict={x_p:x})
print(y)
y = tf.reduce_sum(x_p,1)
with tf.Session() as sess:
y = sess.run(y,feed_dict={x_p:x})
print(y)
y = tf.reduce_sum(x_p,2)
with tf.Session() as sess:
y = sess.run(y,feed_dict={x_p:x})
print(y)
结果:
[[8 10 12]
[14 16 18]]
[[5 7 9]
[17 19 21]]
[[6 15]
[24 33]]
分析:
1.png
从运行结果可以看出,
axis = 0时,表示不同二维数组之间相应位置的数字相加
axis = 1时,表示二维数组内部按列相加
axis = 2时,表示二维数组内部按行相加
转自:https://www.jianshu.com/p/2d7db8b9cec9