tensorflow 封装的函数tf.one_hot():
import tensorflow as tf
import numpy as np
z=np.random.randint(0,10,size=[10])
y=tf.one_hot(z,10,on_value=1,off_value=None,axis=0)
with tf.Session()as sess:
print(z)
print(sess.run(y))
[5 7 7 0 5 5 2 0 0 0]
[[0 0 0 1 0 0 0 1 1 1]
[0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 1 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[1 0 0 0 1 1 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[0 1 1 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 0 0]]
瞅上去很完美,但是,tf.one_hot()会让你的graph动态增添节点,可能导致的后果就是内存泄露,模型越来越慢.....
所以当你发现模型占用内存越来越大,然后执行以下操作tf.get_default_graph().finalize() 报错的时候
就应该考虑自己实现one-hot()功能了。
def one_hot(labels,Label_class):
one_hot_label = np.array([[int(i == int(labels[j])) for i in range(Label_class)] for j in range(len(labels))])
return one_hot_label
举例:
import numpy as np
def one_hot(labels,Label_class):
one_hot_label = np.array([[int(i == int(labels[j])) for i in range(Label_class)] for j in range(len(labels))])
return one_hot_label
y = [2,5,6,7,8]
Label_class = 20
print one_hot(y,Label_class)
[[0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 0]
[0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0]]
https://blog.csdn.net/shanyicheng1111/article/details/80007193