TensorFlow学习－－tf.get_variable使用

tensorflow中tf.get_variable的API为

def get_variable(name,
                 shape=None,
                 dtype=None,
                 initializer=None,
                 regularizer=None,
                 trainable=True,
                 collections=None,
                 caching_device=None,
                 partitioner=None,
                 validate_shape=True,
                 use_resource=None,
                 custom_getter=None):

常用的参数有：
name：变量名称
shape：变量维度
initializer：变量初始化方式
regularizer：正规化
caching_device：可选的设备字符串或函数描述

其中，变量的初始化方式有

tf.constant_initializer–常量初始化

#!/usr/bin/python
# coding:utf-8

import tensorflow as tf
# 默认值为0
v1 = tf.get_variable('v1', shape=[5], initializer=tf.constant_initializer())
# 也可以指定初始化值
v2 = tf.get_variable('v2', shape=[5], initializer=tf.constant_initializer(9.))

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print v1.eval()
    print v2.eval()

输出：

[ 0.  0.  0.  0.  0.]
[ 9.  9.  9.  9.  9.]

tf.random_normal_initializer–正太分布初始化

#!/usr/bin/python
# coding:utf-8

import tensorflow as tf
# 随机正太分布初始化
v1 = tf.get_variable('v1', shape=[5], initializer=tf.random_normal_initializer())
# 指定正太分布的均值方差种子值
v2 = tf.get_variable('v2', shape=[5], initializer=tf.random_normal_initializer(mean=0, stddev=1., seed=1))
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print v1.eval()
    print v2.eval()

输出：

[-0.96925646 -1.34411633  1.27495158  0.10656819 -0.01502592]
[-0.81131822  1.48459876  0.06532937 -2.4427042   0.0992484 ]

tf.truncated_normal_initializer–截取正太分布

#!/usr/bin/python
# coding:utf-8

import tensorflow as tf
# 随机截取正太分布
v1 = tf.get_variable('v1', shape=[5], initializer=tf.truncated_normal_initializer())
# 指定截取正太分布的均值方差种子值
v2 = tf.get_variable('v2', shape=[5], initializer=tf.truncated_normal_initializer(mean=0, stddev=1., seed=1))

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print v1.eval()
    print v2.eval()

输出：

[ 0.33786374 -0.9326936  -0.27475813  0.07207271  1.80768931]
[-0.81131822  1.48459876  0.06532937  0.0992484   0.63969707]

tf.random_uniform_initializer–均匀分布

#!/usr/bin/python
# coding:utf-8

import tensorflow as tf
# 均匀分布
v1 = tf.get_variable('v1', shape=[5], initializer=tf.random_uniform_initializer())
# 指定均匀分布的上下限和种子值
v2 = tf.get_variable('v2', shape=[5], initializer=tf.random_uniform_initializer(maxval=-1., minval=1., seed=0))

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print v1.eval()
    print v2.eval()

输出：

[ 0.62208343  0.81940019  0.67432535  0.07955837  0.59436822]
[ 0.79827476 -0.9403336  -0.69752836  0.90343738  0.90295386]

tf.zeros_initializer–全0初始化

tf.ones_initializer–全1初始化

#!/usr/bin/python
# coding:utf-8

import tensorflow as tf
# 全0
v1 = tf.get_variable('v1', shape=[5], initializer=tf.zeros_initializer())
# 全1
v2 = tf.get_variable('v2', shape=[5], initializer=tf.ones_initializer())

with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    print v1.eval()
    print v2.eval()

输出：

[ 0.  0.  0.  0.  0.]
[ 1.  1.  1.  1.  1.]