import tensorflow as tf
'''
1、scalar(标量)
2、image
3、audio
4、histogram
5、graph
'''
##可视化
#with tf.variable_scope("foo",reuse=tf.AUTO_REUSE):
# with tf.device("/cpu:0"):
# x_init1 = tf.Variable(name="init_x",dtype=tf.float32,initial_value=tf.random_normal(shape=[10]))
# x = tf.Variable(initial_value=x_init1,name='x')
# y = tf.constant(3.0)
# z = x+y
#
#with tf.variable_scope("bar"):
# a = tf.constant(3.0)+4.0
#
#w = z*a
#
##开始记录信息
#tf.summary.scalar(name='scalar_init_x',tensor=x_init1)
#tf.summary.scalar(name='scalar_x',tensor=x)
#tf.summary.scalar(name='scalar_y',tensor=y)
#tf.summary.scalar(name="scalar_z",tensor=z)
#tf.summary.scalar(name='scalar_w',tensor=w)
#
##更新操作
#assign_op = tf.assign(ref=x,value=x+1)
#with tf.control_dependencies([assign_op]):
# with tf.device("/gpu:0"):
# out = x*y
# tf.summary.scalar("scalar_out",out)
#
#
#with tf.Session(config=tf.ConfigProto(log_device_placement=True,allow_soft_placement=True)) as sess:
# #merge all summery
# merged_summary = tf.summary.merge_all()
# #得到输出到文件的对象
# writer = tf.summary.FileWriter('./result',sess.graph)
#
# #变量全局初始化
# sess.run(tf.global_variables_initializer())
#
# #print
# for i in range(1,5):
# #summary = sess.run(merged_summary)
# r_out,r_x,r_w = sess.run([out,x,w])
# #writer.add_summary(summary,i)
# print("{},{},{}".format(r_out,r_x,r_w))
import numpy as np
def add_layer(inputs,in_size,out_size,n_layer,activation_funtion=None):
layer_name='layer%s'%n_layer
with tf.name_scope('layer'):
with tf.name_scope('weight'):
Weights=tf.Variable(tf.random_normal([in_size,out_size]),name='W')
tf.summary.histogram(layer_name+'/Weights',Weights) #各层网络权重,偏置的分布,用histogram_summary函数
with tf.name_scope('biases'):
biases=tf.Variable(tf.zeros([1,out_size])+0.1,name='b')
tf.summary.histogram(layer_name+'/biases',biases)
with tf.name_scope('Wx_plus_b'):
Wx_plus_b=tf.matmul(inputs,Weights)+biases #inputs与weight 顺序不能换
if activation_funtion is None:
output=Wx_plus_b
else:
output=activation_funtion(Wx_plus_b)
tf.summary.histogram(layer_name+'/output',output)
return output
x_data = np.linspace(-1,1,300, dtype=np.float32)[:, np.newaxis]
noise = np.random.normal(0, 0.05, x_data.shape).astype(np.float32)
y_data = np.square(x_data) - 0.5 + noise
with tf.name_scope('inputs'):
xs=tf.placeholder(tf.float32,[None,1],name='x_in')
ys=tf.placeholder(tf.float32,[None,1],name='y_in')
l1=add_layer(xs,1,10,n_layer=1,activation_funtion=tf.nn.relu)
prediction=add_layer(l1,10,1,n_layer=2,activation_funtion=None)
# loss=tf.reduce_mean(tf.reduce_sum(tf.square(ys-prediction),
# reduction_indices=[1]))
with tf.name_scope('loss'):
loss=tf.reduce_mean(tf.square(ys-prediction))
tf.summary.scalar('loss',loss) #数值如学习率,损失函数用scalar_summary函数,tf.scalar_summary(节点名称,获取的数据)
optimizer=tf.train.GradientDescentOptimizer(0.1)
with tf.name_scope('train'):
train=optimizer.minimize(loss)
#变量全局初始化
init = tf.global_variables_initializer()
sess=tf.Session()
sess.run(init)
#整个图经常需要检测许许多多的值,也就是许多值需要summary operation,一个个去run来启动太麻烦了,所以就合并所有获得的值
merged=tf.summary.merge_all()#合并所有的summary data的获取函数,merge_all 可以将所有summary全部保存到磁盘,以便tensorboard显示。如果没有特殊要求,一般用这一句就可一显示训练时的各种信息了。
writer=tf.summary.FileWriter("logs/",sess.graph)#把图保存到一个路径,FileWriter从tensorflow获取summary data,然后保存到指定路径的日志文件中
for i in range(1000):
sess.run(train,feed_dict={xs:x_data,ys:y_data})
if i%50==0:
#summary的操作对于整个图来说相当于是外设,因为tensorflow是由结果驱动的,而图的结果并不依赖于summary操作,所以summary操作需要被run
rs=sess.run(merged,feed_dict={xs:x_data,ys:y_data})#运行所有合并所有的图,获取summary data函数节点和graph是独立的,调用的时候也需要运行session
writer.add_summary(rs,i)#把数据添加到文件中,每一次run的信息和得到的数据加到writer里面,主要是描述数据变化,所以要这样,若是只有流图,就不需要这样
# print(sess.run(loss,feed_dict={xs:x_data,ys:y_data}))