tensorboard可视化

---恢复内容开始---

1. 查看tensorboard的一个简单例子

#tensorflow's tensorboard
import tensorflow as tf

x=tf.constant(1.0,name="input")
w=tf.Variable(0.5,name="weight")
b=tf.Variable(0.1,name="bias")
y=tf.add(tf.multiply(x,w,name="mul_op"),b,name="add_op")
#设置写入的文件夹
summary_writer=tf.summary.FileWriter("./calc_graph")
#获取默认的图
graph=tf.get_default_graph()
summary_writer.add_graph(graph)
#将图结构写入文件
summary_writer.flush()

---

 

在命令行中进到放置calc_graph文件夹的目录下，输入：

tensorboard --logdir=./calc_graph

根据提示打开浏览器，输入地址，便可以查看保存的图结构

至此，我们查看了tensorboard的  graph模块

 

 

2.一个简单的梯度下降例子

import numpy as np
import tensorflow as tf

 

x=tf.placeholder(tf.float32)
y=tf.placeholder(tf.float32)

weight=tf.get_variable("weight",[],tf.float32,initializer=tf.random_normal_initializer())
bias=tf.get_variable("bias",[],tf.float32,initializer=tf.random_normal_initializer())
pred=tf.add(tf.multiply(x,weight,name="mul_op"),bias,name="add_op")

#############定义参数，准备数据###################################

 

loss=tf.square(y - pred,name="loss")
optimizer=tf.train.GradientDescentOptimizer(0.01)

#计算梯度和应用梯度， 可以用optimizer.minimize(loss)代替下面两个语句，效果相同
grads_and_vars=optimizer.compute_gradients(loss)
train_op=optimizer.apply_gradients(grads_and_vars)

##################模型相关设定######################

 

tf.summary.scalar("weight",weight)
tf.summary.scalar("bias",bias)
tf.summary.scalar("loss",loss[0])
#合并summary
merged_summary=tf.summary.merge_all()

summary_writer=tf.summary.FileWriter("./log_graph")
summary_writer.add_graph(tf.get_default_graph())

###############tensorboard中收集训练数据，放入计算图#########################

init_op=tf.global_variables_initializer()

with tf.Session() as sess:
　　sess.run(init_op)
　　for step in range(500):
　　　　train_x=np.random.randn(1)
　　　　train_y=2* train_x+np.random.randn(1)*0.01+10
　　　　_,summary=sess.run([train_op,merged_summary],feed_dict={x:train_x,y:train_y})#sess.run执行了两个操作
　　　　summary_writer.add_summary(summary,step)

我们可以查看tensorboard的Scalars和Graphs模块:

 

 

 

---

 

 3.一个简单的神经网络例子

 增加了命名空间，让图结构更清晰

 

#有一个隐藏层的简单神经网络

#定义方法添加层数

import tensorflow as tf

def add_layer(input_x,in_size,out_size,activation_function=None):

    with tf.name_scope("layer"):

        with tf.name_scope("weights"):

            weights=tf.Variable(tf.random_normal([in_size,out_size]))

        with tf.name_scope("bias"):

            bias=tf.Variable(tf.zeros(out_size)+0.1)

        with tf.name_scope("wx_plus_b"):

            wx_plus_b=tf.matmul(input_x,weights)+bias

        if activation_function is None:

            outputs=wx_plus_b

        else:

            outputs=activation_function(wx_plus_b)

        return outputs

 

import numpy as np

import matplotlib.pyplot as plt

%matplotlib inline

x_data=np.linspace(-1,1,300)[:,np.newaxis]#增加一个维度，变成二维了

noise=np.random.normal(0,0.1,x_data.shape)

y_data=np.square(x_data)-0.5+noise

 

with tf.name_scope("input"):

    xs=tf.placeholder(tf.float32,[None,1],name="x_input")

    ys=tf.placeholder(tf.float32,[None,1],name="y_input")

#############数据准备#####################

 

hidden_layer_out=add_layer(xs,1,10,activation_function=tf.nn.relu)

final_layer_out=add_layer(hidden_layer_out,10,1,activation_function=None)

###################模型构造########################

with tf.name_scope("loss"):

    loss=tf.reduce_mean(tf.reduce_sum(

        tf.square(y_data-final_layer_out),

        reduction_indices=[1]

    ))

optimizer=tf.train.GradientDescentOptimizer(0.01)

with tf.name_scope("train"):

    train_op=optimizer.minimize(loss)

 

fig=plt.figure()

ax=fig.add_subplot(1,1,1)

ax.scatter(x_data,y_data)

plt.ion()

plt.show

 

init=tf.global_variables_initializer()

with tf.Session() as sess:

    summary_writer=tf.summary.FileWriter("./log_graph4")

    summary_writer.add_graph(tf.get_default_graph())

    sess.run(init)

 

 

取消add_layer中的命名空间后，图结构如下：

 

 

 

 

---恢复内容结束---