Most Simple Usage Of Tensorboard

来源:互联网 发布:中国网络发展史插画 编辑:程序博客网 时间:2024/05/18 14:45

Starting to learn tensorflow, some questions confused me so much. The problems are like such. Does it necessary to initialize the tensorflow.summaryFileWriter instance? sometimes I wrote a graph and be able to access the graph, sometimes failed to do so.
I had some try. I figured some of my questions out.

Write A Most Simple Graph

import tensorflow as tfwith tf.name_scope('scp1'):    a = tf.Variable(5.5,name="var0")    b = a + tf.Variable(34.2)with tf.Session() as sess0:    tf.summary.FileWriter('logs',sess0.graph)

save the above code to a .py file, execute it. In the directory where the .py file is stored, there will be a subdirectory logs created. Open command prompt at the dierectory contented the .py file and execute the command tensorboard --logdir logs. It should not be tensorboard --logdir = logs. Then, you can view the graph in chrome.

Problems I Encountered

P1

import tensorflow as tfwith tf.name_scope('scp1'):    a = tf.Variable(5.5,name="var0")    b = a + 34.2with tf.Session() as sess0:    tf.summary.FileWriter('logs',sess0.graph)

I tried this code, I never wrote a graph. The number 3.42 in the code should be tensorflow.constant(34.2) or tensorflow.Variable(34.2) .

P2

import tensorflow as tfwith tf.name_scope('scp1'):    a = tf.Variable(5.5,name="var0")    b = a + 34.2with tf.Session() as sess0:    graf = tf.summary.FileWriter('logs',sess0.graph)    tf.global_variables_initializer().run()

Once, I thought the instance of tensorflow.summary.FileWriter may also need to be initialized. I tried some code like the above. In fact, the last sentence above is not necessary.
As the name implied, it’s the the initializer for variables. The graph is not a variable. So, the initializer is not work for the graph. What’s more. The graph is just depends on the definition of the structure. For showing a graph of the structure, initializing the whole structure seems strange. These were what I perplexed with, when I found a code with tf.global_variables_initializer().run() worked.

原创粉丝点击