tensorboard记录数据

在https://blog.csdn.net/ShenWeiKKX/article/details/101112503我们使用tensorboard展示了网络结构，本文介绍利用tensorboard记录数据，一般记录数据在训练过程中的变化，如loss，accuracy等等。

import tensorflow as tf
import numpy as np
...
out = tf.layers.conv2d(x,filter_nums,filter_size,use_bias=False,
                       kernel_initializer=tf.truncated_normal_initializer(),
                       padding='same',activation=None,name = 'conv1')  #函数的name
...
with tf.name_scope('input'):  #空间范围的name
    image_x = tf.placeholder(imagename_array.dtype,[None],name='image_x') #函数的name
    label_y = tf.placeholder(label_agen_array.dtype,[None,7,7,6],name='label_y') #函数的name
    istraing = tf.placeholder(tf.bool,name='istraing') #函数的name
...
with tf.name_scope('loss'):
	...
    with tf.name_scope('total_loss'):
        total_loss = loss1 + loss2 + loss3
        tf.summary.scalar('total_loss_sca',total_loss)      #记录total_loss的scalar图
        tf.summary.histogram('total_loss_his',total_loss) #记录total_loss的直方图
...
# 合并所有的summary
merged = tf.summary.merge_all()
...
with tf.Session() as sess:
    sess.run(tf.global_variables_initializer())
    writer = tf.summary.FileWriter('logs/',sess.graph)  #'logs/'是写入路径，sess.graph表示将图写入，所以不需要训练
   for epochs in range(1):
	    summary = sess.run(merged, {istraing:True})
	    writer.add_summary(summary,epochs) #使epochs为横坐标
	    ...

1. 与https://blog.csdn.net/ShenWeiKKX/article/details/101112503的tensorboard网络结构相比，此处的代码多了一下，这些多出的代码正是为了记录数据;
2. 接下来的步骤与tensorboard展示网络结构一致;
3. 最后打开tensorboard就可以查看这些scalar和histogram图了。