TensorBoard简单使用

Posted on | In

简单来说

  • Tensorboard可以对静态的图结构和动态的训练结果（accuracy， loss）等进行可视化
  • Tensorboard是通过将数据写入到本地的log文件中，再以本地web的方式呈现

如何使用

1
2
3
4
5
6
7
8
9
10
11
12
# 为训练结果等Operation建立summary
tf.summary.scalar("cost", cross_entropy)
tf.summary.scalar("accuracy", accuracy)
# 合并为一个op，这样sess中只需要运行一次就可以了
summary_op = tf.summary.merge_all()
with tf.Session()  as sess:
	sess.run(tf.initialize_all_variables())
	# create log writer object with static graph structure
	writer = tf.summary.FileWriter(logs_path, graph=tf.get_default_graph())
	_, summary = sess.run([train_op, summary_op], feed_dict={x: batch_x, y_: batch_y})
	# write log
	writer.add_summary(summary, epoch * batch_count + i)

然后运行命令启动TensorBoard服务

1
tensorboard --logdir=your/log/path

参考