TensorBoard 是一个用于可视化 TensorFlow 图形和训练进度的工具，它可以帮助我们更好地理解训练过程中的数据结构、计算图、损失函数、评估指标等。TensorBoard 可以展示多种信息，例如训练过程中绘制的图像、网络结构、直方图、图片和渲染等。
使用 TensorBoard 的方法如下：
- 首先，确保已经安装了 TensorFlow。在安装 TensorFlow 时，TensorBoard 会自动安装。如果使用 conda 安装，可以在 conda 环境中运行 conda install tensorflow。
- 在训练 TensorFlow 模型时，需要将 TensorBoard 的日志记录器添加到训练循环中。以下是一个简单的示例：
import tensorflow as tf
创建一个简单的计算图
a = tf.constant([1.0, 2.0, 3.0], name="input1")
b = tf.constant([1.0, 2.0, 3.0], name="input2")
c = tf.add(a, b)
初始化 TensorBoard 的日志记录器
log_dir = "logs" # 指定日志目录
writer = tf.summary.create_file_writer(log_dir) # 创建日志记录器
训练循环
with tf.GradientTape() as tape:
y = c
loss = tf.reduce_mean(tf.square(y))
tf.summary.scalar("loss", loss, writer=writer) # 记录损失函数
for i in range(1000):
# 训练代码（省略）
#...
关闭 TensorBoard 的日志记录器
writer.close()
CopyCopy
- 在浏览器中打开 http://localhost:6006，就可以看到 TensorBoard 的界面。在界面上，您可以查看训练过程中的指标变化、模型结构、权重分布等详细信息。
请注意，如果使用的是 TensorFlow 2.x 版本，日志记录器的使用方法略有不同。您可以参考官方文档了解更多关于 TensorBoard 的使用方法：https://www.tensorflow.org/tutorials/tensorboard
Using TensorBoard
TensorBoard is a great way to visualize what's happening behind the code.
In this example, we'll loop through some numbers to improve our guess of the average value. Then we can visualize the results on TensorBoard.
Let's just set ourselves up with some data to work with:
import tensorflow as tf
import numpy as np
raw_data = np.random.normal(10, 1, 100)
The moving average is defined as follows:
alpha = tf.constant(0.05)
curr_value = tf.placeholder(tf.float32)
prev_avg = tf.Variable(0.)
update_avg = alpha * curr_value + (1 - alpha) * prev_avg
Here's what we care to visualize:
avg_hist = tf.summary.scalar("running_average", update_avg)
value_hist = tf.summary.scalar("incoming_values", curr_value)
merged = tf.summary.merge_all()
writer = tf.summary.FileWriter("./logs")
Time to compute the moving averages. We'll also run the merged op to track how the values change:
init = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init)
for i in range(len(raw_data)):
summary_str, curr_avg = sess.run([merged, update_avg], feed_dict={curr_value: raw_data[i]})
sess.run(tf.assign(prev_avg, curr_avg))
print(raw_data[i], curr_avg)
writer.add_summary(summary_str, i)
9.247841477069203 0.4623921
10.019298730125382 0.9402374
11.971773672793464 1.4918143
10.702923359431118 1.9523697
11.667057068606786 2.4381042
9.143228690197773 2.7733603
9.457709656523708 3.1075776
12.33999608545561 3.5691986
9.543410229631846 3.8679092
9.251442209932934 4.137086
8.942198790212387 4.3773413
11.019946553148321 4.709471
11.430198193578404 5.0455074
8.6213954795195 5.224302
10.822108995258686 5.504192
10.58310002901428 5.7581377
10.20420365104725 5.9804406
10.312154931419304 6.1970263
10.545111153579882 6.4144306
8.797765458370709 6.5335975
8.56686695526782 6.6352606
12.570525410195215 6.9320235
11.543815331679966 7.162613
10.320920832332627 7.320528
10.423914230722215 7.4756975
10.619258439210187 7.6328754
9.101109809288653 7.7062874
9.841278298991933 7.813037
9.099955845561944 7.877383
9.41973125623955 7.9545
11.082836040691273 8.110917
10.116690980009775 8.2112055
9.402594289154155 8.270775
10.925993106488145 8.403536
10.243254438024696 8.495522
9.477769687949733 8.544634
9.351362392482848 8.58497
9.242191906408548 8.617831
12.123667719477677 8.793122
10.076009517273803 8.857266
9.74900667301667 8.901854
10.830363231386094 8.998279
8.861116004341559 8.991421
10.007389057190906 9.042218
10.769369554012615 9.128575
12.971561516039255 9.3207245
9.875042913748056 9.34844
9.64616462712992 9.363326
9.76634851758219 9.383477
9.326634526001623 9.380634
8.492294014699189 9.336217
10.006073094467316 9.369709
9.442892778881891 9.373368
9.56787198816676 9.383093
9.961494974707488 9.412013
9.572285501643822 9.420026
11.851354361154291 9.541592
10.833573476171445 9.606191
11.836376240592454 9.7177
11.047626672901409 9.784197
10.913818292308468 9.840677
10.60857743486623 9.879072
9.883074005285522 9.879272
8.227633816367192 9.79669
9.788906167639809 9.796301
9.001469197788671 9.756559
8.918205933440774 9.714642
9.885320274459133 9.723175
10.77521268535355 9.775776
9.68349427673202 9.771162
10.113753965038361 9.788292
9.6597232190883 9.781863
9.323572053812015 9.758949
9.618532841629188 9.751928
9.011944462852757 9.714929
8.323719148832197 9.645369
9.442883485401897 9.635244
10.430287903497137 9.674997
10.838671174170663 9.733181
9.346134056876938 9.713829
10.234079103904495 9.739841
9.692786236742311 9.737489
8.675916172925552 9.68441
9.7006074487691 9.68522
10.064675943184373 9.704192
9.4021612098359 9.689091
11.124410899430886 9.760857
10.034575898474612 9.774543
9.793431430576485 9.775487
10.889420930759462 9.831183
9.253518007206916 9.8023
11.114827470151916 9.867927
9.378996323459113 9.843479
9.864306640072803 9.844521
11.803316169037448 9.94246
10.103049011008196 9.95049
8.723187258083906 9.889125
8.985505621881307 9.843944
10.690261212066178 9.88626
8.426969249944442 9.813295
Check out the visualization by running TensorBoard from the terminal:
$ tensorboard --logdir=path/to/logs
#made the logs be written successfully
writer.close()
