Tensorflow MNIST 数据集測试代码入门

測试代码已上传至GitHub：yhlleo/mnist

将MNIST数据集，下载后复制到目录Mnist_data中，假设已经配置好tensorflow环境，基本的四个測试代码文件，都能够直接编译执行：

• mnist_softmax.py: MNIST机器学习入门
• mnist_deep.py: 深入MNIST
• fully_connected_feed.py: TensorFlow运作方式入门
• mnist_with_summaries.py: Tensorboard训练过程可视化

mnist_softmax.py执行结果比較简单，就不列举。

mnist_deep.py迭代执行较为耗时，结果已显示在博客： 深入MNIST code測试 。

fully_connected_feed.py的执行结果例如以下（本人电脑为2 CPU，没有使用GPU）：

Extracting Mnist_data/train-images-idx3-ubyte.gz
Extracting Mnist_data/train-labels-idx1-ubyte.gz
Extracting Mnist_data/t10k-images-idx3-ubyte.gz
Extracting Mnist_data/t10k-labels-idx1-ubyte.gz
I tensorflow/core/common_runtime/local_device.cc:25] Local device intra op parallelism threads: 2
I tensorflow/core/common_runtime/local_session.cc:45] Local session inter op parallelism threads: 2
Step 0: loss = 2.33 (0.023 sec)
Step 100: loss = 2.09 (0.007 sec)
Step 200: loss = 1.76 (0.009 sec)
Step 300: loss = 1.36 (0.007 sec)
Step 400: loss = 1.12 (0.007 sec)
Step 500: loss = 0.74 (0.008 sec)
Step 600: loss = 0.78 (0.006 sec)
Step 700: loss = 0.69 (0.007 sec)
Step 800: loss = 0.67 (0.007 sec)
Step 900: loss = 0.52 (0.010 sec)
Training Data Eval:
  Num examples: 55000  Num correct: 47532  Precision @ 1: 0.8642
Validation Data Eval:
  Num examples: 5000  Num correct: 4360  Precision @ 1: 0.8720
Test Data Eval:
  Num examples: 10000  Num correct: 8705  Precision @ 1: 0.8705
Step 1000: loss = 0.56 (0.013 sec)
Step 1100: loss = 0.50 (0.145 sec)
Step 1200: loss = 0.33 (0.007 sec)
Step 1300: loss = 0.44 (0.006 sec)
Step 1400: loss = 0.39 (0.006 sec)
Step 1500: loss = 0.33 (0.009 sec)
Step 1600: loss = 0.56 (0.008 sec)
Step 1700: loss = 0.50 (0.007 sec)
Step 1800: loss = 0.42 (0.006 sec)
Step 1900: loss = 0.41 (0.006 sec)
Training Data Eval:
  Num examples: 55000  Num correct: 49220  Precision @ 1: 0.8949
Validation Data Eval:
  Num examples: 5000  Num correct: 4520  Precision @ 1: 0.9040
Test Data Eval:
  Num examples: 10000  Num correct: 9014  Precision @ 1: 0.9014
[Finished in 22.8s]

mnist_with_summaries.py主要提供了一种在Tensorboard可视化方法，首先。编译执行代码：

执行完成后，打开终端Terminal，输入tensorboard --logdir=/tmp/mnist_logs（与writer = tf.train.SummaryWriter(‘/tmp/mnist_logs‘, sess.graph_def)中的文件路径一致），终端中就会执行显示：Starting TensorBoard on port 6006 (You can navigate to http://localhost:6006)

然后，打开浏览器，输入链接http://localhost:6006：

当中，有一些选项。比如菜单条里包含EVENTS, IMAGES, GRAPH, HISTOGRAMS，都能够一一点开查看~

另外，此时假设不关闭该终端，是无法在其它终端中又一次生成可视化结果的，会出现端口占用的错误。很多其它具体信息能够查看英文原文：TensorBoard: Visualizing Learning。

如有纰漏，欢迎指正！