请求指导：在Google Colab复现MNIST网络并启动TensorBoard

Hey there, let's get your MNIST + TensorBoard setup working smoothly in Colab— I've walked through this dozens of times, so here's a step-by-step breakdown:

1. 适配代码到Colab环境 #

First off, since your reference code is from TensorFlow 1.8, you'll want to make sure Colab runs the compatible TF version. At the top of your notebook, run this to switch to TF 1.x:

%tensorflow_version 1.x

Then either copy the full code from mnist_with_summaries.py into a code cell, or pull the repo directly:

!git clone https://github.com/tensorflow/tensorflow.git

Navigate to the example directory and tweak the log directory in the code to use Colab's accessible temp storage— update the LOG_DIR variable to:

LOG_DIR = '/tmp/tensorflow/mnist/logs/mnist_with_summaries'

2. 确认日志文件生成 #

After running the training script, verify summary files were created by checking the directory:

!ls -la /tmp/tensorflow/mnist/logs/mnist_with_summaries

You should see subdirectories like train and test with .tfevents files inside— that means summaries are being written correctly.

3. 启动TensorBoard（原生Colab方法） #

Forget the old ngrok workarounds— Colab has built-in TensorBoard support now. Run these two commands in separate cells:
First load the extension:

%load_ext tensorboard

Then launch TensorBoard pointing to your log directory:

%tensorboard --logdir /tmp/tensorflow/mnist/logs/mnist_with_summaries

A panel will pop up in your notebook with the full TensorBoard interface ready to use.

常见问题排查 #

• If TensorBoard shows "No dashboards are active": Double-check your log directory path matches what's in the code, and confirm the training script ran long enough to write summaries.
• If you get TF version errors: Make sure you ran %tensorflow_version 1.x before importing TensorFlow— Colab defaults to TF 2.x these days, which breaks TF 1.x syntax.
• If the log directory is empty: Ensure the summary ops in the code are properly merged and written using tf.summary.FileWriter.

内容的提问来源于stack exchange，提问作者Vikas Dhiman