TensorBoard

TensorBoard is a tool for providing the measurements and visualizations required during machine learning workflows. It enables you to track experimental metrics such as loss and accuracy, visualize model graphs, project embeddings into lower dimensional spaces, and more.

Create an instance mirrored as TensorFlow 2. After the instance is created and started, you can click the link in the instance list to open JupyterLab.

Run the notebook code in Getting Started with TensorBoard.

First download the get_started.ipynb notepad file. Upload to the server via JupyterLab.

Double-click the file to open Notepad, and you can see the code content in Notepad in the work area on the right.

The instance has already provided a started TensorBoard, so there are two commands to start tensorboard in the code that need to be commented out with #.

There is an error in the code that is marked as code and needs to be corrected. Click the area in front of the code block in the figure below to select it, and modify the code to Markdown above.

Select Run - Run All Cells in the menu to run all the code.

After waiting for all cells to finish running, right-click the generated logs folder and select Cut.

The folder read by TensorBoard in the instance is the /tb_logs folder, right-click and paste into this directory.

Click the link in the instance list to open TensorBoard.

warning

Some new versions of TensorBoard may display a white screen in the Safari browser and need to be opened with Chrome.

After opening, you can see the results obtained after running the code just now.

In the actual operation process, you can let the program output the results directly to the /tb_logs folder. You can also create a soft link in the /tb_logs folder to point to the actual training result directory.