Google Colab provides free GPU acceleration for Machine Learning. During neural network training, we wanna see training progress and result. Using Tensorboard is convenient way to visualize the progress. This tutorial will show you how to do so in the most simplest way.
Principle
Run Tensorboard in the background of Google Colab, use ngrok to tunnel traffic to localhost, so that it can be accessed outside Google Colab. Detail of code can be found in this stackoverflow. Credit to Joope 🙏🏻
Feel little bit complicated? Here is a library to do it easier
Installation
pip install tensorboardcolab
Initialization
tbc=TensorBoardColab()
After initialization, TensorBoard link will be shown in Colab Google Juyter output.
Done! One line of code can start up TensorBoard in Google Colab, easy, right? :) But how to use it? Read below …
Add to Keras callback
model.fit(x,y,epochs=100000,callbacks=[TensorBoardColabCallback(tbc)])
That is all! Now you can read the progress graph
Oh! Wait, you want more control? Easy, read below …
Save picture to TensorBoard
tbc.save_image(title=”test_title”, image=image)
Save a value to graph of TensorBoard
tbc.save_value(“graph_name”, “line_name”, epoch, value)tbc.save_value(“graph_name2”, “line_name2”, epoch, value2)
.
.
.tbc.flush_line(line_name)
tbc.flush_line(line_name2)
.
.tbc.close()
Hope you enjoyed!
Reference
About Me
LinkedIn: https://www.linkedin.com/in/tommy-tao-a5961156/
Twitter: https://x.com/TaoManWai