Google Colab mount Google Drive and connect with Weights and Biases to track model training progress(cool TensorBoard alternative)

I use Google Colab since it provides a free GPU, and it’s pretty easy to connect to your google drive. It feels like having another computer. But my TensorBoard code didn’t work when I opened a Jupyter notebook. This is because TensorBoard runs locally, and my Colab “computer” hasn’t set it up yet. Then I found a great alternative toolkit, Weights and Biases. I love the name, and it is so easy to use; here are some notes on how to use it.

Mount Google Drive

It is easy, just authorize it as below

from google.colab import drive

# Prompt for authorization.
drive.mount('/content/drive')

# Check your drive
!ls "/content/drive/My Drive"

You can check the file route on the left

Save your folder path for convenience

Connect to Weights and Biase

Install the package and log in, paste the API key pop up from your WB page.

!pip install wandb
!wandb login

#initial a project for record
import wandb

wandb.init(project="my-test-project")

Press display, you could track the information directly from the notebook

Now you can log and visualise the value easily

wandb.log({"Critic":critic_loss, "Actor":actor_loss})

Here is a simple example:

We could check the progress, I must have done something wrong with my model, the loss is increasing…