How To Use DetaMetrics (TensorBoard Alternative for Cloud Notebooks)
Disclaimer: I created DetaMetrics, and thought it’d be nice to explain how to use it. I am not affiliated in any way with Deta Space.
TensorBoard is a great resource, allowing you to view your machine learning models’ training progress, however you often run into issues when running your models on Kaggle or Google Colab, as most people do as we do not have the hardware for training larger models.
To solve this, I created DetaMetrics, a simple to use alternative to TensorBoard that works with all your cloud notebooks, it takes 5 minutes to set up and will save you hours of matplotlib graphing and frustration in the future.
How To Use
You only have to do the following once.
- Sign in or register for Deta Space, a free microservice hosting service
- Go to the DetaMetrics discover page
- Click Install
- Now, go back to your Deta Space canvas (this link will work if you’re signed in to Deta)
- Click the 3 dots on the DetaMetrics tile
- Go to Keys > Create New API Key. It is important you’re creating an API key and not a Data Key. Copy and save the given key somewhere.
Configuring your notebook..
- Install the SDK in your note book with…
%pip install detametrics- You now have it all set up (but please read on for a brief usage tutorial).
- Refer to the Docs for info on the TensorFlow call-back and Base API that’s compatible with any custom training loops, including PyTorch.
- If you click on the DetaMetrics Tile on your Deta Canvas (this link works if you’re signed into Deta Space) it will come up with the DetaMetrics dashboard. You need to paste in your API key into the input rectangle at the top middle.
- The dashboard has helpful controls like data deletion, a dark theme and live reload. It also has a button to find the URL ID of your dashboard.
Helpful Notes: Use Kaggle’s Secrets function (in the Add-ons tab) to store your API key for security.
