Reinforcement Learning Training Dashboard
To show you how you can use custom dashboards in your Reinforcement Learning projects, we prepared an example dashboard for the training phase.
-> Not sure what a custom dashboard in Neptune is? You’ll find this info below.
See this dashboard in the app (no registration needed).
What widgets are added to the dashboard?
- Episode duration
- Episode return
- Episode recordings
- Hardware utilization
When is such a dashboard useful?
- When you want to have one source of truth to keep an eye on the RL training session,
- Especially with rewards, average-duration plots, and gifs of how the agent works in the first and final episodes.
Custom dashboards in Neptune
When analyzing your metadata in the Neptune app, you can create custom dashboards.
They let you combine different metadata types (charts, parameters, hardware utilization, videos, etc.) in a single view.
Why would you need dashboards?
- To see all metadata related to a specific part of the project in one place,
- To see training and validation results on one chart,
- To see validation metrics and image predictions on a subset of a validation set next to each other,
- To divide metadata by stages of the project and display each group in a different dashboard,
- To analyze the same group of charts weekly or for each of your projects, for example, when you want to have a template dashboard to check it every time and/or to show it to other stakeholders
These are the most common use cases that we hear about from our users.
But for sure, these are not the only use cases. We call them “custom” dashboards for a reason – you can set them up however you want.
How to create a custom dashboard?
Here’s documentation that explains how to create a custom dashboard in the Neptune app.
Other useful resources
Check a full RL-focused example project in Neptune.
See other example dashboards in the Neptune app (including a general overview of metadata, dashboards for different stages of the ML project, and more).
Couldn’t find the use case you were looking for?
Just get in touch, and our ML team will create a custom demo for you.