Reinforcement Learning Example Project

Here’s a public example project to give you a taste of neptune.ai’s API and the app.

It’s a Reinforcement Learning project that shows how to use Neptune in the training, evaluation, and deployment phases of an RL project.

You can just open the project and play with the app, no registration is needed. And if you want to see the code behind it, go to the example’s GitHub repo. 

What’s in this Reinforcement Learning example project? 

Here are a few most important things you can see in the project: 

• Runs table – when you open the project, you will land in the view that shows all the experiments in a table format.

• You can easily change the columns that are visible in the runs table or filter the runs. You can also save custom views for later. In this project, we have saved several custom views, e.g., “evaluated”, “in prod”, “top”, etc. To switch between views, open the dropdown menu above the runs table.

• To compare multiple runs, go to the Compare runs tab. You can switch from Charts comparison to Parallel coordinates, Images comparison, or Side-by-side comparison. If you’ve logged artifacts, you can also compare datasets between runs. 

By the way, you can choose which runs you want to compare by clicking the eye icon in the runs table. 

• You can also inspect a single run. Click the Runs details tab at the top of your screen, and you’ll land in the single run’s view. Just below the name of the run, you can see different dashboards.
  
  The first one is the All metadata dashboard with everything that was logged to Neptune (organized in a folder-like structure). Then, there are Charts, Images, Monitoring, Source code, and more. 

• At the end of the list, you can see some custom dashboards that combine different metadata types in one place. In this project, we created four different dashboards, one for every stage of the project. We dive deeper into this dashboard here.