Run workflow

New Project / New Experiment

• Users start a new experiment to try a different Reward Function.
  • Can Save Draft of the reward function.
  • Can perform a Test Run
    • Test Runs can't be stopped. Some internal users have asked for it, but I want to get more feedback before making it more complicated.
    • Once a test run is started, the 'Reward Function' can't be changed in that experiment. Users will need to start a new experiment.

Experiment View

• Users track the progress of the experiment on this page.
  • The first use is to confirm the test run + reward function is working (going up)
  • Once the user is satisfied with the test run results, the can start a discovery run.
    • Start Discovery Run
      • Only available after a test run has started. Doesn't have to wait until the test run is completed.
      • This starts training for 12 policies with different configurations.
  • Users can download the trained policies from any of the policies in the table and use them in AnyLogic.
  • If they like the results of a specific policy, they will choose one and have it train longer to see if the results can be improved.
  • Run Full Training
    • This is available for any policy in the table
      • That is a Test Run or Discovery Run (is not a Full Run Policy)
      • Has not already had a full run started from it. (This is optional for beta if it makes it too complicated)

Questions:

• Policies are created for each type of run:
  • Test run -> One policy
  • Discovery run -> 12 policies (old spec showed 35)
  • Full run -> One policy
• For the states I assume we're looking at for example Run Type equals Discovery Run and that the status is Completed? Yep.
• Are these all the possible states and options?