#2656 - Committing current state before lunch. New ActionPenalty reward added. Basic implementation returns a -1 reward if last_action_response.action isn't DONOTHING. Minor change in data_manipulation so I can see it working in the data_manipulation notebook. Need to use configured values but so far, promising?. Looks to result in a better average reward than without which is good, I think.

2024-06-25 12:29:01 +01:00
parent 4a81dc3b2c
commit db27bea4ec
2 changed files with 33 additions and 1 deletions
--- a/src/primaite/config/_package_data/data_manipulation.yaml
+++ b/src/primaite/config/_package_data/data_manipulation.yaml
@@ -740,7 +740,6 @@ agents:
            agent_name: client_2_green_user


-
    agent_settings:
      flatten_obs: true