Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Changes for using dynamic orders in harl #71

Closed
wants to merge 95 commits into from

Conversation

bmielnicki
Copy link
Collaborator

@bmielnicki bmielnicki commented Mar 11, 2021

Changes that are needed so my harl changes HumanCompatibleAI/human_aware_rl#16 can work.
Includes changes from #38 and #57 and:

  • ending episode after fulfilling n orders
  • max number of orders at same time
  • GreedyHumanModel can be pedagogical in the way it picks ingredients; it is generally improved and compatible with dynamic orders Dynamic orders #57
  • SlowedDownAgent (agent that does wait action fraction of the time)
  • fix of SampleAgent
  • possibility to ignore choosed dispensers/counters by mlam
  • multiple typos fixes
  • moving all of preprocessing/encoding/featurizations functions and state spaces to OvercookedGridworld/OvercookedEnv (used in harl code More flexible training human_aware_rl#16)

… every object, add events_list field in trajectory
- Configuring recipes if needed (as they are used inside orders)
- Fixing save_trajectories methiod
- change outdated comment/variable name inside save_traj_as_json
Agent that is slowed down version of another agent.
Can be used to produce assymetry of power between agents.
- Prioritization of pots where there is less ingredients left for ideal recipe.
- Choosing next ingredient in the way that reveal intention for entire recipe.
- Tests for slowed down agents
- Tests for pedagogical ingredients picking
- Testing both GreedyHumanModel and SimpleGreedyHumanModel
- Standarization of ingredients list (sorted lisf of tuples) as method
- ingredients_diff (used in GreedyHumanModel)
- More tests
- one_hot_joint_action_encoding
- sparse_categorical_joint_action_encoding
- multi_hot_orders_encoding
- check for featurize_state if layout can be featurized
- gym space for every encoding method (besides action ones)
Probably nothing changed besides order_ids. This version was used in last tests
without this change there are false positives sometimes
@micahcarroll
Copy link
Member

Given that this is a lower priority issue right now, I'll temporarily close this PR for bookeeping – we can re-open it in the future if we are interested in exploring this direction further.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants