Here is a list of existing Parity tests we have.
Canonically, the Agent 1
agent type is constant in each experiment while Agent 2
changes.
yaml | Agent1 | Agent2 | Environment | Outcome |
---|---|---|---|---|
ipd/naive |
Naive |
Any |
Finite |
Agent2: TitForTat -> ALL-C; [Naive, Defect, Altruistic] -> ALL-D |
ipd/PPO |
PPO |
Any |
Finite |
Agent2: TitForTat -> ALL-C; [PPO, Defect, Altruistic] -> ALL-D |
ipd/PPO_memory |
PPO_memory |
Any |
Finite |
Agent2: TitForTat -> ALL-C; [PPO_memory, Defect, Altruistic] -> ALL-D |
ipd/mfos_fixed |
PPO |
Fixed |
Infinite |
Agent2: TitForTat -> ALL-C; [Defect, Altruistic] -> ALL-D |
ipd/mfos_nl |
PPO |
Naive |
Infinite |
Agent2: Naive -> ZD-Extortion (Player 1 has a payoff greater than CC) |
ipd/marl2_fixed |
PPO_memory |
Fixed |
Meta |
Agent2: TitForTat -> ALL-C; [Defect, Altruistic] -> ALL-D |
ipd/marl2_nl |
PPO_memory |
Naive |
Meta |
Agent2: Naive -> Mixture of Cooperation and Defection |
ipd/earl_fixed |
PPO_memory |
Fixed |
Meta |
Agent2: TitForTat -> ALL-C; [Defect, Altruistic] -> ALL-D |
ipd/earl_nl_cma |
PPO_memory |
Naive |
Meta |
Agent2: Naive -> ZD-Extortion |
ipd/earl_nl_open |
PPO_memory |
Naive |
Meta |
Agent2: Naive -> ZD-Extortion |
ipd/earl_nl_pgpe |
PPO_memory |
Naive |
Meta |
Agent2: Naive -> TBD |