scripted baseline agents

scripts/eval_oc/scripted_overcooked.py

@@ -0,0 +1,113 @@
+from datetime import datetime
+import itertools
+import math
+from pathlib import Path
+import numpy as np
+import pickle
+from src.agent.albatross import AlbatrossAgent, AlbatrossAgentConfig
+from src.agent.initialization import get_agent_from_config
+from src.agent.one_shot import NetworkAgent, NetworkAgentConfig, bc_agent_from_file
+from src.agent.scripted import PlaceDishEverywhereAgentConfig, PlaceOnionAgentConfig, PlaceOnionDeliverAgentConfig, PlaceOnionEverywhereAgentConfig
+from src.game.overcooked.config import AsymmetricAdvantageOvercookedConfig, CoordinationRingOvercookedConfig, \
+    CounterCircuitOvercookedConfig, CrampedRoomOvercookedConfig, ForcedCoordinationOvercookedConfig, OvercookedRewardConfig
+from src.game.overcooked.overcooked import OvercookedGame
+from src.misc.utils import set_seed
+from src.network.initialization import get_network_from_file
+from src.trainer.az_evaluator import do_evaluation
+
+
+def eval_scripted_oc(seed: int):
+    print(f'{datetime.now()} - Started eval script', flush=True)
+    save_path = Path(__file__).parent.parent.parent / 'a_data' / 'scripted'
+    game_cfg, prefix = AsymmetricAdvantageOvercookedConfig(), 'aa'
+    # game_cfg, prefix = CoordinationRingOvercookedConfig(), 'co'
+    # game_cfg, prefix = CounterCircuitOvercookedConfig(), 'cc'
+
+    # init scripted agent
+    scripted_cfg = PlaceOnionAgentConfig()
+    # scripted_cfg = PlaceOnionDeliverAgentConfig()
+    # scripted_cfg = PlaceOnionEverywhereAgentConfig()
+    # scripted_cfg = PlaceDishEverywhereAgentConfig()
+
+    scripted_agent = get_agent_from_config(scripted_cfg)
+
+    # fname = f'{prefix}_dish_everywhere_{seed}.pkl'
+    fname = f'tmp.pkl'
+    print(f"{fname=}", flush=True)
+
+    set_seed(seed)
+
+    net_path = Path(__file__).parent.parent.parent / 'a_saved_runs' / 'overcooked'
+    proxy_path = net_path / f'proxy_{prefix}_{seed}' / 'latest.pt'
+    resp_path = net_path / f'resp_{prefix}_{seed}' / 'latest.pt'
+
+    net = get_network_from_file(resp_path).eval()
+    alb_network_agent_cfg = NetworkAgentConfig(
+        net_cfg=net.cfg,
+        temperature_input=True,
+        single_temperature=False,
+        init_temperatures=[0, 0],
+    )
+    alb_online_agent_cfg = AlbatrossAgentConfig(
+        num_player=2,
+        agent_cfg=alb_network_agent_cfg,
+        device_str='cpu',
+        response_net_path=str(resp_path),
+        proxy_net_path=str(proxy_path),
+        noise_std=None,
+        # fixed_temperatures=[9, 9],
+        num_samples=1,
+        init_temp=0,
+        # num_likelihood_bins=int(2e3),
+        # sample_from_likelihood=True,
+    )
+    alb_online_agent = AlbatrossAgent(alb_online_agent_cfg)
+
+    reward_cfg = OvercookedRewardConfig(
+        placement_in_pot=0,
+        dish_pickup=0,
+        soup_pickup=0,
+        soup_delivery=20,
+        start_cooking=0,
+    )
+    game_cfg.reward_cfg = reward_cfg
+    game_cfg.temperature_input = True
+    game_cfg.single_temperature_input = True
+    game_cfg.automatic_cook_start = False
+    game = OvercookedGame(game_cfg)
+
+    print(f'{datetime.now()} - Started evaluation of {prefix} with {seed=}', flush=True)
+    results, _ = do_evaluation(
+        game=game,
+        evaluee=scripted_agent,
+        opponent_list=[alb_online_agent],
+        num_episodes=[100],
+        enemy_iterations=0,
+        temperature_list=[0.5],
+        own_temperature=1,
+        prevent_draw=False,
+        switch_positions=False,
+        verbose_level=1,
+    )
+    with open(save_path / fname, 'wb') as f:
+        pickle.dump(results, f)
+
+
+def compute_avg():
+    path = Path(__file__).parent.parent.parent / 'a_data' / 'scripted'
+    res_list = []
+    for seed in range(5):
+        with open(path / f'cc_dish_everywhere_{seed}.pkl', 'rb') as f:
+            res = pickle.load(f)
+        res_list.append(res)
+    full_arr = np.asarray(res_list)[:, 0]
+    arr = full_arr.mean(axis=-1)
+    print(arr)
+    print(f"{arr.mean()=}")
+    print(f"{arr.std()=}")
+
+
+
+if __name__ == '__main__':
+    # eval_scripted_oc(0)
+    compute_avg()

scripts/temp_est/matrix_game.py

@@ -0,0 +1,69 @@
+
+
+import numpy as np
+from src.equilibria.logit import compute_logit_equilibrium
+from src.modelling.mle import compute_likelihood, compute_temperature_mle
+
+
+def main_matrix_game_repeated():
+    # compute le with temperature 10 as ground truth best play
+    ja_vals = np.asarray([[4, 4], [0, 0], [1, 1], [2, 2]])
+
+    val, pol, err = compute_logit_equilibrium(
+        available_actions=[[0, 1], [0, 1]],
+        joint_action_list=[(0, 0), (0, 1), (1, 0), (1, 1)],
+        joint_action_value_arr=ja_vals,
+        num_iterations=int(1e6),
+        epsilon=0,
+        temperatures=[10, 10],
+    )
+    p1_logit_pol = pol[0]
+    q_a = p1_logit_pol[0] * ja_vals[0, 1] + p1_logit_pol[1] * ja_vals[2, 1]
+    q_b = p1_logit_pol[0] * ja_vals[1, 1] + p1_logit_pol[1] * ja_vals[3, 1]
+
+    p2_actions = [0, 1, 0, 0]
+    utils = [[q_a, q_b] for _ in range(len(p2_actions))]
+
+    cur_temperature = compute_temperature_mle(
+        min_temp=-10,
+        max_temp=10,
+        num_iterations=20,
+        chosen_actions=p2_actions,
+        utilities=utils,
+    )
+
+    val_log, pol_log, err = compute_logit_equilibrium(
+        available_actions=[[0, 1], [0, 1]],
+        joint_action_list=[(0, 0), (0, 1), (1, 0), (1, 1)],
+        joint_action_value_arr=ja_vals,
+        num_iterations=int(1e6),
+        epsilon=0,
+        temperatures=[cur_temperature, cur_temperature],
+    )
+    q_a_p1 = pol_log[1][0] * ja_vals[0, 1] + pol_log[1][1] * ja_vals[1, 1]
+    q_b_p1 = pol_log[1][0] * ja_vals[2, 1] + pol_log[1][1] * ja_vals[3, 1]
+
+    a = 1
+
+    # for t in range(1, num_steps + 1):
+    #     utils = [[q_a, q_b] for _ in range(t)]
+
+    #     p2_actions = [0 for _ in range(t)]
+
+    #     cur_temperature = compute_temperature_mle(
+    #         min_temp=0,
+    #         max_temp=10,
+    #         num_iterations=20,
+    #         chosen_actions=p2_actions,
+    #         utilities=utils,
+    #     )
+    #     all_estimates.append(cur_temperature)        
+
+    a = 1
+
+
+
+
+
+if __name__ == '__main__':
+    main_matrix_game_repeated()

src/agent/initialization.py

@@ -4,6 +4,7 @@
     NetworkAgentConfig, LegalRandomAgent, LegalRandomAgentConfig, BCNetworkAgent, BCNetworkAgentConfig
 from src.agent.overcooked import GreedyHumanOvercookedAgent, GreedyHumanOvercookedAgentConfig
 from src.agent.planning import AStarAgent, AStarAgentConfig
+from src.agent.scripted import PlaceDishEverywhereAgent, PlaceDishEverywhereAgentConfig, PlaceOnionAgent, PlaceOnionAgentConfig, PlaceOnionDeliverAgent, PlaceOnionDeliverAgentConfig, PlaceOnionEverywhereAgent, PlaceOnionEverywhereAgentConfig
 from src.agent.search_agent import SearchAgent, LookaheadAgent, SearchAgentConfig, LookaheadAgentConfig, \
     DoubleSearchAgent, DoubleSearchAgentConfig
 
@@ -29,6 +30,14 @@ def get_agent_from_config(agent_cfg: AgentConfig) -> Agent:
         return AlbatrossAgent(agent_cfg)
     elif isinstance(agent_cfg, BCNetworkAgentConfig):
         return BCNetworkAgent(agent_cfg)
+    elif isinstance(agent_cfg, PlaceOnionAgentConfig):
+        return PlaceOnionAgent(agent_cfg)
+    elif isinstance(agent_cfg, PlaceOnionEverywhereAgentConfig):
+        return PlaceOnionEverywhereAgent(agent_cfg)
+    elif isinstance(agent_cfg, PlaceDishEverywhereAgentConfig):
+        return PlaceDishEverywhereAgent(agent_cfg)
+    elif isinstance(agent_cfg, PlaceOnionDeliverAgentConfig):
+        return PlaceOnionDeliverAgent(agent_cfg)
     else:
         raise ValueError(f"Unknown agent type: {agent_cfg}")