1from argparse import ArgumentParser
2
3import wandb
4
5from amago.envs.builtin.metaworld_ml import Metaworld
6from amago.nets.tstep_encoders import FFTstepEncoder
7from amago.envs.exploration import BilevelEpsilonGreedy
8from amago.cli_utils import *
9
10
11def add_cli(parser):
12 parser.add_argument(
13 "--benchmark",
14 type=str,
15 default="reach-v2",
16 help="`name-v2` for ML1, or `ml10`/`ml45`",
17 )
18 parser.add_argument("--k", type=int, default=3, help="K-Shots")
19 parser.add_argument("--max_seq_len", type=int, default=256)
20 parser.add_argument(
21 "--hide_rl2s",
22 action="store_true",
23 help="hides the 'rl2 info' (previous actions, rewards)",
24 )
25 return parser
26
27
28if __name__ == "__main__":
29 parser = ArgumentParser()
30 add_common_cli(parser)
31 add_cli(parser)
32 args = parser.parse_args()
33
34 config = {
35 "amago.nets.tstep_encoders.FFTstepEncoder.hide_rl2s": args.hide_rl2s,
36 # delete the next three lines to use the paper settings, which were
37 # intentionally left wide open to avoid reward-specific tuning.
38 "amago.nets.actor_critic.NCriticsTwoHot.min_return": -100.0,
39 "amago.nets.actor_critic.NCriticsTwoHot.max_return": 5000 * args.k,
40 "amago.nets.actor_critic.NCriticsTwoHot.output_bins": 96,
41 }
42 traj_encoder_type = switch_traj_encoder(
43 config,
44 arch=args.traj_encoder,
45 memory_size=args.memory_size,
46 layers=args.memory_layers,
47 )
48 agent_type = switch_agent(
49 config, args.agent_type, reward_multiplier=1.0, num_critics=4
50 )
51 exploration_type = switch_exploration(
52 config, "bilevel", steps_anneal=2_000_000, rollout_horizon=args.k * 500
53 )
54 use_config(config, args.configs)
55
56 make_train_env = lambda: Metaworld(args.benchmark, "train", k_episodes=args.k)
57 make_test_env = lambda: Metaworld(args.benchmark, "test", k_episodes=args.k)
58
59 group_name = (
60 f"{args.run_name}_metaworld_{args.benchmark}_K_{args.k}_L_{args.max_seq_len}"
61 )
62 for trial in range(args.trials):
63 run_name = group_name + f"_trial_{trial}"
64 experiment = create_experiment_from_cli(
65 args,
66 make_train_env=make_train_env,
67 make_val_env=make_train_env,
68 max_seq_len=args.max_seq_len,
69 traj_save_len=min(500 * args.k + 1, args.max_seq_len * 4),
70 group_name=group_name,
71 run_name=run_name,
72 tstep_encoder_type=FFTstepEncoder,
73 traj_encoder_type=traj_encoder_type,
74 agent_type=agent_type,
75 val_timesteps_per_epoch=15 * args.k * 500 + 1,
76 learning_rate=5e-4,
77 grad_clip=2.0,
78 exploration_wrapper_type=exploration_type,
79 )
80
81 experiment = switch_async_mode(experiment, args.mode)
82 experiment.start()
83 if args.ckpt is not None:
84 experiment.load_checkpoint(args.ckpt)
85 experiment.learn()
86 experiment.evaluate_test(make_test_env, timesteps=20_000, render=False)
87 experiment.delete_buffer_from_disk()
88 wandb.finish()
