COIN/coin_eval.py at master · marmotlab/COIN · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
import torch
import argparse
import numpy as np

from utils import set_eval_env, save_as_json
from models.coin import TD3

if __name__ == "__main__":

    parser = argparse.ArgumentParser()
    parser.add_argument("--policy", default="TD3")  # Policy name (TD3)
    parser.add_argument("--env", default="intersection")  # MetaDrive environment name
    parser.add_argument("--seed", default=0, type=int)  # Sets Gym, PyTorch and Numpy seeds
    parser.add_argument("--hidden_size", default=256, type=int, help='The number of hidden units')
    parser.add_argument("--expl_noise", default=0.01, type=float)  # Std of Gaussian exploration noise
    parser.add_argument("--discount", default=0.99, type=float)  # Discount factor
    parser.add_argument("--tau", default=0.005, type=float)  # Target network update rate
    parser.add_argument("--policy_noise", default=0.2)  # Noise added to target policy during critic update
    parser.add_argument("--noise_clip", default=0.5)  # Range to clip target policy noise
    parser.add_argument("--policy_freq", default=2, type=int)  # Frequency of delayed policy updates
    parser.add_argument("--model_path", default="", required=True)  # Model load file name, "" doesn't load, "default" uses file_name
    parser.add_argument("--num_eval", default=20, type=int)  # Number of episodes for evaluation
    parser.add_argument("--num_agents", default=30, type=int) # Number of agents
    parser.add_argument("--gui", default=True, type=bool)  # Activate Metadrive gui window
    parser.add_argument("--generate_gif", default=False, type=bool) # Whether to save gifs
    args = parser.parse_args()

    print("---------------------------------------")
    print("Policy: {}, Env: {}, Seed: {}, Num Agents: {}".format(args.policy, args.env, args.seed, args.num_agents))
    print("---------------------------------------")

    env = set_eval_env(args.env, args.seed, args.num_agents)
    # Set seeds
    env.seed(args.seed)
    np.random.seed(args.seed)
    torch.manual_seed(args.seed)

    state_dim = 10
    obs_dim = env.observation_space['agent0'].shape[0]
    action_dim = env.action_space['agent0'].shape[0]
    max_action = float(env.action_space['agent0'].high[0])

    kwargs = {
        "obs_dim": obs_dim,
        "state_dim": state_dim,
        "action_dim": action_dim,
        "max_action": max_action,
        "discount": args.discount,
        "tau": args.tau,
    }

    # Initialize policy
    if args.policy == "TD3":
        # Target policy smoothing is scaled wrt the action scale
        kwargs["policy_noise"] = args.policy_noise * max_action
        kwargs["noise_clip"] = args.noise_clip * max_action
        kwargs["policy_freq"] = args.policy_freq
        kwargs['hidden_dim'] = args.hidden_size
        policy = TD3(**kwargs)
    else:
        raise NotImplementedError

    # load last
    # file_name = "{}_{}_{}".format(args.policy, args.env, args.seed)
    # policy_file = file_name
    # policy.load(args.model_path + "/model/" + policy_file)

    # load best
    policy.load_best(args.model_path + "/model")

    eval_metrics = dict(avg_arr_dest=[], avg_crash_veh=[], avg_run_out=[], avg_effi=[], avg_reach_step=[],avg_safety=[])
    for i in range(int(args.num_eval)):
        env.close()
        seed = 1000 + i * 100
        env = set_eval_env(args.env, seed, args.num_agents)
        # Reset environment
        obs_n, _ = env.reset()
        if args.env == 'intersection':
            env.render(mode="top_down", num_stack=25, camera_position=(75, 5), screen_record=True, window=args.gui)
        elif args.env == 'roundabout':
            env.render(mode="top_down", num_stack=25, camera_position=(105, 5), scaling=3.6, screen_record=True,
                        window=args.gui)
        elif args.env == 'bottleneck':
            env.render(mode='top_down', num_stack=25, camera_position=(95, 5), screen_record=True, window=args.gui)

        episode_reward = 0
        episode_timesteps = 0
        episode_crash_veh = 0
        episode_arr_dest = 0
        episode_run_out = 0
        episode_efficiency = 0
        episode_veh_step_dict = dict()
        episode_veh_finish_step_dict = dict()
        while True:
            episode_timesteps += 1
            a_n = dict()
            # Select according to policy
            for agent, obs in obs_n.items():
                a_n[agent] = (policy.select_action(np.array(obs)) +
                                np.random.normal(0, max_action * args.expl_noise,
                                                size=action_dim)).clip(-max_action, max_action)
                # a_n[agent] = (policy.select_action(np.array(obs))).clip(-max_action, max_action)

            # Perform action
            next_obs_n, r_n, d_n, _, info = env.step(a_n)
            if args.env == 'intersection':
                env.render(mode="top_down", num_stack=25, camera_position=(75, 5), screen_record=True,
                            window=args.gui)
            elif args.env == 'roundabout':
                env.render(mode="top_down", num_stack=25, camera_position=(105, 5), scaling=3.6, screen_record=True,
                            window=args.gui)
            elif args.env == 'bottleneck':
                env.render(mode='top_down', num_stack=25, camera_position=(95, 5), screen_record=True,
                            window=args.gui)

            for agent in list(obs_n.keys()):
                # Remove the observations of finished agents
                if d_n[agent]:
                    del next_obs_n[agent]
                    # Count the trajectory length of vehicles that not collide
                    if info[agent]['arrive_dest']:
                        episode_veh_finish_step_dict[agent] = episode_veh_step_dict[agent]
                    else:
                        episode_veh_finish_step_dict[agent] = env.config['horizon']

                # Record episode metrics
                if agent not in episode_veh_step_dict:
                    episode_veh_step_dict[agent] = 1
                else:
                    episode_veh_step_dict[agent] += 1
                if info[agent]['crash_vehicle']:
                    episode_crash_veh += 1
                if info[agent]['out_of_road']:
                    episode_run_out += 1
                if info[agent]['arrive_dest']:
                    episode_arr_dest += 1

            obs_n = next_obs_n
            episode_reward += sum(r_n.values())

            if d_n['__all__']:
                episode_num_veh = len(episode_veh_step_dict)
                episode_avg_reward = episode_reward / episode_num_veh
                episode_avg_crash_veh = episode_crash_veh / episode_num_veh
                episode_avg_run_out_veh = episode_run_out / episode_num_veh
                episode_avg_arr_des = episode_arr_dest / episode_num_veh
                episode_avg_efficiency = (episode_arr_dest - episode_crash_veh - episode_run_out) / env.config['horizon']
                episode_avg_safety = -(episode_run_out+episode_crash_veh)

                if len(episode_veh_finish_step_dict) > 0:
                    episode_avg_reach_step = np.mean(list(episode_veh_finish_step_dict.values()))
                else:
                    episode_avg_reach_step = env.config['horizon']
                # +1 to account for 0 indexing. +0 on ep_timesteps since it will increment +1 even if done=True
                print("Episode Num: {} Episode T: {} Success: {:.3f}".format(i,
                                                                                episode_timesteps,
                                                                                episode_avg_arr_des))

                # Save evaluation metrics
                eval_metrics['avg_arr_dest'].append(episode_avg_arr_des)
                eval_metrics['avg_crash_veh'].append(episode_avg_crash_veh)
                eval_metrics['avg_run_out'].append(episode_avg_run_out_veh)
                eval_metrics['avg_effi'].append(episode_avg_efficiency)
                eval_metrics['avg_reach_step'].append(episode_avg_reach_step)
                eval_metrics['avg_safety'].append(episode_avg_safety)

                # Save gifs
                if args.generate_gif:
                    env.top_down_renderer.generate_gif(args.model_path + '/gifs/epi{}_success{:.2f}.gif'.format(i, episode_avg_arr_des))

                break

    save_as_json(args.model_path+'/result/{}_{}_{}.json'.format(args.policy, args.num_agents, args.env), eval_metrics)
    print("Average Success Rate: {}\n".format(np.mean(eval_metrics['avg_arr_dest'])))
    env.close()