marl_transfer/main.py at master · sumitsk/marl_transfer · 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
import os
import json
import datetime
import numpy as np
import torch
import utils
import random
from copy import deepcopy
from arguments import get_args
from tensorboardX import SummaryWriter
from eval import evaluate
from learner import setup_master
from pprint import pprint


np.set_printoptions(suppress=True, precision=4)


def train(args, return_early=False):
    writer = SummaryWriter(args.log_dir)
    envs = utils.make_parallel_envs(args)
    master = setup_master(args)
    # used during evaluation only
    eval_master, eval_env = setup_master(args, return_env=True)
    obs = envs.reset() # shape - num_processes x num_agents x obs_dim
    master.initialize_obs(obs)
    n = len(master.all_agents)
    episode_rewards = torch.zeros([args.num_processes, n], device=args.device)
    final_rewards = torch.zeros([args.num_processes, n], device=args.device)

    # start simulations
    start = datetime.datetime.now()
    for j in range(args.num_updates):
        for step in range(args.num_steps):
            with torch.no_grad():
                actions_list = master.act(step)
            agent_actions = np.transpose(np.array(actions_list),(1,0,2))
            obs, reward, done, info = envs.step(agent_actions)
            reward = torch.from_numpy(np.stack(reward)).float().to(args.device)
            episode_rewards += reward
            masks = torch.FloatTensor(1-1.0*done).to(args.device)
            final_rewards *= masks
            final_rewards += (1 - masks) * episode_rewards
            episode_rewards *= masks

            master.update_rollout(obs, reward, masks)

        master.wrap_horizon()
        return_vals = master.update()
        value_loss = return_vals[:, 0]
        action_loss = return_vals[:, 1]
        dist_entropy = return_vals[:, 2]
        master.after_update()

        if j%args.save_interval == 0 and not args.test:
            savedict = {'models': [agent.actor_critic.state_dict() for agent in master.all_agents]}
            ob_rms = (None, None) if envs.ob_rms is None else (envs.ob_rms[0].mean, envs.ob_rms[0].var)
            savedict['ob_rms'] = ob_rms
            savedir = args.save_dir+'/ep'+str(j)+'.pt'
            torch.save(savedict, savedir)

        total_num_steps = (j + 1) * args.num_processes * args.num_steps

        if j%args.log_interval == 0:
            end = datetime.datetime.now()
            seconds = (end-start).total_seconds()
            mean_reward = final_rewards.mean(dim=0).cpu().numpy()
            print("Updates {} | Num timesteps {} | Time {} | FPS {}\nMean reward {}\nEntropy {:.4f} Value loss {:.4f} Policy loss {:.4f}\n".
                  format(j, total_num_steps, str(end-start), int(total_num_steps / seconds),
                  mean_reward, dist_entropy[0], value_loss[0], action_loss[0]))
            if not args.test:
                for idx in range(n):
                    writer.add_scalar('agent'+str(idx)+'/training_reward', mean_reward[idx], j)

                writer.add_scalar('all/value_loss', value_loss[0], j)
                writer.add_scalar('all/action_loss', action_loss[0], j)
                writer.add_scalar('all/dist_entropy', dist_entropy[0], j)

        if args.eval_interval is not None and j%args.eval_interval==0:
            ob_rms = (None, None) if envs.ob_rms is None else (envs.ob_rms[0].mean, envs.ob_rms[0].var)
            print('===========================================================================================')
            _, eval_perstep_rewards, final_min_dists, num_success, eval_episode_len = evaluate(args, None, master.all_policies,
                                                                                               ob_rms=ob_rms, env=eval_env,
                                                                                               master=eval_master)
            print('Evaluation {:d} | Mean per-step reward {:.2f}'.format(j//args.eval_interval, eval_perstep_rewards.mean()))
            print('Num success {:d}/{:d} | Episode Length {:.2f}'.format(num_success, args.num_eval_episodes, eval_episode_len))
            if final_min_dists:
                print('Final_dists_mean {}'.format(np.stack(final_min_dists).mean(0)))
                print('Final_dists_var {}'.format(np.stack(final_min_dists).var(0)))
            print('===========================================================================================\n')

            if not args.test:
                writer.add_scalar('all/eval_success', 100.0*num_success/args.num_eval_episodes, j)
                writer.add_scalar('all/episode_length', eval_episode_len, j)
                for idx in range(n):
                    writer.add_scalar('agent'+str(idx)+'/eval_per_step_reward', eval_perstep_rewards.mean(0)[idx], j)
                    if final_min_dists:
                        writer.add_scalar('agent'+str(idx)+'/eval_min_dist', np.stack(final_min_dists).mean(0)[idx], j)

            curriculum_success_thres = 0.9
            if return_early and num_success*1./args.num_eval_episodes > curriculum_success_thres:
                savedict = {'models': [agent.actor_critic.state_dict() for agent in master.all_agents]}
                ob_rms = (None, None) if envs.ob_rms is None else (envs.ob_rms[0].mean, envs.ob_rms[0].var)
                savedict['ob_rms'] = ob_rms
                savedir = args.save_dir+'/ep'+str(j)+'.pt'
                torch.save(savedict, savedir)
                print('===========================================================================================\n')
                print('{} agents: training complete. Breaking.\n'.format(args.num_agents))
                print('===========================================================================================\n')
                break

    writer.close()
    if return_early:
        return savedir

if __name__ == '__main__':
    args = get_args()
    if args.seed is None:
        args.seed = random.randint(0,10000)
    args.num_updates = args.num_frames // args.num_steps // args.num_processes
    torch.manual_seed(args.seed)
    torch.set_num_threads(1)
    np.random.seed(args.seed)
    if args.cuda:
        torch.cuda.manual_seed(args.seed)

    pprint(vars(args))
    if not args.test:
        with open(os.path.join(args.save_dir, 'params.json'), 'w') as f:
            params = deepcopy(vars(args))
            params.pop('device')
            json.dump(params, f)
    train(args)