-
Notifications
You must be signed in to change notification settings - Fork 5
/
train_sac.py
192 lines (156 loc) · 7.22 KB
/
train_sac.py
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
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import sys
import time
import shutil
from offpolicy.video import VideoRecorder
from offpolicy.logger import Logger
from offpolicy.replay_buffer import ReplayBuffer
from offpolicy.agent.sac import SACAgent
import offpolicy.utils as utils
from crowd_nav.configs.config import Config
import crowd_sim
import os
from collections import deque
class Workspace(object):
def __init__(self):
self.config = Config()
utils.set_seed_everywhere(self.config.env.seed)
env_name = self.config.env.env_name
task = self.config.env.task
policy_name = self.config.robot.policy + '_sac'
self.output_dir = os.path.join(self.config.training.output_dir, task, policy_name)
# save policy to output_dir
if os.path.exists(self.output_dir) and self.config.training.overwrite: # if I want to overwrite the directory
shutil.rmtree(self.output_dir) # delete an entire directory tree
if not os.path.exists(self.output_dir):
os.makedirs(self.output_dir)
shutil.copytree('crowd_nav/configs', os.path.join(self.output_dir, 'configs'))
self.model_dir = os.path.join(self.output_dir, 'checkpoints')
if not os.path.exists(self.model_dir):
os.makedirs(self.model_dir)
self.work_dir = self.output_dir
print(f'workspace: {self.work_dir}')
self.logger = Logger(self.work_dir,
save_tb=True,
log_frequency=10000,
agent='sac')
self.device = torch.device("cuda" if self.config.training.cuda and torch.cuda.is_available() else "cpu")
self.env = utils.make_env(self.config)
self.eval_env = utils.make_eval_env(self.config)
obs_shape = self.env.observation_space.spaces
action_shape = self.env.action_space.shape
self.agent = SACAgent(self.config, obs_shape, action_shape, self.device)
self.replay_buffer = ReplayBuffer(obs_shape,
action_shape,
int(self.config.sac.num_train_steps),
self.device)
self.video_recorder = VideoRecorder(
self.work_dir if self.config.sac.save_video else None)
self.step = 0
self.interaction = 0
self.max_success_rate = 0.
def evaluate(self):
success = 0
collision = 0
timeout = 0
average_episode_reward = 0
for episode in range(self.config.sac.num_eval_episodes):
self.eval_env.case_counter['test'] = 0
obs = self.eval_env.reset()
self.agent.reset()
done = False
episode_reward = 0
while not done:
with utils.eval_mode(self.agent):
action = self.agent.act(obs, sample=False)
obs, reward, done, info = self.eval_env.step(action)
done = done[0]
episode_reward += reward[0]
average_episode_reward += episode_reward
status = str(info['info'])
if status == 'Reaching goal':
success += 1
elif status == 'Collision':
collision += 1
elif status == 'Timeout':
timeout += 1
average_episode_reward /= self.config.sac.num_eval_episodes
success_rate = success / self.config.sac.num_eval_episodes
collision_rate = collision / self.config.sac.num_eval_episodes
timeout_rate = timeout / self.config.sac.num_eval_episodes
if success_rate > self.max_success_rate:
self.max_success_rate = success_rate
torch.save(self.agent.actor.state_dict(), os.path.join(self.model_dir, './best_sac_actor.pt'))
print('eval', average_episode_reward)
self.logger.log('eval/episode_reward', average_episode_reward,
self.step)
self.logger.log('eval/success_rate', success_rate,
self.step)
self.logger.log('eval/collision_rate', collision_rate,
self.step)
self.logger.log('eval/timeout_rate', timeout_rate,
self.step)
self.logger.dump(self.step)
def run(self):
episode, episode_step, episode_reward, done = 0, 0, 0, True
start_time = time.time()
episode_rewards = deque(maxlen=100)
reward_list = []
while self.step < self.config.sac.num_train_steps:
if done:
if self.step > 0:
self.logger.log('train/duration',
time.time() - start_time, self.step)
start_time = time.time()
self.logger.dump(
self.step, save=(self.step > self.config.sac.num_seed_steps))
if self.interaction > self.config.sac.save_interval:
self.interaction = 0
filename = 'sac_actor' + str(self.step) + '.pt'
torch.save(self.agent.actor.state_dict(), os.path.join(self.model_dir, filename))
np.save(os.path.join(self.output_dir, 'reward.npy'), reward_list)
self.logger.log('eval/episode', episode, self.step)
self.evaluate()
self.logger.log('train/episode_reward', episode_reward,
self.step)
if self.step >= self.config.sac.num_seed_steps:
episode_rewards.append(episode_reward)
reward_list.append(np.mean(episode_rewards))
print('%d/%d, %d, %f' % (self.step, self.config.sac.num_train_steps, episode_step, np.mean(episode_rewards)))
obs = self.env.reset()
self.agent.reset()
done = False
episode_reward = 0
episode_step = 0
episode += 1
self.logger.log('train/episode', episode, self.step)
# sample action for data collection
if self.step < self.config.sac.num_seed_steps:
action = self.env.action_space.sample()
action = utils.clip_action(action, clip_norm=True, max_norm=self.config.robot.v_pref)
else:
with utils.eval_mode(self.agent):
action = self.agent.act(obs, sample=True)
action = utils.clip_action(action, clip_norm=True, max_norm=self.config.robot.v_pref)
# run training update
if self.step >= self.config.sac.num_seed_steps:
self.agent.update(self.replay_buffer, self.logger, self.step)
next_obs, reward, done, info = self.env.step(action)
# allow infinite bootstrap
done = float(done[0])
done_no_max = done
episode_reward += reward[0]
self.replay_buffer.add(obs, action, reward, next_obs, done,
done_no_max)
obs = next_obs
episode_step += 1
self.step += 1
self.interaction += 1
def main():
workspace = Workspace()
workspace.run()
if __name__ == '__main__':
main()