-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtrain.py
executable file
·212 lines (179 loc) · 8.56 KB
/
train.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
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
"""
Script to train the PPO policy for dexterous grasping
Check scripts/train.sh for training scripts
"""
import os
from a2c_ppo_acktr.arguments import get_args
args = get_args()
import time
import shutil
import os.path as osp
from collections import deque
import numpy as np
import torch
from torch.utils.tensorboard import SummaryWriter
from a2c_ppo_acktr import algo, utils
from a2c_ppo_acktr.envs import make_vec_envs
from a2c_ppo_acktr.model import Policy
from a2c_ppo_acktr.storage import RolloutStorage
def main():
np.random.seed(args.seed)
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
if args.cuda and torch.cuda.is_available() and args.cuda_deterministic:
torch.backends.cudnn.benchmark = False
torch.backends.cudnn.deterministic = True
exp_dir = args.exp
log_dir = osp.join(exp_dir, 'monitor')
tb_dir = osp.join(exp_dir, 'logs')
save_dir = osp.join(exp_dir, 'models')
os.makedirs(tb_dir, exist_ok=True)
os.makedirs(save_dir, exist_ok=True)
utils.cleanup_log_dir(log_dir)
writer = SummaryWriter(tb_dir)
# copy important codes into exp dir
code_dir = osp.join(exp_dir, 'codes')
os.makedirs(code_dir, exist_ok=True)
with open(osp.join(code_dir, 'args.txt'), 'w') as f:
f.write(str(args)+'\n')
curr_dir = osp.dirname(osp.abspath(__file__))
shutil.copy(osp.join(curr_dir, 'train.py'), code_dir)
shutil.copy(osp.join(curr_dir, 'a2c_ppo_acktr/arguments.py'), code_dir)
shutil.copy(osp.join(curr_dir, 'a2c_ppo_acktr/envs.py'), code_dir)
shutil.copy(osp.join(curr_dir, 'a2c_ppo_acktr/model.py'), code_dir)
shutil.copy(osp.join(curr_dir, 'envs/mj_envs/dex_manip/graff.py'), code_dir)
torch.set_num_threads(1)
device = torch.device("cuda" if args.cuda else "cpu")
grasp_attrs_dict = {'dataset': args.dataset,
'obj': args.obj,
'policy': args.policy,
'cnn_arch': args.cnn_arch,
'noise': args.noise,
'inputs': args.inputs,
'cameras': args.cameras,
'img_res': args.img_res,
'rewards': args.rewards,
'reward_dst_thr': args.reward_dst_thr,
'obj_mass': args.obj_mass,
'obj_rot': args.obj_rot,
'obj_tr': args.obj_tr,
'gravity': args.gravity,
'debug': args.debug}
envs = make_vec_envs(args.env_name, args.seed, args.num_processes,
args.gamma, log_dir, device, int(args.gpu_env), False, dataset=args.dataset,
object=args.obj, grasp_attrs_dict=grasp_attrs_dict)
if args.load_model == 0:
actor_critic = Policy(
envs.observation_space,
envs.action_space,
policy=args.policy,
cnn_args={'arch': args.cnn_arch,
'pretrained': args.cnn_pretrained,
'cameras': args.cameras},
base_kwargs={'recurrent': args.recurrent_policy})
start_update_num = 0
else:
from a2c_ppo_acktr.utils import get_vec_normalize
actor_critic, ob_rms = \
torch.load(os.path.join(args.exp, 'models', str(args.load_model) + ".pt"))
start_update_num = args.load_model + 1
vec_norm = get_vec_normalize(envs)
if vec_norm is not None:
vec_norm.eval()
vec_norm.ob_rms = ob_rms
actor_critic.to(device)
agent = algo.PPO(
actor_critic,
args.clip_param,
args.ppo_epoch,
args.num_mini_batch,
args.value_loss_coef,
args.entropy_coef,
lr=args.lr,
eps=args.eps,
max_grad_norm=args.max_grad_norm)
rollouts = RolloutStorage(args.num_steps, args.num_processes,
envs.observation_space, envs.action_space,
actor_critic.recurrent_hidden_state_size)
obs = envs.reset()
rollouts.obs.copy_(0, obs)
rollouts.to(device)
start = time.time()
num_updates = int(args.num_env_steps) // args.num_steps // args.num_processes
train_rewards = deque(maxlen=10)
episode_successes = deque(maxlen=10)
episode_successes_orig = deque(maxlen=10)
best_ep_rews = -np.inf
for j in range(start_update_num, num_updates):
if args.use_linear_lr_decay:
# decrease learning rate linearly
utils.update_linear_schedule(
agent.optimizer, j, num_updates, args.lr)
for step in range(args.num_steps):
# Sample actions
with torch.no_grad():
value, action, action_log_prob, recurrent_hidden_states = actor_critic.act(
rollouts.obs[step], rollouts.recurrent_hidden_states[step],
rollouts.masks[step])
# Obser reward and next obs
obs, reward, done, infos = envs.step(action)
for info in infos:
if 'episode' in info.keys():
train_rewards.append(info['episode']['r'])
episode_successes.append(info['episode']['obj_lift'])
episode_successes_orig.append(info['episode']['obj_grab'])
# If done then clean the history of observations.
masks = torch.FloatTensor(
[[0.0] if done_ else [1.0] for done_ in done])
bad_masks = torch.FloatTensor(
[[0.0] if 'bad_transition' in info.keys() else [1.0]
for info in infos])
rollouts.insert(obs, recurrent_hidden_states, action,
action_log_prob, value, reward, masks, bad_masks)
with torch.no_grad():
next_value = actor_critic.get_value(
rollouts.obs[-1], rollouts.recurrent_hidden_states[-1],
rollouts.masks[-1]).detach()
rollouts.compute_returns(next_value, args.use_gae, args.gamma,
args.gae_lambda, args.use_proper_time_limits)
value_loss, action_loss, dist_entropy = agent.update(rollouts)
rollouts.after_update()
if j % args.log_interval == 0 and len(train_rewards) > 1:
total_num_steps = (j + 1) * args.num_processes * args.num_steps
end = time.time()
print(
"Updates {}, num timesteps {}, FPS {} \n Last {} training episodes: mean/median reward {:.1f}/{:.1f}, min/max reward {:.1f}/{:.1f}\n"
.format(j, total_num_steps,
int(total_num_steps / (end - start)),
len(train_rewards), np.mean(train_rewards),
np.median(train_rewards), np.min(train_rewards),
np.max(train_rewards), dist_entropy, value_loss,
action_loss))
writer.add_scalar('Loss/total', value_loss * args.value_loss_coef + action_loss -
dist_entropy * args.entropy_coef, total_num_steps)
writer.add_scalar('Loss/value', value_loss, total_num_steps)
writer.add_scalar('Loss/action', action_loss, total_num_steps)
writer.add_scalar('Loss/entropy', dist_entropy, total_num_steps)
writer.add_scalar('Rewards/mean', np.mean(train_rewards), total_num_steps)
writer.add_scalar('Rewards/median', np.median(train_rewards), total_num_steps)
writer.add_scalar('Rewards/max', np.max(train_rewards), total_num_steps)
writer.add_scalar('Rewards/min', np.min(train_rewards), total_num_steps)
writer.add_scalar('Success_rate/hand-obj-notable', np.mean(episode_successes), total_num_steps)
writer.add_scalar('Success_rate/obj-notable', np.mean(episode_successes_orig), total_num_steps)
# save model for every interval-th episode or for the last epoch
if ((j + 1) % args.save_interval == 0 or j == num_updates - 1):
# save model
torch.save([
actor_critic,
getattr(utils.get_vec_normalize(envs), 'ob_rms', None)
], os.path.join(save_dir, str(j + 1) + ".pt"))
# save best model if available
if np.mean(train_rewards) > best_ep_rews:
print('Best model found. Saving.')
best_ep_rews = np.mean(train_rewards)
torch.save([
actor_critic,
getattr(utils.get_vec_normalize(envs), 'ob_rms', None)
], os.path.join(save_dir, "best.pt"))
if __name__ == "__main__":
main()