-
Notifications
You must be signed in to change notification settings - Fork 1
/
tsrl_train.py
56 lines (51 loc) · 2.06 KB
/
tsrl_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
import os
import sys
sys.path.append(os.getcwd())
from turtle import done
import wandb
import argparse
from tsrl_algos import TSRL
import datetime
def main():
wandb.init(project="your project name", entity="your entity id")
# Parameters
parser = argparse.ArgumentParser()
parser.add_argument('--device', default='cuda', help='cuda or cpu')
parser.add_argument('--env_name', default='hopper-medium-v2', help='choose your mujoco env')
parser.add_argument('--ratio', default=1, type=float, help='choose the data ratio')
parser.add_argument('--alpha', default=2.5, type=float)
parser.add_argument('--gamma', default=0.99, type=float)
parser.add_argument("--seed", default=111, type=int)
parser.add_argument("--num_hidden", default=512, type=int)
parser.add_argument('--lr_actor', default=3e-4, type=float)
parser.add_argument('--lr_critic', default=3e-4, type=float)
parser.add_argument('--z_act_weight', default=0.0, type=float)
parser.add_argument('--inconsis_weight', default=0.0, type=float)
parser.add_argument('--quantile', default=0.0, type=float)
parser.add_argument('--drop_prob', default=0.1, type=float)
parser.add_argument('--eval_iter', default=5, type=int)
parser.add_argument('--augment', default=True, type=bool)
parser.add_argument('--store_prams', default=False, type=bool)
args = parser.parse_args()
wandb.config.update(args)
# setup mujoco environment
env_name = args.env_name
wandb.run.name = f"TSRL-alpha_{args.alpha}-{env_name}-seed_{args.seed}-ratio_{args.ratio}"
agent_tsrl = TSRL(env_name,
device=args.device,
ratio=args.ratio,
gamma=args.gamma,
alpha=args.alpha,
num_hidden=args.num_hidden,
z_act_weight=args.z_act_weight,
inconsis_weight = args.inconsis_weight,
quantile=args.quantile,
drop_prob=args.drop_prob,
store_prams=args.store_prams,
augment=args.augment,
eval_iter=args.eval_iter,
seed=args.seed,
)
agent_tsrl.tsrl_learn(total_time_step=int(1e+6))
if __name__ == '__main__':
main()