-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
81 lines (67 loc) · 4.12 KB
/
main.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
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
import pybullet as p
import pybullet_data
import time
import argparse
import os
from train import Trainer
from network import DQNetwork
from data import ReplayMemory
from environment import CarEnvironment
parser = argparse.ArgumentParser()
parser.add_argument('--batch_size', action='store', type=int, default=32)
parser.add_argument('--workers', action='store', type=int, default=6)
parser.add_argument('--seed_val', action='store', type=int, default=1)
# Model parameters
parser.add_argument('--num_episodes', action='store', type=int, default=100)
parser.add_argument('--max_steps', action='store', type=int, default=100)
parser.add_argument('--num_actions', action='store', type=int, default=4) # 0-L, 1-R, 2-F, 3-B
parser.add_argument('--replay_memory_size', action='store', type=int, default=10000)
parser.add_argument('--gamma', action='store', type=float, default=0.999, help='future reward discount')
parser.add_argument('--epsilon_start', action='store', type=float, default=0.9, help='start epsilon')
parser.add_argument('--epsilon_stop', action='store', type=float, default=0.05, help='stop epsilon')
parser.add_argument('--epsilon_decay', action='store', type=int, default=250, help='epsilon decay')
parser.add_argument('--threshold_dist', action='store', type=float, default=2)
parser.add_argument('--input_channels', action='store', type=int, default=3)
parser.add_argument('--target_update', action='store', type=int, default=10)
parser.add_argument('--reward_average_window', action='store', type=int, default=10)
parser.add_argument('--averageRewardThreshold', action='store', type=int, default=4500)
# Environment parameters
parser.add_argument('--start_x', action='store', type=int, default=0)
parser.add_argument('--start_y', action='store', type=int, default=0)
parser.add_argument('--start_z', action='store', type=float, default=0.1)
parser.add_argument('--final_x', action='store', type=int, default=-3)
parser.add_argument('--final_y', action='store', type=int, default=8)
parser.add_argument('--final_z', action='store', type=int, default=0)
parser.add_argument('--reward_goal', action='store', type=int, default=5000)
parser.add_argument('--reward_collision', action='store', type=float, default=-0.05)
parser.add_argument('--reward_outside_boundary', action='store', type=int, default=-100)
parser.add_argument('--screen_height', action='store', type=int, default=200)
parser.add_argument('--screen_width', action='store', type=int, default=200)
# Agent parameters
parser.add_argument('--targetVel', action='store', type=int, default=5, help='rad/s')
parser.add_argument('--maxForce', action='store', type=int, default=50)
# Default parameters
parser.add_argument('--lr', action='store', type=float,default=0.001)
parser.add_argument('--step_size', action='store', type=float,default=20)
parser.add_argument('--LR_gamma', action='store', type=float,default=0.1)
parser.add_argument('--experiment', action='store', type=str,default='trial')
parser.add_argument('--save_path', action='store', type=str,default='./checkpoints/')
args = parser.parse_args()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
PolicyNet = DQNetwork(args.input_channels, args.num_actions).to(device)
TargetNet = DQNetwork(args.input_channels, args.num_actions).to(device)
TargetNet.load_state_dict(PolicyNet.state_dict())
TargetNet.eval()
optimizer = optim.Adam(PolicyNet.parameters(), lr=args.lr, betas=(0.9, 0.999), eps=1e-08, weight_decay=0, amsgrad=False)
scheduler = lr_scheduler.MultiStepLR(optimizer, milestones=[10, 20, 40, 80], gamma=args.LR_gamma)
memory = ReplayMemory(args, args.replay_memory_size)
env = CarEnvironment(args, device)
env.reset()
trainer = Trainer(args, env, memory, PolicyNet, TargetNet, optimizer, scheduler, device)
trainer.train()
print("Training Complete!")
env.close()