-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_pcn.py
198 lines (183 loc) · 8.15 KB
/
train_pcn.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
import datetime
from pathlib import Path
import random
import mo_gymnasium as mo_gym
from motndp.city import City
from motndp.constraints import MetroConstraints
import numpy as np
import torch
import envs
import argparse
from morl_baselines.multi_policy.pcn.pcn import PCN, BasePCNModel, DiscreteActionsDefaultModel
from morl_baselines.multi_policy.pcn.pcn_tndp import PCNTNDP, PCNTNDPModel
def main(args):
def make_env(gym_env):
if gym_env == 'deep-sea-treasure-concave-v0':
return mo_gym.make(gym_env)
city = City(
args.city_path,
groups_file=args.groups_file,
ignore_existing_lines=args.ignore_existing_lines
)
env = mo_gym.make(args.gym_env,
city=city,
constraints=MetroConstraints(city),
nr_stations=args.nr_stations)
return env
env = make_env(args.gym_env)
if args.gym_env == 'deep-sea-treasure-concave-v0':
agent = PCN(
env,
scaling_factor=args.scaling_factor,
learning_rate=args.lr,
batch_size=args.batch_size,
project_name=args.project_name,
experiment_name=args.experiment_name,
log=not args.no_log,
seed=args.seed,
hidden_dim=args.hidden_dim,
model_class=DiscreteActionsDefaultModel
)
else:
agent = PCNTNDP(
env,
scaling_factor=args.scaling_factor,
learning_rate=args.lr,
batch_size=args.batch_size,
project_name="MORL-TNDP",
experiment_name=args.experiment_name,
log=not args.no_log,
seed=args.seed,
nr_layers=args.nr_layers,
hidden_dim=args.hidden_dim,
model_class=PCNTNDPModel
)
if args.starting_loc is None:
print('NOTE: Training is running with random starting locations.')
save_dir = Path(f"./results/pcn_{args.env}_{datetime.datetime.today().strftime('%Y%m%d_%H_%M_%S.%f')}")
if args.gym_env == 'deep-sea-treasure-concave-v0':
agent.train(
total_timesteps=args.timesteps,
eval_env=make_env(args.gym_env),
ref_point=args.ref_point,
num_er_episodes=args.num_er_episodes,
num_step_episodes=args.num_step_episodes,
max_buffer_size=args.max_buffer_size,
num_model_updates=args.num_model_updates,
max_return=args.max_return,
)
else:
agent.train(
total_timesteps=args.timesteps,
eval_env=make_env(args.gym_env),
ref_point=args.ref_point,
num_er_episodes=args.num_er_episodes,
num_step_episodes=args.num_step_episodes,
max_buffer_size=args.max_buffer_size,
num_model_updates=args.num_model_updates,
starting_loc=args.starting_loc,
nr_stations=args.nr_stations,
max_return=args.max_return,
save_dir=save_dir,
pf_plot_limits=args.pf_plot_limits,
n_policies=args.num_policies,
cd_threshold=args.cd_threshold,
# known_pareto_front=env.unwrapped.pareto_front(gamma=1.0),
)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="MO PCN - TNDP")
# Acceptable values: 'dilemma', 'margins', 'amsterdam'
parser.add_argument('--env', default='dilemma', type=str)
# For amsterdam environment we have different groups files (different nr of objectives)
parser.add_argument('--nr_groups', default=5, type=int)
# Starting location of the agent
parser.add_argument('--starting_loc_x', default=None, type=int)
parser.add_argument('--starting_loc_y', default=None, type=int)
# Episode horizon -- used as a proxy of both the budget and the number of stations (stations are not really costed)
parser.add_argument('--nr_stations', type=int, required=True)
parser.add_argument('--lr', default=1e-2, type=float)
parser.add_argument('--batch_size', default=256, type=int)
parser.add_argument('--num_er_episodes', default=50, type=int)
parser.add_argument('--num_step_episodes', default=10, type=int)
parser.add_argument('--num_model_updates', default=10, type=int)
parser.add_argument('--num_policies', default=10, type=int)
parser.add_argument('--max_buffer_size', default=50, type=int)
parser.add_argument('--nr_layers', default=1, type=int)
parser.add_argument('--hidden_dim', default=64, type=int)
parser.add_argument('--timesteps', default=2000, type=int)
parser.add_argument('--no_log', action='store_true', default=False)
parser.add_argument('--seed', default=42, type=int)
parser.add_argument('--cd_threshold', default=0.2, type=float, help='controls the threshold for crowdedness distance.')
args = parser.parse_args()
print(args)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
random.seed(args.seed)
# Some values are hardcoded for each environment (this is flexible, but we don't want to have to pass 100 arguments to the script)
if args.env == 'dilemma':
args.city_path = Path(f"./envs/mo-tndp/cities/dilemma_5x5")
args.nr_stations = 9
args.gym_env = 'motndp_dilemma-v0'
args.groups_file = "groups.txt"
args.ignore_existing_lines = True
args.experiment_name = "PCN-Dilemma"
args.scaling_factor = np.array([1, 1, 0.1])
args.ref_point = np.array([0, 0])
args.max_return=np.array([1, 1])
args.pf_plot_limits = [0, 0.5]
elif args.env == 'margins':
args.city_path = Path(f"./envs/mo-tndp/cities/margins_5x5")
args.nr_stations = 9
args.gym_env = 'motndp_margins-v0'
args.groups_file = f"groups.txt"
args.ignore_existing_lines = True
args.experiment_name = "PCN-Margins"
args.scaling_factor = np.array([1, 1, 0.1])
args.ref_point = np.array([0, 0])
args.max_return=np.array([1, 1])
args.pf_plot_limits = [0, 0.5]
elif args.env == 'amsterdam':
args.city_path = Path(f"./envs/mo-tndp/cities/amsterdam")
args.gym_env = 'motndp_amsterdam-v0'
args.groups_file = f"price_groups_{args.nr_groups}.txt"
args.ignore_existing_lines = True
args.experiment_name = "PCN-Amsterdam"
args.scaling_factor = np.array([100] * args.nr_groups + [0.01])
args.ref_point = np.array([0] * args.nr_groups)
args.max_return=np.array([1] * args.nr_groups)
# args.pf_plot_limits = [0, 0.015]
args.pf_plot_limits = None
elif args.env == 'xian':
args.city_path = Path(f"./envs/mo-tndp/cities/xian")
args.gym_env = 'motndp_xian-v0'
args.groups_file = f"price_groups_{args.nr_groups}.txt"
args.ignore_existing_lines = True
args.experiment_name = "PCN-Xian"
args.scaling_factor = np.array([100] * args.nr_groups + [0.01])
args.ref_point = np.array([0] * args.nr_groups)
args.max_return=np.array([1] * args.nr_groups)
args.pf_plot_limits = None
elif args.env == 'amsterdam_10x10':
args.city_path = Path(f"./envs/mo-tndp/cities/amsterdam_10x10")
args.gym_env = 'motndp_amsterdam_10x10-v0'
args.groups_file = f"price_groups_{args.nr_groups}.txt"
args.ignore_existing_lines = True
args.experiment_name = "PCN-Amsterdam_10x10"
args.scaling_factor = np.array([100] * args.nr_groups + [0.01])
args.ref_point = np.array([0] * args.nr_groups)
args.max_return=np.array([1] * args.nr_groups)
# args.pf_plot_limits = [0, 0.015]
args.pf_plot_limits = None
elif args.env == 'dst':
args.gym_env = 'deep-sea-treasure-concave-v0'
args.project_name = "DST"
args.experiment_name = "PCN-DST"
args.scaling_factor = np.array([0.1, 0.1, 0.01])
args.ref_point = np.array([0.0, -200.0])
args.max_return=np.array([124, -1])
args.pf_plot_limits = None
if args.starting_loc_x is not None and args.starting_loc_y is not None:
args.starting_loc = (args.starting_loc_x, args.starting_loc_y)
else:
args.starting_loc = None
main(args)