-
Notifications
You must be signed in to change notification settings - Fork 12
/
train_stable_baselines.py
183 lines (157 loc) · 7.9 KB
/
train_stable_baselines.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
# this file is used to evalaute the performance of the ev2gym environment with various stable baselines algorithms.
from stable_baselines3 import PPO, A2C, DDPG, SAC, TD3
from stable_baselines3.common.callbacks import BaseCallback
from stable_baselines3.common.callbacks import EvalCallback
from sb3_contrib import TQC, TRPO, ARS, RecurrentPPO
from ev2gym.models.ev2gym_env import EV2Gym
from ev2gym.rl_agent.reward import SquaredTrackingErrorReward, ProfitMax_TrPenalty_UserIncentives
from ev2gym.rl_agent.reward import profit_maximization
from ev2gym.rl_agent.state import V2G_profit_max, PublicPST, V2G_profit_max_loads
import gymnasium as gym
import argparse
import wandb
from wandb.integration.sb3 import WandbCallback
import os
import yaml
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument('--algorithm', type=str, default="ddpg")
parser.add_argument('--device', type=str, default="cuda:0")
parser.add_argument('--run_name', type=str, default="")
parser.add_argument('--config_file', type=str,
# default="ev2gym/example_config_files/V2GProfitMax.yaml")
default="ev2gym/example_config_files/PublicPST.yaml")
algorithm = parser.parse_args().algorithm
device = parser.parse_args().device
run_name = parser.parse_args().run_name
config_file = parser.parse_args().config_file
config = yaml.load(open(config_file, 'r'), Loader=yaml.FullLoader)
if config_file == "ev2gym/example_config_files/V2GProfitMax.yaml":
reward_function = profit_maximization
state_function = V2G_profit_max
group_name = f'{config["number_of_charging_stations"]}cs_V2GProfitMax'
elif config_file == "ev2gym/example_config_files/PublicPST.yaml":
reward_function = SquaredTrackingErrorReward
state_function = PublicPST
group_name = f'{config["number_of_charging_stations"]}cs_PublicPST'
elif config_file == "ev2gym/example_config_files/V2GProfitPlusLoads.yaml":
reward_function = ProfitMax_TrPenalty_UserIncentives
state_function = V2G_profit_max_loads
group_name = f'{config["number_of_charging_stations"]}cs_V2GProfitPlusLoads'
run_name += f'{algorithm}_{reward_function.__name__}_{state_function.__name__}'
run = wandb.init(project='ev2gym',
sync_tensorboard=True,
group=group_name,
name=run_name,
save_code=True,
)
gym.envs.register(id='evs-v0', entry_point='ev2gym.models.ev2gym_env:EV2Gym',
kwargs={'config_file': config_file,
'verbose': False,
'save_plots': False,
'generate_rnd_game': True,
'reward_function': reward_function,
'state_function': state_function,
})
env = gym.make('evs-v0')
eval_log_dir = "./eval_logs/"
os.makedirs(eval_log_dir, exist_ok=True)
os.makedirs(f"./saved_models/{group_name}", exist_ok=True)
eval_callback = EvalCallback(env, best_model_save_path=eval_log_dir,
log_path=eval_log_dir,
eval_freq=config['simulation_length']*50,
n_eval_episodes=10, deterministic=True,
render=False)
if algorithm == "ddpg":
model = DDPG("MlpPolicy", env, verbose=1,
learning_rate = 1e-3,
buffer_size = 1_000_000, # 1e6
learning_starts = 100,
batch_size = 100,
tau = 0.005,
gamma = 0.99,
device=device, tensorboard_log="./logs/")
elif algorithm == "td3":
model = TD3("MlpPolicy", env, verbose=1,
device=device, tensorboard_log="./logs/")
elif algorithm == "sac":
model = SAC("MlpPolicy", env, verbose=1,
device=device, tensorboard_log="./logs/")
elif algorithm == "a2c":
model = A2C("MlpPolicy", env, verbose=1,
device=device, tensorboard_log="./logs/")
elif algorithm == "ppo":
model = PPO("MlpPolicy", env, verbose=1,
device=device, tensorboard_log="./logs/")
elif algorithm == "tqc":
model = TQC("MlpPolicy", env, verbose=1,
device=device, tensorboard_log="./logs/")
elif algorithm == "trpo":
model = TRPO("MlpPolicy", env, verbose=1,
device=device, tensorboard_log="./logs/")
elif algorithm == "ars":
model = ARS("MlpPolicy", env, verbose=1,
device=device, tensorboard_log="./logs/")
elif algorithm == "rppo":
model = RecurrentPPO("MlpLstmPolicy", env, verbose=1,
device=device, tensorboard_log="./logs/")
else:
raise ValueError("Unknown algorithm")
model.learn(total_timesteps=4_000_000,
progress_bar=True,
callback=[
WandbCallback(
gradient_save_freq=100000,
model_save_path=f"models/{run.id}",
verbose=2),
eval_callback])
model.save(f"./saved_models/{group_name}/{run_name}")
env = model.get_env()
obs = env.reset()
stats = []
for i in range(96*1000):
action, _states = model.predict(obs, deterministic=True)
obs, reward, done, info = env.step(action)
# env.render()
# VecEnv resets automatically
if done:
stats.append(info)
obs = env.reset()
# print average stats
print("=====================================================")
print(f' Average stats for {algorithm} algorithm, {len(stats)} episodes')
print("total_ev_served: ", sum(
[i[0]['total_ev_served'] for i in stats])/len(stats))
print("total_profits: ", sum(
[i[0]['total_profits'] for i in stats])/len(stats))
print("total_energy_charged: ", sum(
[i[0]['total_energy_charged'] for i in stats])/len(stats))
print("total_energy_discharged: ", sum(
[i[0]['total_energy_discharged'] for i in stats])/len(stats))
print("average_user_satisfaction: ", sum(
[i[0]['average_user_satisfaction'] for i in stats])/len(stats))
print("power_tracker_violation: ", sum(
[i[0]['power_tracker_violation'] for i in stats])/len(stats))
print("tracking_error: ", sum(
[i[0]['tracking_error'] for i in stats])/len(stats))
print("energy_user_satisfaction: ", sum(
[i[0]['energy_user_satisfaction'] for i in stats])/len(stats))
print("total_transformer_overload: ", sum(
[i[0]['total_transformer_overload'] for i in stats])/len(stats))
print("reward: ", sum([i[0]['episode']['r'] for i in stats])/len(stats))
run.log({
"test/total_ev_served": sum([i[0]['total_ev_served'] for i in stats])/len(stats),
"test/total_profits": sum([i[0]['total_profits'] for i in stats])/len(stats),
"test/total_energy_charged": sum([i[0]['total_energy_charged'] for i in stats])/len(stats),
"test/total_energy_discharged": sum([i[0]['total_energy_discharged'] for i in stats])/len(stats),
"test/average_user_satisfaction": sum([i[0]['average_user_satisfaction'] for i in stats])/len
(stats),
"test/power_tracker_violation": sum([i[0]['power_tracker_violation'] for i in stats])/len(stats),
"test/tracking_error": sum([i[0]['tracking_error'] for i in stats])/len(stats),
"test/energy_user_satisfaction": sum([i[0]['energy_user_satisfaction'] for i in stats])/len
(stats),
"test/total_transformer_overload": sum([i[0]['total_transformer_overload'] for i in stats])/len
(stats),
"test/reward": sum([i[0]['episode']['r'] for i in stats])/len(stats),
})
run.finish()