-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_IQL.py
150 lines (125 loc) · 5.95 KB
/
train_IQL.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
import numpy as np
import d4rl
import torch
import gym
import argparse
import os
#import d4rl_pybullet
from tqdm import tqdm
import utils
import IQL
def eval_policy(policy, env_name, seed, mean, std, seed_offset=0, eval_episodes=10):
eval_env = gym.make('HalfCheetah-v2')
eval_env.seed(seed + seed_offset)
policy.actor.eval()
avg_reward = 0.
for _ in range(eval_episodes):
state, done = eval_env.reset(), False
while not done:
state = (np.array(state).reshape(1, -1) - mean) / std
action = policy.select_action(state)
eval_env.render()
state, reward, done, _ = eval_env.step(action)
avg_reward += reward
eval_env.close()
avg_reward /= eval_episodes
return avg_reward
if __name__ == "__main__":
parser = argparse.ArgumentParser()
# Experiment
parser.add_argument("--policy", default="IQL") # Policy name
parser.add_argument("--env", default="halfcheetah-expert-v2") # OpenAI gym environment name
parser.add_argument("--seed", default=1, type=int) # Sets Gym, PyTorch and Numpy seeds
parser.add_argument("--eval_freq", default=1e2, type=int) # How often (time steps) we evaluate
parser.add_argument("--max_timesteps", default=1e4, type=int) # Max time steps to run environment
parser.add_argument("--save_model", default="True",action="store_true") # Save model and optimizer parameters
parser.add_argument("--load_model", default="") # Model load file name, "" doesn't load, "default" uses file_name
# IQL
parser.add_argument("--batch_size", default=256, type=int) # Batch size for both actor and critic
parser.add_argument("--discount", default=0.99) # Discount factor
parser.add_argument("--tau", default=0.005) # Target network update rate
parser.add_argument("--expectile", default=0.7) # Expectile parameter Tau
parser.add_argument("--beta", default=3.0) # Temperature parameter Beta
parser.add_argument("--max_weight", default=100.0) # Max weight for actor update
parser.add_argument("--normalize_data", default=True)
args = parser.parse_args()
file_name = f"{args.policy}_{args.env}_{args.seed}"
print("---------------------------------------")
print(f"Policy: {args.policy}, Env: {args.env}, Seed: {args.seed}")
print("---------------------------------------")
if not os.path.exists("./results"):
os.makedirs("./results")
if args.save_model and not os.path.exists("./models"):
os.makedirs("./models")
env = gym.make(args.env)
# Set seeds
env.seed(args.seed)
env.action_space.seed(args.seed)
torch.manual_seed(args.seed)
np.random.seed(args.seed)
state_dim = env.observation_space.shape[0]
action_dim = env.action_space.shape[0]
max_action = float(env.action_space.high[0])
hidden = (256, 256)
kwargs = {
"state_dim": state_dim,
"action_dim": action_dim,
"max_action": max_action,
"discount": args.discount,
"tau": args.tau,
"expectile": args.expectile,
"beta": args.beta,
"max_weight": args.max_weight,
"actor_hidden": hidden,
"critic_hidden": hidden,
"value_hidden": hidden,
}
# Initialize policy
policy = IQL.IQL(**kwargs)
if args.load_model != "":
policy_file = file_name if args.load_model == "default" else args.load_model
policy.load(f"./models/{policy_file}")
replay_buffer = utils.ReplayBuffer(state_dim, action_dim)
dataset = env.get_dataset()
# In the case of D4RL-Pybullet dataset, else use conver_D4RL method
replay_buffer.convert_D4RL(dataset)
if args.normalize_data:
mean, std = replay_buffer.normalize_states()
else:
mean, std = 0, 1
n_epochs = int(args.max_timesteps) // int(args.eval_freq)
evaluations = []
eval = eval_policy(policy, args.env, args.seed, mean, std)
print('initial eval: {}'.format(eval))
# for epoch in range(n_epochs):
# range_gen = tqdm(
# range(int(args.eval_freq)),
# desc=f"Epoch {int(epoch)}/{n_epochs}",
# )
# mean_val = 0
# mean_q = 0
# mean_v_loss = 0
# mean_c_loss = 0
# mean_a_loss = 0
# for itr in range_gen:
# val, q, value_loss, critic_loss, actor_loss = policy.train(replay_buffer, args.batch_size)
# info = policy.train(replay_buffer, args.batch_size)
# mean_val += info['value']
# mean_q += info['q_val']
# mean_c_loss += info['critic_loss']
# mean_v_loss += info['value_loss']
# mean_a_loss += info['actor_loss']
# eval = eval_policy(policy, args.env, args.seed, mean, std)
# evaluations.append(eval)
# policy.actor_scheduler.step()
# print('Epoch {}/{}: value: {:.3f}. Q: {:.3f}. value_loss: {:.3f}. critic_loss: {:.3f}. actor_loss: {:.3f} env: {:.2f}'.format(
# epoch, n_epochs,
# mean_val / len(range_gen),
# mean_q / len(range_gen),
# mean_v_loss / len(range_gen),
# mean_c_loss / len(range_gen),
# mean_a_loss / len(range_gen),
# eval))
# np.save(f"./results/{file_name}", evaluations)
# if args.save_model: policy.save(f"./models/{file_name}")
env.close()