-
Notifications
You must be signed in to change notification settings - Fork 2
/
train_ddqn.py
79 lines (75 loc) · 2.78 KB
/
train_ddqn.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
# PYTHON_ARGCOMPLETE_OK
import numpy.random
import tensorflow as tf
import numpy as np
import gym
from tensorflow.keras.models import load_model
import dill
from memory import NormalMemory
gym_env_path = '/home/thai/eclipse-workspace/FightingICEv4.5'
# java_env_path='/home/thai/jdk1.8.0_271/bin/java'
import gym_fightingice
import argparse
action_space_num = 56
from agent import AgentWithNormalMemory, AgentWithPER, AgentWithPERAndMultiRewards, AgentNormalMultiReward
from datetime import datetime
def train_with_agent(agent, epsilon, multi_rewards, steps):
agentoo7 = agent(epsilon=epsilon)
# try:
# agentoo7.load_model()
# agentoo7.load_memory()
# except Exception as ex:
# print(ex)
# agentoo7 = agent(epsilon=epsilon)
env_name = 'FightingiceDataNoFrameskip-v0'
# env_name = 'FightingiceDataFrameskip-v0'
env = gym.make(env_name, java_env_path=gym_env_path, freq_restart_java=1, multi_rewards=multi_rewards)
for s in range(steps):
done = False
state = env.reset(p2='MctsAi')
total_reward = 0
start_time = datetime.now()
while not done:
env.render()
if len(state) == 4:
state = state[0]
action = agentoo7.act(state)
next_state, reward, done, _ = env.step(action)
agentoo7.update_mem(state, action, reward, next_state, done)
agentoo7.train()
state = next_state
if isinstance(reward, int):
total_reward += reward
else:
total_reward += sum(reward)
if done:
print("total reward after {} episode is {} and epsilon is {}".format(s, total_reward, agentoo7.epsilon))
print("time for an episode {}", datetime.now() - start_time)
print('Save model')
agentoo7.save_model()
agentoo7.save_memory()
env.close()
if __name__ == '__main__':
# epsilon = 1.0
# agent = AgentWithPER
parser = argparse.ArgumentParser()
parser.add_argument('--epsilon', type=float, default=1)
parser.add_argument('--agent', type=str, choices=['normal', 'per', 'per_multi', 'normal_multi'], default='normal')
parser.add_argument('--step', type=int, default=50)
args = parser.parse_args()
print('parsed args', args)
epsilon = args.epsilon
agents = {
'normal': AgentWithNormalMemory,
'per': AgentWithPER,
'per_multi': AgentWithPERAndMultiRewards,
'normal_multi': AgentNormalMultiReward,
}
multi_reward_types = {
'normal': False,
'per': False,
'per_multi': True,
'normal_multi': True
}
agent = agents[args.agent]
train_with_agent(agent, epsilon, multi_rewards=multi_reward_types[args.agent], steps=args.step)