-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathrunner.py
200 lines (136 loc) · 7.79 KB
/
runner.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
199
200
import numpy as np
import os
import ray
import time
import matplotlib.pyplot as plt
from common.rollout import RolloutWorker
from agent.agent import Agents
from common.replay_buffer import ReplayBuffer
import matplotlib.pyplot as plt
from tensorboardX import SummaryWriter
ray.init()
@ray.remote(num_cpus=3)
class RayRolloutWorker():
def __init__(self, id, args):
self.id = id
self.args = args
self.rayrolloutWorker = RolloutWorker(self.id, self.args)
def rollout(self, model):
return self.rayrolloutWorker.generate_episode(model)
def rollout_test(self, model):
return self.rayrolloutWorker.generate_episode_test(model)
def close(self):
pass
class Runner:
def __init__(self, args):
self.args = args
self.agents = Agents(args, master = True)
self.buffer = ReplayBuffer(self.args)
self.workers = None
stamp = int(time.time())
if not self.args.dr_coef:
self.save_path = self.args.result_dir + '/' + args.alg + '/no_dr/' + '%f'%args.level + '/' + (time.strftime("%Y_%m_%d_%H", time.localtime(stamp)))
else:
self.save_path = self.args.result_dir + '/' + args.alg + '/dr/' + '%f'%args.level + '/' + (time.strftime("%Y_%m_%d_%H", time.localtime(stamp)))
if not os.path.exists(self.save_path):
os.makedirs(self.save_path)
if not self.args.dr_coef:
self.log_path = self.args.log_dir + '/' + self.args.action_space + '/' + args.alg + '/no_dr/' + '%f'%args.level + '/' + (time.strftime("%Y_%m_%d_%H", time.localtime(stamp)))
else:
self.log_path = self.args.log_dir + '/' + self.args.action_space + '/' + args.alg + '/dr/' + '%f'%args.level + '/' + (time.strftime("%Y_%m_%d_%H", time.localtime(stamp)))
if not os.path.exists(self.log_path):
os.makedirs(self.log_path)
self.writer = SummaryWriter(self.log_path)
np.random.seed(self.args.seed)
def run(self):
self.workers = [RayRolloutWorker.remote(i, self.args) for i in range(1, self.args.env_num + 1)]
time_steps = 0
episode_rewards_train = 0
is_wins = 0
for eps in range(int(self.args.n_eps)):
rnn_para = self.agents.policy.get_para()
model_id = ray.put(rnn_para)
buffers_ids = [
worker.rollout.remote(model_id) for worker in self.workers
]
start_time = time.time()
episodes = []
for batch in range(self.args.env_num):
[buffers_id], buffers_ids = ray.wait(buffers_ids)
episode, episode_reward_train, is_win, steps = ray.get(buffers_id)
episodes.append(episode)
time_steps += steps
is_wins += is_win
episode_rewards_train += episode_reward_train
if (eps % self.args.print_cycle == 0) :
if eps != 0 :
print("ep : ", eps, " step : ", time_steps,
" win_rate : ", is_wins / self.args.env_num / self.args.print_cycle,
" reward : ",episode_rewards_train / self.args.env_num/ self.args.print_cycle)
self.writer.add_scalar('train_win_rate', is_wins / self.args.env_num / self.args.print_cycle, global_step = eps * self.args.env_num)
self.writer.add_scalar('train_rewards', episode_rewards_train / self.args.env_num / self.args.print_cycle, global_step = eps * self.args.env_num)
else:
print("ep : ", eps, " step : ", time_steps,
" win_rate : ", is_wins / self.args.env_num,
" reward : ",episode_rewards_train / self.args.env_num)
self.writer.add_scalar('train_win_rate', is_wins / self.args.env_num, global_step = eps * self.args.env_num)
self.writer.add_scalar('train_rewards', episode_rewards_train / self.args.env_num, global_step = eps * self.args.env_num)
episode_rewards_train = 0
is_wins = 0
episode_batch = episodes[0]
episodes.pop(0)
for episode in episodes:
for key in episode_batch.keys():
episode_batch[key] = np.concatenate((episode_batch[key], episode[key]), axis=0)
# train qmix here
self.buffer.store_episode(episode_batch)
if time_steps > 0:
need_train_step = 2 * int (self.buffer.current_size // self.args.batch_size)
# if need_train_step < 20:
if need_train_step < 20:
need_train_step = 20
if eps % self.args.target_update_cycle == 0:
target_update = True
else:
target_update = False
for train_step in range(need_train_step):
mini_batch = self.buffer.sample(min(self.buffer.current_size, self.args.batch_size))
self.agents.train(mini_batch, train_step, target_update)
if eps % self.args.save_cycle == 0:
self.agents.policy.save_model(eps)
if eps % self.args.evaluate_cycle == 0:
rnn_para = self.agents.policy.get_para()
model_id = ray.put(rnn_para)
win_rates = 0
episode_rewards = 0
ep_steps = 0
for i in range(self.args.evaluate_epoch):
buffers_ids = [worker.rollout_test.remote(model_id) for worker in self.workers]
for batch in range(self.args.env_num):
[buffers_id], buffers_ids = ray.wait(buffers_ids)
win_rate, episode_reward, ep_step = ray.get(buffers_id)
win_rates += win_rate
episode_rewards += episode_reward
ep_steps += ep_step
print( "testing ...... ep : ", eps,
" win_rate : ", win_rates / self.args.env_num / self.args.evaluate_epoch,
" reward : ",episode_rewards / self.args.env_num / self.args.evaluate_epoch)
self.writer.add_scalar('test_win_rate', win_rates / self.args.env_num / self.args.evaluate_epoch, global_step = eps * self.args.env_num)
self.writer.add_scalar('test_rewards', episode_rewards / self.args.env_num / self.args.evaluate_epoch, global_step = eps * self.args.env_num)
# def test(self, port, eps):
# self.agents.policy.load_model(eps)
# print("self.args.no_graphics : ********************************* ", self.args.no_graphics)
# rolloutWorker = RolloutWorker(port, self.args, no_graphics=self.args.no_graphics, time_scale = self.args.time_scale)
# rnn_para = self.agents.policy.get_para()
# win_rates = 0
# episode_rewards = 0
# ep_steps = 0
# for i in range(20):
# print(i)
# win_rate, episode_reward, ep_step = rolloutWorker.generate_episode_test_easy(rnn_para)
# win_rates += win_rate
# episode_rewards += episode_reward
# ep_steps += ep_step
# print( "testing ...... ep : ", eps,
# " win_rate : ", win_rates / 20,
# " reward : ",episode_rewards / 20)