-
Notifications
You must be signed in to change notification settings - Fork 4
/
train.py
277 lines (238 loc) · 11.4 KB
/
train.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
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
import argparse
import numpy as np
import os
import time
import pickle
import torch
import sys
from algorithm import ATOC_trainer
import matplotlib.pyplot as plt
import xlwt
import math
import pandas as pd
def parse_args():
parser = argparse.ArgumentParser("Reinforcement Learning experiments for multiagent environments")
# Environment
parser.add_argument("--scenario", type=str, default="simple_spread", help="name of the scenario script")
parser.add_argument("--max_episode_len", type=int, default=50, help="maximum episode length")
parser.add_argument("--num_episodes", type=int, default=100, help="number of episodes")
parser.add_argument("--T", type=int, default=15, help="number of step to initiate a communicagtion group")
parser.add_argument("--m", type=int, default=2, help="number agents in a communicagtion group")
# Core training parameters
parser.add_argument("--actor_lr", type=float, default=3e-4, help="learning rate for actor")
parser.add_argument("--critic_lr", type=float, default=1e-3, help="learning rate for critic")
parser.add_argument("--gamma", type=float, default=0.96, help="discount factor")
parser.add_argument("--actor_hidden_size", type=int, default=128, help="number of units in the actor network")
parser.add_argument("--critic_hidden_size", type=int, default=128, help="number of units in the critic network")
parser.add_argument("--tau", type=float, default=0.001, metavar='G', help='discount factor for model (default: 0.001)')
parser.add_argument("--memory_size", type=int, default=2000, help='size of the replay memory')
parser.add_argument("--warmup_size", type=int, default=300, help='number of steps before training, must larger than batch_size')
parser.add_argument("--batch_size", type=int, default=64, help="number of steps to optimize at the same time")
# Random process
parser.add_argument("--ou_theta", type=float, default=0.15, help="noise theta")
parser.add_argument("--ou_sigma", type=float, default=0.2, help="noise sigma")
parser.add_argument("--ou_mu", type=float, default=0.0, help="noise mu")
# Checkpointing
parser.add_argument("--exp_name", type=str, default='test', help="name of the experiment")
parser.add_argument("--save_path", type=str, default="", help="directory in which training state and model should be saved")
parser.add_argument("--save_rate", type=int, default=100, help="save model once every time this many episodes are completed")
# Evaluation
parser.add_argument("--load", type=str, default="", help="which model to load")
parser.add_argument("--restore", action="store_true", default=False)
parser.add_argument("--display", action="store_true", default=False)
parser.add_argument("--benchmark", action="store_true", default=False)
parser.add_argument("--benchmark-iters", type=int, default=100000, help="number of iterations run for benchmarking")
parser.add_argument("--benchmark-dir", type=str, default="./benchmark_files/", help="directory where benchmark data is saved")
parser.add_argument("--plots_dir", type=str, default="./results/", help="directory where plot data is saved")
return parser.parse_args()
def make_env(scenario_name, arglist, benchmark=False):
from multiagent.environment import MultiAgentEnv
import multiagent.scenarios as scenarios
# load scenario from script
scenario = scenarios.load(scenario_name + ".py").Scenario()
world = scenario.make_world()
if benchmark:
env = MultiAgentEnv(world, scenario.reset_world, scenario.reward, scenario.observation, scenario.benchmark_data)
else:
env = MultiAgentEnv(world, scenario.reset_world, scenario.reward, scenario.observation) # reset, reward, obs are callbacks
return env
def train(arglist):
arglist.benchmark = True
env = make_env(arglist.scenario, arglist, arglist.benchmark)
trainer = ATOC_trainer(arglist.gamma, arglist.tau, arglist.actor_hidden_size, arglist.critic_hidden_size, env.observation_space[0], env.action_space[0], arglist)
# if arglist.display or arglist.restore or arglist.benchmark:
# trainer.load_model(arglist.exp_name, suffix=arglist.load)
action_noise = False if arglist.display else True
episode_step = 0
train_step = 0
agent_rewards = [[0.0] for _ in range(env.n)]
episode_rewards = [0.0]
final_save_rewards = [] # sum of rewards for training curve
time_start = time.time()
obs_n = env.reset()
C = None # Communication group
reward = []
delay = [0.0]
energy = [0.0]
record_C = []
occupy = [0.0]
nagents = len(obs_n)
if int(nagents/2)==2:
max_num = 3
else:
max_num = int(nagents/2)
arglist.m = np.random.randint(2, max_num)
listC = []
for i in range(nagents):
listC.append(0)
record_C.append(listC)
print("arglist.m",arglist.m)
print('Starting iterations...')
while True:
env.render()
thoughts = trainer.get_thoughts(obs_n) # tensor(nagents, actor_hidden_size)
if (episode_step % arglist.T == 0) or (C == None):
C,total_dist = trainer.initiate_group(obs_n, arglist.m, thoughts)
# TODO
inter_thoughts = trainer.update_thoughts(thoughts, C) # (nagents, actor_hidden_size)
action_n = trainer.select_action(thoughts, inter_thoughts, C)
# TODO calc delta Q for the training of the attention unit
trainer.calc_delta_Q(obs_n, action_n, thoughts, C)
# if energy_n!= 0:
# print(energy_n)
listC = []
for c_index in range(C.shape[0]):
if C[c_index][c_index]:
# print(record_C[-1])
# print(record_C[-1][c_index])
listC.append(1+int(record_C[-1][c_index]))
else:
listC.append(0+int(record_C[-1][c_index]))
# listC.append(C[c_index][c_index])
record_C[-1] = listC
# add noise to the action for exploring
# action_n += action_noise * trainer.random_process.sample()
# action_n = np.clip(action_n, 0.0, 1.0)
new_obs_n, reward_n, done_n, info_n = env.step(action_n)
occupy[-1] += info_n[0]
# print(info_n[0])
# print("reward_n", reward_n)
episode_step += 1
train_step += 1
done = all(done_n)
terminal = (episode_step >= arglist.max_episode_len)
# collect experience
trainer.memory.push(obs_n, action_n, reward_n, new_obs_n, C.data.numpy())
obs_n = new_obs_n
# print("reward_n",reward_n)
delay_n, energy_n = trainer.calc_delayAndenergy(C, total_dist)
delay[-1] += delay_n
energy[-1] += energy_n
for i, rew in enumerate(reward_n):
episode_rewards[-1] += rew
agent_rewards[i][-1] += rew
# print(episode_rewards)
# print(len(episode_rewards))
if done or terminal:
# TODO: not to train attention unit
if len(episode_rewards) % 10 == 0:
trainer.update_attention_unit()
reward.append(episode_rewards[-1]/(arglist.max_episode_len*4))
print("steps: {}, episodes: {}, mean_episode_reward: {}, time: {}".format(
train_step, len(episode_rewards), episode_rewards[-1]/(arglist.max_episode_len*nagents), round(time.time() - time_start, 3)))
time_start = time.time()
obs_n = env.reset(mode = 1) #
episode_step = 0
episode_rewards.append(0)
delay.append(0)
print(occupy)
occupy.append(0)
energy.append(0)
listC = []
for z in range(nagents):
listC.append(0)
record_C.append(listC)
# print(len(record_C))
# print(record_C)
for a in agent_rewards:
a.append(0)
# for displaying learned policies
if arglist.display:
time.sleep(0.1)
env.render()
continue
# update trainer every step, if not in display mode
loss = None
if len(trainer.memory) >= arglist.warmup_size and (train_step % 50 == 0):
loss = trainer.update_parameters()
# save model and display training output
# and (len(episode_rewards) % arglist.save_rate == 0)
# if terminal:
# trainer.save_model(arglist.exp_name, suffix=str(len(episode_rewards)//arglist.save_rate))
# print("steps: {}, episodes: {}, mean_episode_reward: {}, time: {}".format(
# train_step, len(episode_rewards), episode_rewards, round(time.time() - time_start, 3)))
# time_start = time.time()
# final_save_rewards.append(np.mean(episode_rewards[-arglist.save_rate:]))
# np.mean(episode_rewards[-arglist.save_rate-1:-1])
# save final episodes rewards for plotting training results
if len(episode_rewards) > arglist.num_episodes:
# reward_file_name = arglist.plots_dir + arglist.exp_name + '_rewards.pkl'
# with open(reward_file_name, 'wb') as fp:
# pickle.dump(final_save_rewards, fp)
plt.figure(5)
plt.title("rewards")
plt.plot(episode_rewards, ls="-", lw=2, label="plot figure")
plt.legend()
plt.show()
# workbook = xlwt.Workbook(encoding='utf-8')
# # 创建一个worksheet
# worksheet = workbook.add_sheet('My Worksheet')
# worksheet1 = workbook.add_sheet('My C')
#
# # 写入excel
# # 参数对应 行, 列, 值
# for value in range(len(episode_rewards)):
# worksheet.write(value, 0, label=episode_rewards[value])
# for value in range(len(delay)):
# worksheet.write(value, 1, label=delay[value])
# for value in range(len(energy)):
# worksheet.write(value, 2, label=energy[value])
# worksheet.write(0, 3, label=arglist.m)
#
#
#
# for value_x in range(len(record_C)):
# for value_y in range(len(record_C[value_x])):
# worksheet1.write(value_x, value_y, label=record_C[value_x][value_y])
# workbook.save('Excel_test.xls')
# result.append(arglist.m)
data1 = pd.DataFrame(episode_rewards)
data2 = pd.DataFrame(delay)
data3 = pd.DataFrame(energy)
data4 = pd.DataFrame(record_C)
data6 = pd.DataFrame(occupy)
m = []
m.append(arglist.m)
data5 = pd.DataFrame(m)
writer = pd.ExcelWriter('data.xlsx')
data1.to_excel(writer, 'episode_rewards')
data2.to_excel(writer, 'delay')
data3.to_excel(writer, 'energy')
data4.to_excel(writer, 'record_C')
data5.to_excel(writer, 'arglist.m')
data6.to_excel(writer,'occupy')
writer.save()
# writer.close()
# print(dataframe)
print("Finish total of {} episodes.".format(len(episode_rewards)))
print(record_C[-10:])
# plt.figure(6)
# plt.title("reward")
# plt.plot(reward, ls="-", lw=2, label="plot figure")
# plt.legend()
#
# plt.show()
break
if __name__=="__main__":
arglist = parse_args()
train(arglist)