-
Notifications
You must be signed in to change notification settings - Fork 2
/
mini_network_add_map_bn.py
213 lines (166 loc) · 9.91 KB
/
mini_network_add_map_bn.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
import numpy as np
import tensorflow as tf
import param as P
from algo.ppo_add_map_bn import Policy_net, PPOTrain
# for mini game
_SIZE_MINI_INPUT = 20
_SIZE_MINI_ACTIONS = 10
class MiniNetwork(object):
def __init__(self, sess=None, summary_writer=tf.summary.FileWriter("logs/"), rl_training=False,
reuse=False, cluster=None, index=0, device='/gpu:0',
ppo_load_path=None, ppo_save_path=None,
ob_space_add=0, act_space_add=0,
freeze_head=False, use_bn=True,
use_sep_net=True, load_latest=False,
restore_model=False, restore_from=None, restore_to=None,
add_image=False, partial_restore=True, weighted_sum_type='AddWeight',
initial_type="original"):
self.policy_model_path_load = ppo_load_path + "mini"
self.latest_model_path_load = ppo_load_path + "latest"
self.policy_model_path_save = ppo_save_path + "mini"
self.latest_model_path_save = ppo_save_path + "latest"
self.rl_training = rl_training
self.reuse = reuse
self.sess = sess
self.cluster = cluster
self.index = index
self.device = device
self.ob_space_add = ob_space_add
self.act_space_add = act_space_add
self.add_image = add_image
self.freeze_head = freeze_head
self.use_bn = use_bn
self.use_sep_net = use_sep_net
self.restore_model = restore_model
self.restore_from = restore_from
self.restore_to = restore_to
self.load_latest = load_latest
self.partial_restore = partial_restore
self.weighted_sum_type = weighted_sum_type
self.initial_type = initial_type
if self.ob_space_add == 0 and self.act_space_add == 0 and self.add_image == False:
self.use_add = False
self.lr=P.mini_lr
self.epoch_num = P.mini_epoch_num
else:
self.use_add = True
self.lr=P.mini_lr_add
self.epoch_num = P.mini_epoch_num
self._create_graph()
self.rl_saver = tf.train.Saver()
self.summary_writer = summary_writer
def initialize(self):
init_op = tf.global_variables_initializer()
self.sess.run(init_op)
def reset_old_network(self):
self.policy_ppo.assign_policy_parameters()
self.policy_ppo.reset_mean_returns()
self.sess.run(self.results_sum.assign(0))
self.sess.run(self.game_num.assign(0))
def _create_graph(self):
if self.reuse:
tf.get_variable_scope().reuse_variables()
assert tf.get_variable_scope().reuse
worker_device = "/job:worker/task:%d" % self.index + self.device
with tf.device(tf.train.replica_device_setter(worker_device=worker_device, cluster=self.cluster)):
self.results_sum = tf.get_variable(name="results_sum", shape=[], initializer=tf.zeros_initializer)
self.game_num = tf.get_variable(name="game_num", shape=[], initializer=tf.zeros_initializer)
self.global_step = tf.get_variable(name="global_step", shape=[], dtype=tf.int32,
initializer=tf.zeros_initializer, trainable=False)
self.global_steps = tf.get_variable(name="iter_steps", shape=[], dtype=tf.int32,
initializer=tf.zeros_initializer, trainable=False)
self.win_rate = self.results_sum / self.game_num
self.mean_win_rate = tf.summary.scalar('mean_win_rate_dis', self.results_sum / self.game_num)
self.merged = tf.summary.merge([self.mean_win_rate])
mini_scope = "MiniPolicyNN"
with tf.variable_scope(mini_scope):
ob_space = _SIZE_MINI_INPUT
act_space_array = _SIZE_MINI_ACTIONS
self.policy = Policy_net('policy', self.sess, ob_space, self.ob_space_add,
act_space_array, self.act_space_add, self.freeze_head, self.use_bn, self.use_sep_net,
self.add_image, self.weighted_sum_type, self.initial_type)
self.policy_old = Policy_net('old_policy', self.sess, ob_space, self.ob_space_add,
act_space_array, self.act_space_add, self.freeze_head, self.use_bn, self.use_sep_net,
self.add_image, self.weighted_sum_type, self.initial_type)
self.policy_ppo = PPOTrain('PPO', self.sess, self.policy, self.policy_old, lr=self.lr, epoch_num=self.epoch_num)
var_train_list = tf.get_collection(tf.GraphKeys.TRAINABLE_VARIABLES)
var_all_list = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES)
if self.restore_model:
print('restore_model')
if self.restore_from == 'mini' and self.restore_to == 'mini':
print('restore_model: mini to mini')
self.old_policy_saver = tf.train.Saver(var_list=var_all_list)
elif self.restore_from == 'mini' and self.restore_to == 'source':
print('restore_model: mini to source')
if self.use_add:
variables_to_restore = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope=self.policy.scope)
old_variables_to_restore = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope=self.policy_old.scope)
variables_to_restore += old_variables_to_restore
# remove layers for added obs
variables_to_restore = [v for v in variables_to_restore if len(v.name.split('/')) > 2 and 'DenseLayer3' not in v.name.split('/')]
# remove layer weight for added action
variables_to_restore = [v for v in variables_to_restore if len(v.name.split('/')) > 2 and v.name.split('/')[-2] != 'add_output_layer']
if not self.partial_restore:
variables_to_restore = [v for v in variables_to_restore if len(v.name.split('/')) > 2 and v.name.split('/')[-2] != 'output']
# remove layer weight for weighted sum
variables_to_restore = [v for v in variables_to_restore if len(v.name.split('/')) > 2 and
'AdaptiveWeight:0' not in v.name.split('/') and 'AttentionWeight' not in v.name.split('/')]
if self.add_image:
map_variables_to_restore = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope=self.policy.map_variable_scope)
old_map_variables_to_restore = tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope=self.policy_old.map_variable_scope)
# remove layers for added map
variables_to_restore = [v for v in variables_to_restore if v not in map_variables_to_restore and v not in old_map_variables_to_restore]
print('restore_model: mini to source, use_add')
#print('variables_to_restore:', variables_to_restore)
self.old_policy_saver = tf.train.Saver(var_list=variables_to_restore)
else:
self.old_policy_saver = tf.train.Saver(var_list=var_all_list)
elif self.restore_from == 'source' and self.restore_to == 'source':
self.old_policy_saver = tf.train.Saver(var_list=var_all_list)
else:
self.old_policy_saver = tf.train.Saver(var_list=var_all_list)
else:
self.old_policy_saver = tf.train.Saver(var_list=var_all_list)
self.new_policy_saver = tf.train.Saver(var_list=var_all_list)
def Update_result(self, result_list):
win = 0
for i in result_list:
if i > 0:
win += 1
self.sess.run(self.results_sum.assign_add(win))
self.sess.run(self.game_num.assign_add(len(result_list)))
def Update_summary(self, counter):
print("Update summary........")
policy_summary = self.policy_ppo.get_summary_dis()
self.summary_writer.add_summary(policy_summary, counter)
summary = self.sess.run(self.merged)
self.summary_writer.add_summary(summary, counter)
print("Update summary finished!")
self.sess.run(self.global_steps.assign(counter))
steps = int(self.sess.run(self.global_steps))
win_game = int(self.sess.run(self.results_sum))
all_game = int(self.sess.run(self.game_num))
#print('all_game:', all_game)
win_rate = win_game / float(all_game) if all_game != 0 else 0.
return steps, win_rate
def get_win_rate(self):
return float(self.sess.run(self.win_rate))
def Update_policy(self, buffer):
self.policy_ppo.ppo_train_dis(buffer.observations, buffer.obs_add, buffer.obs_map, buffer.tech_actions,
buffer.rewards, buffer.values, buffer.values_next, buffer.gaes, buffer.returns,
buffer.return_values, self.index, self.summary_writer)
def get_global_steps(self):
return int(self.sess.run(self.global_steps))
def save_policy(self):
self.new_policy_saver.save(self.sess, self.policy_model_path_save)
print("policy has been saved in", self.policy_model_path_save)
def save_latest_policy(self):
self.new_policy_saver.save(self.sess, self.latest_model_path_save)
print("latest policy has been saved in", self.latest_model_path_save)
def restore_policy(self):
if self.load_latest:
self.old_policy_saver.restore(self.sess, self.latest_model_path_load)
print("Restore policy from", self.latest_model_path_load)
else:
self.old_policy_saver.restore(self.sess, self.policy_model_path_load)
print("Restore policy from", self.policy_model_path_load)