-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
executable file
·188 lines (174 loc) · 8.83 KB
/
main.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
import gc
import multiprocessing as mp
import os
from functools import partial
from os.path import isfile
from shutil import copy
import tensorflow as tf
from numpy import zeros
from tensorflow import keras
import MillEnv
import Network
import configs
import encoders
import logger
import mcts
import memory
def execute_generate_play(nnet_path, multiplikator=configs.SIMS_FAKTOR,
exponent=configs.SIMS_EXPONENT):
gc.collect()
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
nnet = Network.get_net(configs.FILTERS, configs.HIDDEN_SIZE, configs.OUT_FILTERS,
configs.NUM_ACTIONS, configs.INPUT_SIZE, None, configs.NUM_RESIDUAL)
nnet.load_weights(nnet_path)
env = MillEnv.MillEnv()
mcts_ = mcts.MonteCarloTreeSearch(mcts.State(zeros((1, 24)), 0, -env.isPlaying, env))
stmem = mcts_.generatePlay(nnet, multiplikator, exponent)
del mcts_
del env
keras.backend.clear_session()
tf.compat.v1.reset_default_graph()
del nnet
gc.collect()
return stmem
def execute_pit(oldNet_path, newNet_path, begins, multiplikator=configs.SIMS_FAKTOR,
exponent=configs.SIMS_EXPONENT):
gc.collect()
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
oldNet = Network.get_net(configs.FILTERS, configs.HIDDEN_SIZE, configs.OUT_FILTERS,
configs.NUM_ACTIONS, configs.INPUT_SIZE, None, configs.NUM_RESIDUAL)
newNet = Network.get_net(configs.FILTERS, configs.HIDDEN_SIZE, configs.OUT_FILTERS,
configs.NUM_ACTIONS, configs.INPUT_SIZE, None, configs.NUM_RESIDUAL)
oldNet.load_weights(oldNet_path)
newNet.load_weights(newNet_path)
winner = mcts.pit(oldNet, newNet, begins, multiplikator, exponent)
keras.backend.clear_session()
tf.compat.v1.reset_default_graph()
del oldNet
del newNet
gc.collect()
return winner
def train_net(in_path, out_path, train_data, tensorboard_path):
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
physical_devices = tf.config.list_physical_devices('GPU')
for gpu_instance in physical_devices:
tf.config.experimental.set_memory_growth(gpu_instance, True)
tensorboard_callback = keras.callbacks.TensorBoard(tensorboard_path, update_freq=10)
current_Network = Network.get_net(configs.FILTERS, configs.HIDDEN_SIZE, configs.OUT_FILTERS,
configs.NUM_ACTIONS, configs.INPUT_SIZE, None, configs.NUM_RESIDUAL)
current_Network.load_weights(in_path)
current_Network.compile(optimizer='adam',
loss={'policy_output': Network.cross_entropy_with_logits, 'value_output': 'mse'},
loss_weights=[0.5, 0.5],
metrics=['accuracy'])
current_Network.fit(
encoders.prepareForNetwork(train_data[0], train_data[1], train_data[2], train_data[3],
train_data[4]),
{'policy_output': train_data[5],
'value_output': train_data[6]}, epochs=configs.EPOCHS,
batch_size=configs.BATCH_SIZE, callbacks=[tensorboard_callback])
current_Network.save_weights(out_path)
def save_first_net(path):
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
current_Network = Network.get_net(configs.FILTERS, configs.HIDDEN_SIZE, configs.OUT_FILTERS,
configs.NUM_ACTIONS, configs.INPUT_SIZE, None, configs.NUM_RESIDUAL)
print(current_Network.summary())
current_Network.save_weights(path)
def save_whole_net(weights_path, model_path):
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
current_Network = Network.get_net(configs.FILTERS, configs.HIDDEN_SIZE, configs.OUT_FILTERS,
configs.NUM_ACTIONS, configs.INPUT_SIZE, None, configs.NUM_RESIDUAL)
current_Network.load_weights(weights_path)
current_Network.save(model_path)
if __name__ == "__main__":
logger_handle = logger.Logger(configs.LOGGER_PATH)
current_mem_size = configs.MIN_MEMORY
mem = memory.Memory(current_mem_size, configs.MAX_MEMORY)
episode = 0
init_net_p = mp.Process(target=save_first_net, args=(configs.BEST_PATH,))
init_net_p.start()
init_net_p.join()
del init_net_p
gc.collect()
copy("configs.py", configs.INTERMEDIATE_SAVE_PATH + "configs.py")
copy(configs.BEST_PATH, configs.NEW_NET_PATH)
if isfile("interrupt_array.npy") and isfile("interrupted_vars.obj"):
episode = mem.loadState("interrupt_array.npy", "interrupted_vars.obj")
copy(configs.NETWORK_PATH + str(episode) + ".h5", configs.BEST_PATH)
logger_handle.log("=========== restarting training ==========")
try:
while episode <= configs.TRAINING_LOOPS:
logger_handle.log("============== starting episode " + str(episode) + " ===============")
current_Network_path = configs.NETWORK_PATH + str(episode) + ".h5"
copy(configs.BEST_PATH, current_Network_path)
logger_handle.log("saving actual net to " + current_Network_path)
logger_handle.log("saving intermediate arrays")
mem.saveState(episode, configs.INTERMEDIATE_SAVE_PATH + "interrupt_array.npy",
configs.INTERMEDIATE_SAVE_PATH + "interrupted_vars.obj")
logger_handle.log("============== starting selfplay ================")
with mp.Pool(configs.NUM_CPUS, maxtasksperchild=10) as pool:
for stmem in pool.imap_unordered(partial(execute_generate_play, configs.BEST_PATH, configs.SIMS_FAKTOR),
[
configs.SIMS_EXPONENT for play in
range(configs.EPISODES)]):
mem.addToMem(stmem)
logger_handle.log("player won: " + str(stmem[0][6]) + " turns played: " + str(len(stmem) // 8))
del stmem
gc.collect()
del pool
gc.collect()
logger_handle.log("saving intermediate arrays")
mem.saveState(episode, configs.INTERMEDIATE_SAVE_PATH + "interrupt_array.npy",
configs.INTERMEDIATE_SAVE_PATH + "interrupted_vars.obj")
logger_handle.log("============== starting training ================")
train_data = mem.getTrainSamples()
train_p = mp.Process(
target=train_net, args=(configs.NEW_NET_PATH, configs.NEW_NET_PATH, train_data,
configs.TENSORBOARD_PATH + str(episode)))
train_p.start()
train_p.join()
del train_p
gc.collect()
logger_handle.log("============ starting pit =============")
oldWins = 0
newWins = 0
with mp.Pool(configs.NUM_CPUS, maxtasksperchild=10) as pool:
for win in pool.imap_unordered(
partial(execute_pit, configs.BEST_PATH, configs.NEW_NET_PATH, exponent=configs.SIMS_EXPONENT,
multiplikator=configs.SIMS_FAKTOR),
[1 if pit_iter % 2 == 0 else -1 for pit_iter in range(configs.EVAL_EPISODES)]):
if win == 1:
newWins += 1
elif win == -1:
oldWins += 1
logger_handle.log("player won: " + str(win))
del pool
gc.collect()
if newWins < oldWins * configs.SCORING_THRESHOLD: # not better then previus
logger_handle.log("falling back to old network")
else:
logger_handle.log("new network accepted")
copy(configs.NEW_NET_PATH, configs.BEST_PATH)
if episode <= configs.MEMORY_ITERS:
current_mem_size = int(
configs.MIN_MEMORY + (configs.MAX_MEMORY - configs.MIN_MEMORY) * (episode / configs.MEMORY_ITERS))
logger_handle.log("changed mem size to " + str(current_mem_size))
mem.changeMaxSize(current_mem_size)
episode += 1
logger_handle.log("============ finisched AlphaZero ===========")
save_p = mp.Process(target=save_whole_net,
args=(configs.BEST_PATH, configs.INTERMEDIATE_SAVE_PATH + "models/whole_net"))
save_p.start()
save_p.join()
mem.saveState(episode, "finished_array.npy", "finisched_vars.obj")
except BaseException as e:
print(e.__doc__)
print(e)
logger_handle.log("============ interupted training ===========")
logger_handle.log(str(e.__doc__))
logger_handle.log(str(e))
save_p = mp.Process(target=save_whole_net,
args=(configs.BEST_PATH, configs.INTERMEDIATE_SAVE_PATH + "models/whole_net"))
save_p.start()
save_p.join()
mem.saveState(episode, "interrupt_array.npy", "interrupted_vars.obj")