# ====================
# 학습 사이클 실행
# ====================
# 패키지 임포트
from dual_network import dual_network
from self_play import self_play
from train_network import train_network
from evaluate_network import evaluate_network
from evaluate_best_player import evaluate_best_player
# 듀얼 네트워크 생성
dual_network()
for i in range(10):
print('Train', i, '====================')
# 셀프 플레이 파트
self_play()
# 파라미터 갱신 파트
train_network()
# 신규 파라미터 평가 파트
update_best_player = evaluate_network()
# 베스트 플레이어 평가
if update_best_player:
evaluate_best_player()
from dual_network import dual_network
from self_play import self_play
from train_network import train_network
from evaluate_network import evaluate_network
# from evaluate_best_player import evaluate_best_player
dual_network()
for i in range(10):
print('Train', i, '===================')
self_play()
train_network()
evaluate_network()
from evaluate_network import evaluate_network
from train_network import train_network
from self_play import self_play
for i in range(10):
print('Train', i, '====================')
self_play() # セルフプレイ部
train_network() # パラメータ更新部
evaluate_network() # 新パラメータ評価部
from evaluate_network import evaluate_network, update_best_player
from evaluate_best_player import evaluate_best_player
dual_network()
count = 0
fail_count = 0
for i in range(30):
print('Train', i, '======================')
self_play(fail_count)
train_network(fail_count)
#skip = True
#updated = True
#if i%10 == 0 and i != 0:
updated = evaluate_network()
# skip = False
#else:
# update_best_player()
if updated == True: # and skip == False:
count += 1
fail_count = 0
else:
fail_count += 1
if count > 4:
evaluate_best_player()
count = 0
if fail_count > 3:
print("mean var: ", np.mean(np.var(acc, axis=-1)))
elif args.validate == "pfnn_random":
pfnn = pfnn_random_layers.load(
target_file,
args.dimension_random_noise,
args.layers_add_random_noise,
args.params_random_noise,
args.sample_noise_each_phase
)
pfnn.start_tf()
dataset_config_file = args.dataset #"data/dataset.json"
with open(dataset_config_file, "r") as f:
config_store = json.load(f)
c = DirectionalController(pfnn, config_store)
evaluate_network(pfnn, config_store)