target_model = Model()
memory = Memory() # メモリ
agent = Agent() # エージェント
learned_flag = False # 学習が終了したか否か
for epoch in range(N_EPOCHS):
if learned_flag:
break
print('Epoch: {}'.format(epoch + 1))
# 初期状態設定
env.reset()
state, reward, finished = env.random_step()
target_model.model.set_weights(main_model.model.get_weights())
while not finished:
action = agent.get_action(state, epoch, main_model)
learned_flag = env.is_learned()
next_state, reward, finished = env.step(action.argmax())
memory.add((state, action, reward, next_state))
state = next_state
if len(memory.buffer) > S_BATCH:
main_model.replay(memory, S_BATCH, GAMMA, target_model)
target_model.model.set_weights(main_model.model.get_weights())
