if end:
end_user.add(user)
env.set_state(user, s_)
reward += r
memory.store_transition(s, a, r, s_)
# 网络训练和参数更新
b_s, b_a, b_r, b_s_ = memory.sample(BATCH_SIZE)
q_ = target_AC.get_q(b_s_)
target_q = b_r + GAMMA * q_
train_AC.train(target_q, s=b_s)
target_AC.set_params(train_AC.a_params, train_AC.c_params, t=0.2)
# 统计用户的指标
user_auc = 0
for user in range(1, 944):
if user not in end_user:
items = env._find_positive_history(user)
s = env.state[user]
a = train_AC.get_action(s)
all_label_is, auc = env.auc(user, a)
if auc >= 0:
user_auc += auc
elif all_label_is == 0:
neg_user.add(user)
else:
pos_user.add(user)
if 943 - len(end_user) - len(neg_user) - len(pos_user) == 0:
break
# 记录各项指标
all_neg_user_num.append(len(neg_user))
all_pos_user_num.append(len(pos_user))
end_user_num.append(len(end_user))
