Deck Building Games (Pokemon is the ultra goal, similar to this) 卡组构建游戏（口袋妖怪是最终目标，类似于此） Model: Agent will firstly select tools/cards from a big set. Then play with that custom set. 模型：智能体首先从一个大型集合中选择工具/卡牌，然后使用该自定义集合进行游戏。 things to consider: Single card strength, short time combo, long time combo 考虑事项：单卡强度、短期连携、长期连携 idea 1: Play with all cards available, then choose from set: too many cards, maybe hundreds length vector. 想法1：使用所有可用卡牌进行游戏，然后从中选择——卡牌数量太多，可能是一个数百维的向量。 idea 2: Use vector database to find the inner connection between cards. 想法2：使用向量数据库寻找卡牌之间的内在联系。 idea 3: First train play model, then cards selection model. Hard to train play model with unknown set. 想法3：先训练游戏模型，再训练卡牌选择模型。由于卡组未知，训练游戏模型较为困难。