https://github.com/NM512/dreamerv3-torch/issues/18
参考 DeepMind Dreamer 在这个任务上栽了 这个任务考验AI记忆能力的泛化
欢迎改进提升AI的记忆能力;
{"step": 601000, "dataset_size": 300500.0, "train_return": 6.0, "train_length": 500.0, "train_episodes": 601.0}
{"step": 704000, "dataset_size": 352000.0, "train_return": 6.0, "train_length": 500.0, "train_episodes": 704.0}
{"step": 454000, "dataset_size": 227000.0, "train_return": 6.0, "train_length": 500.0, "train_episodes": 454.0}
{"step": 528000, "dataset_size": 264000.0, "train_return": 6.0, "train_length": 500.0, "train_episodes": 528.0}
{"step": 545000, "eval_return": 2.9, "eval_length": 500.0, "eval_episodes": 10.0}
{"step": 555000, "dataset_size": 277500.0, "train_return": 6.0, "train_length": 500.0, "train_episodes": 555.0}
{"step": 581000, "dataset_size": 290500.0, "train_return": 6.0, "train_length": 500.0, "train_episodes": 581.0}
{"step": 608000, "dataset_size": 304000.0, "train_return": 6.0, "train_length": 500.0, "train_episodes": 608.0}
{"step": 616000, "dataset_size": 308000.0, "train_return": 6.0, "train_length": 500.0, "train_episodes": 616.0}
{"step": 649000, "dataset_size": 324500.0, "train_return": 6.0, "train_length": 500.0, "train_episodes": 649.0}
{"step": 693000, "dataset_size": 346500.0, "train_return": 6.0, "train_length": 500.0, "train_episodes": 693.0}
torch版本效果对比:
相关推荐:
code:通过进化、可塑性和 元 元学习 获得认知能力(4个时间维度的学习迭代)
代码:Learning to Learn and Forget (华为)