polish(nyz): polish dqn and ppo comments #732
Conversation
PaParaZz1
changed the title
| """ | ||
| # Data preprocessing operations, such as stack data, cpu to cuda device | ||
| data = default_preprocess_learn( |
There was a problem hiding this comment.
数据预处理这里或许可以详细说明，里面做了哪些操作
| R2D2 proposes that several tricks should be used to improve upon DRQN, | ||
| namely some recurrent experience replay tricks such as burn-in. | ||
| R2D2 proposes that several tricks should be used to improve upon DRQN, namely some recurrent experience replay \ | ||
| tricks and the burn-in mechanism for off-policy training. |
There was a problem hiding this comment.
The R2D2 policy class is inspired by the paper "Recurrent Experience Replay in Distributed Reinforcement Learning". R2D2 suggests the incorporation of several enhancements over DRQN, specifically the application of novel recurrent experience replay strategies and the implementation of a burn-in mechanism for off-policy training.
| - data (:obj:`List[Dict[str, Any]`): The trajectory data(a list of transition), each element is the same \ | ||
| format as the return value of ``self._process_transition`` method. | ||
| - transitions (:obj:`List[Dict[str, Any]`): The trajectory data (a list of transition), each element is \ | ||
| the same format as the return value of ``self._process_transition`` method. | ||
| Returns: |
There was a problem hiding this comment.
The trajectory data, which is a list of transitions. Each element is in the same format as the return value of the self._process_transition method.
| And the user can customize the this data processing procecure by overriding this two methods and collector \ | ||
| itself. | ||
| - samples (:obj:`List[Dict[str, Any]]`): The processed train samples, each element is the similar format \ | ||
| as input transitions, but may contain more data for training, such as nstep reward and target obs. | ||
| """ |
There was a problem hiding this comment.
The processed training samples. Each element is similar in format to the input transitions, but may contain additional data for training, such as n-step reward and target observations.
Description
Related Issue
TODO
Check List