fangchenliu/maskdp_data
收藏资源简介:
该数据集用于论文《Masked Autoencoding for Scalable and Generalizable Decision Making》,包含四种类型的数据:无监督(使用纯内在奖励的训练经验)、半监督(使用内在奖励加外在奖励的训练经验)、监督(使用纯外在奖励的训练经验)和专家数据(来自接近专家策略的TD3检查点的数据)。数据集的组织格式包括训练数据和评估数据,分别存储在不同的文件夹中。
This dataset is intended for the paper titled *Masked Autoencoding for Scalable and Generalizable Decision Making*. It contains four types of data: unsupervised data (training experiences using purely intrinsic rewards), semi-supervised data (training experiences using both intrinsic and extrinsic rewards), supervised data (training experiences using purely extrinsic rewards), and expert data (data sourced from TD3 checkpoints that approximate expert policies). The dataset is structured into training and evaluation subsets, which are stored in separate directories.
数据集概述
数据集名称
Dataset for Masked Autoencoding for Scalable and Generalizable Decision Making
数据集来源
该数据集用于论文《Masked Autoencoding for Scalable and Generalizable Decision Making》。
数据集格式
数据集的组织结构如下:
├── maskdp_train │ ├── cheetah │ │ ├── expert # 来自TD3策略的近专家轨迹 | | | ├── cheetah_run | | | | ├── 0.npy | | | | ├── 1.npy | | | | ├── ... | | | ├── cheetah_run_backwards │ │ ├── sup # 监督数据,完整经验回放与外在奖励 | | | ├── cheetah_run | | | ├── cheetah_run_backwards │ │ ├── semi # 半监督数据,完整经验回放与内在+外在奖励 | | | ├── cheetah_run | | | ├── cheetah_run_backwards │ │ ├── unsup # 无监督数据,完整经验回放与内在奖励 | | | ├── 0.npy | | | ├── 1.npy | | | ├── ... │ ├── walker ... │ ├── quadruped ... ├── maskdp_eval │ ├── expert │ │ ├── cheetah_run │ │ ├── cheetah_run_backwards │ │ ├── ... │ │ ├── walker_stand │ │ ├── quadruped_walk │ │ ├── ... │ ├── unsup │ │ ├── cheetah │ │ ├── walker │ │ ├── quadruped
数据类型
数据集包含以下四种类型:
- 无监督数据(仅使用内在奖励的训练经验)
- 半监督数据(使用内在奖励+外在奖励的训练经验)
- 监督数据(仅使用外在奖励的训练经验)
- 专家数据(来自近专家TD3检查点的轨迹)




