使用说明
依赖项
安装包装 requirements.txt 文件
conda create -n pbrl python=3.10
conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia
conda install tensorboard ipykernel matplotlib seaborn
pip install "gym[mujoco_py,classic_control]==0.23.0"
pip install pyrallis tqdm
pip install git+https://github.com/Farama-Foundation/Metaworld.git@master#egg=metaworld
pip install -r requirements.txt
数据集
元世界 medium-replay 数据集可在官方存储库中获得 里拉.
训练
通过修改配置文件设置学习率、网络架构、批处理大小和其他算法超参数。
要训练奖励模型：
python train/learn_reward.py --config=configs/medium-replay/task-name-v2/reward.yaml
To train transition model,
python train/learn_transition.py--配置=配置/媒体回放/任务名称-v2/transition.yaml
To run PbRL algorithm,
python main.py--配置=配置/媒体回放/任务名称-v2/pbrl.yaml
## Results
The training results are stored in `log/`.
## Reference
Our code is based on the official implementation of \<APPO : Adversarial Preference-based Policy Optimization\> : [https://github.com/oh-lab/APPO](https://github.com/oh-lab/APPO)