NathanGavenski/How-Resilient-are-Imitation-Learning-Methods-to-Sub-Optimal-Experts

---
annotations_creators:
- machine-generated
language_creators:
- expert-generated
language: []
license:
- mit
multilinguality: []
size_categories:
- 100B<n<1T
source_datasets:
- original
task_categories:
- other
task_ids: []
pretty_name: How Resilient are Imitation Learning Methods to Sub-Optimal Experts?
tags:
- Imitation Learning
- Expert Trajectories
- Classic Control
---

# How Resilient are Imitation Learning Methods to Sub-Optimal Experts?

## Related Work
Trajectories used in [How Resilient are Imitation Learning Methods to Sub-Optimal Experts?]()
The code that uses this data is on GitHub: https://github.com/NathanGavenski/How-resilient-IL-methods-are

# Structure
These trajectories are formed by using [Stable Baselines](https://stable-baselines.readthedocs.io/en/master/).
Each file is a dictionary of a set of trajectories with the following keys:

* actions: the action in the given timestamp `t`
* obs: current state in the given timestamp `t`
* rewards: reward retrieved after the action in the given timestamp `t`
* episode_returns: The aggregated reward of each episode (each file consists of 5000 runs)
* episode_Starts: Whether that `obs` is the first state of an episode (boolean list)

## Citation Information
```
@inproceedings{gavenski2022how,
title={How Resilient are Imitation Learning Methods to Sub-Optimal Experts?},
author={Nathan Gavenski and Juarez Monteiro and Adilson Medronha and Rodrigo Barros},
booktitle={2022 Brazilian Conference on Intelligent Systems (BRACIS)},
year={2022},
organization={IEEE}
}
```

## Contact:
- [Nathan Schneider Gavenski]([e-mail])
- [Juarez Monteiro]([e-mail])
- [Adilson Medronha]([e-mail])
- [Rodrigo C. Barros]([e-mail])

原始信息汇总

数据集概述

基本信息

• 数据集名称: How Resilient are Imitation Learning Methods to Sub-Optimal Experts?
• 注释创建者: 机器生成
• 语言创建者: 专家生成
• 许可证: MIT
• 大小分类: 100B<n<1T
• 源数据集: 原始数据
• 任务类别: 其他
• 美观名称: How Resilient are Imitation Learning Methods to Sub-Optimal Experts?
• 标签:
  • Imitation Learning
  • Expert Trajectories
  • Classic Control

数据结构

• 数据由Stable Baselines生成，每个文件包含以下键值：
  • actions: 给定时间戳t的动作
  • obs: 给定时间戳t的当前状态
  • rewards: 给定时间戳t执行动作后的奖励
  • episode_returns: 每个情节的总奖励（每个文件包含5000次运行）
  • episode_Starts: 指示obs是否为情节的第一个状态（布尔列表）

引用信息

@inproceedings{gavenski2022how, title={How Resilient are Imitation Learning Methods to Sub-Optimal Experts?}, author={Nathan Gavenski and Juarez Monteiro and Adilson Medronha and Rodrigo Barros}, booktitle={2022 Brazilian Conference on Intelligent Systems (BRACIS)}, year={2022}, organization={IEEE} }