Rainbow
=======
[![MIT License](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE.md)
Rainbow: Combining Improvements in Deep Reinforcement Learning [[1]](#references).
Results and pretrained models can be found in the [releases](https://github.com/Kaixhin/Rainbow/releases).
- [x] DQN [[2]](#references)
- [x] Double DQN [[3]](#references)
- [x] Prioritised Experience Replay [[4]](#references)
- [x] Dueling Network Architecture [[5]](#references)
- [x] Multi-step Returns [[6]](#references)
- [x] Distributional RL [[7]](#references)
- [x] Noisy Nets [[8]](#references)
Run the original Rainbow with the default arguments:
```
python main.py
```
Data-efficient Rainbow [[9]](#references) can be run using the following options (note that the "unbounded" memory is implemented here in practice by manually setting the memory capacity to be the same as the maximum number of timesteps):
```
python main.py --target-update 2000 \
--T-max 100000 \
--learn-start 1600 \
--memory-capacity 100000 \
--replay-frequency 1 \
--multi-step 20 \
--architecture data-efficient \
--hidden-size 256 \
--learning-rate 0.0001 \
--evaluation-interval 10000
```
Note that pretrained models from the [`1.3`](https://github.com/Kaixhin/Rainbow/releases/tag/1.3) release used a (slightly) incorrect network architecture. To use these, change the padding in the first convolutional layer from 0 to 1 (DeepMind uses "valid" (no) padding).
Requirements
------------
- [atari-py](https://github.com/openai/atari-py)
- [OpenCV Python](https://pypi.python.org/pypi/opencv-python)
- [Plotly](https://plot.ly/)
- [PyTorch](http://pytorch.org/)
To install all dependencies with Anaconda run `conda env create -f environment.yml` and use `source activate rainbow` to activate the environment.
Available Atari games can be found in the [`atari-py` ROMs folder](https://github.com/openai/atari-py/tree/master/atari_py/atari_roms).
Acknowledgements
----------------
- [@floringogianu](https://github.com/floringogianu) for [categorical-dqn](https://github.com/floringogianu/categorical-dqn)
- [@jvmancuso](https://github.com/jvmancuso) for [Noisy layer](https://github.com/pytorch/pytorch/pull/2103)
- [@jaara](https://github.com/jaara) for [AI-blog](https://github.com/jaara/AI-blog)
- [@openai](https://github.com/openai) for [Baselines](https://github.com/openai/baselines)
- [@mtthss](https://github.com/mtthss) for [implementation details](https://github.com/Kaixhin/Rainbow/wiki/Matteo's-Notes)
References
----------
[1] [Rainbow: Combining Improvements in Deep Reinforcement Learning](https://arxiv.org/abs/1710.02298)
[2] [Playing Atari with Deep Reinforcement Learning](http://arxiv.org/abs/1312.5602)
[3] [Deep Reinforcement Learning with Double Q-learning](http://arxiv.org/abs/1509.06461)
[4] [Prioritized Experience Replay](http://arxiv.org/abs/1511.05952)
[5] [Dueling Network Architectures for Deep Reinforcement Learning](http://arxiv.org/abs/1511.06581)
[6] [Reinforcement Learning: An Introduction](http://www.incompleteideas.net/sutton/book/ebook/the-book.html)
[7] [A Distributional Perspective on Reinforcement Learning](https://arxiv.org/abs/1707.06887)
[8] [Noisy Networks for Exploration](https://arxiv.org/abs/1706.10295)
[9] [When to Use Parametric Models in Reinforcement Learning?](https://arxiv.org/abs/1906.05243)
没有合适的资源?快使用搜索试试~ 我知道了~
Rainbow:Rainbow:结合深度强化学习的改进
共13个文件
py:6个
md:4个
gitignore:1个
5星 · 超过95%的资源 需积分: 49 21 下载量 45 浏览量
2021-02-04
22:49:26
上传
评论 3
收藏 17KB ZIP 举报
温馨提示
彩虹 Rainbow:结合深度强化学习的改进 。 结果和预先训练的模型可以在找到。 DQN Double DQN 优先体验重播 决斗网络体系结构 多步骤退货 分布式RL 吵网 使用默认参数运行原始Rainbow: python main.py 可以使用以下选项运行数据有效的Rainbow (请注意,实际上,此处通过手动设置内存容量与最大时间步数相同来实现“无界”内存): python main.py --target-update 2000 \ --T-max 100000 \ --learn-star
资源详情
资源评论
资源推荐
收起资源包目录
Rainbow-master.zip (13个子文件)
Rainbow-master
env.py 3KB
agent.py 6KB
results
README.md 76B
test.py 3KB
main.py 9KB
model.py 4KB
requirements.txt 41B
CONTRIBUTING.md 930B
environment.yml 126B
LICENSE.md 1KB
README.md 3KB
.gitignore 1KB
memory.py 10KB
共 13 条
- 1
NinglingPan
- 粉丝: 19
- 资源: 4645
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功
评论1