Updated on 2023.04.11 DI-engine-v0.4.7
Documentation | 中文文档 | Tutorials | Feature | Task & Middleware | TreeTensor | Roadmap
DI-engine is a generalized decision intelligence engine for PyTorch and JAX.
It provides python-first and asynchronous-native task and middleware abstractions, and modularly integrates several of the most important decision-making concepts: Env, Policy and Model. Based on the above mechanisms, DI-engine supports various deep reinforcement learning algorithms with superior performance, high efficiency, well-organized documentation and unittest:
- Most basic DRL algorithms, such as DQN, PPO, SAC, R2D2, IMPALA
- Multi-agent RL algorithms like QMIX, MAPPO, ACE
- Imitation learning algorithms (BC/IRL/GAIL) , such as GAIL, SQIL, Guided Cost Learning, Implicit BC
- Offline RL algorithms: CQL, TD3BC, Decision Transformer
- Model-based RL algorithms: SVG, MVE, STEVE / MBPO, DDPPO
- Exploration algorithms like HER, RND, ICM, NGU
DI-engine aims to standardize different Decision Intelligence environments and applications, supporting both academic research and prototype applications. Various training pipelines and customized decision AI applications are also supported:
(Click to Collapse)
- Traditional academic environments
- DI-zoo: various decision intelligence demonstrations and benchmark environments with DI-engine.
- Tutorial courses
- PPOxFamily: PPO x Family DRL Tutorial Course
- Real world decision AI applications
- DI-star: Decision AI in StarCraftII
- DI-drive: Auto-driving platform
- GoBigger: [ICLR 2023] Multi-Agent Decision Intelligence Environment
- DI-sheep: Decision AI in 3 Tiles Game
- DI-smartcross: Decision AI in Traffic Light Control
- DI-bioseq: Decision AI in Biological Sequence Prediction and Searching
- DI-1024: Deep Reinforcement Learning + 1024 Game
- Research paper
- InterFuser: [CoRL 2022] Safety-Enhanced Autonomous Driving Using Interpretable Sensor Fusion Transformer
- ACE: [AAAI 2023] ACE: Cooperative Multi-agent Q-learning with Bidirectional Action-Dependency
- Docs and Tutorials
- DI-engine-docs: Tutorials, best practice and the API reference.
- awesome-model-based-RL: A curated list of awesome Model-Based RL resources
- awesome-exploration-RL: A curated list of awesome exploration RL resources
- awesome-decision-transformer: A curated list of Decision Transformer resources
- awesome-RLHF: A curated list of reinforcement learning with human feedback resources
- awesome-multi-modal-reinforcement-learning: A curated list of Multi-Modal Reinforcement Learning resources
- awesome-AI-based-protein-design: a collection of research papers for AI-based protein design
- awesome-diffusion-model-in-rl: A curated list of Diffusion Model in RL resources
- awesome-end-to-end-autonomous-driving: A curated list of awesome End-to-End Autonomous Driving resources
- awesome-driving-behavior-prediction: A collection of research papers for Driving Behavior Prediction
On the low-level end, DI-engine comes with a set of highly re-usable modules, including RL optimization functions, PyTorch utilities and auxiliary tools.
BTW, DI-engine also has some special system optimization and design for efficient and robust large-scale RL training:
(Click for Details)
- treevalue: Tree-nested data structure
- DI-treetensor: Tree-nested PyTorch tensor Lib
- DI-orchestrator: RL Kubernetes Custom Resource and Operator Lib
- DI-hpc: RL HPC OP Lib
- DI-store: RL Object Store
Have fun with exploration and exploitation.
- Introduction to DI-engine
- Outline
- Installation
- Quick Start
- Feature
- Feedback and Contribution
- Supporters
- Citation
- License
You can simply install DI-engine from PyPI with the following command:
pip install DI-engine
If you use Anaconda or Miniconda, you can install DI-engine from conda-forge through the following command:
conda install -c opendilab di-engine
For more information about installation, you can refer to installation.
And our dockerhub repo can be found here,we prepare base image
and env image
with common RL environments.
(Click for Details)
- base: opendilab/ding:nightly
- rpc: opendilab/ding:nightly-rpc
- atari: opendilab/ding:nightly-atari
- mujoco: opendilab/ding:nightly-mujoco
- dmc: opendilab/ding:nightly-dmc2gym
- metaworld: opendilab/ding:nightly-metaworld
- smac: opendilab/ding:nightly-smac
- grf: opendilab/ding:nightly-grf
- cityflow: opendilab/ding:nightly-cityflow
- evogym: opendilab/ding:nightly-evogym
- d4rl: opendilab/ding:nightly-d4rl
The detailed documentation are hosted on doc | 中文文档.
How to migrate a new RL Env | 如何迁移一个新的强化学习环境
How to customize the neural network model | 如何定制策略使用的神经网络模型
Bonus: Train RL agent in one line code:
ding -m serial -e cartpole -p dqn -s 0
(Click to Collapse)
discrete means discrete action space, which is only label in normal DRL algorithms (1-23)
means continuous action space, which is only label in normal DRL algorithms (1-23)
means hybrid (discrete + continuous) action space (1-23)
Distributed Reinforcement Learning|分布式强化学习
Multi-Agent Reinforcement Learning|多智能体强化学习
Exploration Mechanisms in Reinforcement Learning|强化学习中的探索机制
Offiline Reinforcement Learning|离线强化学习
Model-Based Reinforcement Learning|基于模型的强化学习
means other sub-direction algorithms, usually as plugin-in in the whole pipeline
P.S: The .py
file in Runnable Demo
can be found in dizoo
No. | Algorithm | Label | Doc and Implementation | Runnable Demo |
---|---|---|---|---|
1 | DQN | DQN doc DQN中文文档 policy/dqn |
python3 -u cartpole_dqn_main.py / ding -m serial -c cartpole_dqn_config.py -s 0 | |
2 | C51 | C51 doc policy/c51 |
ding -m serial -c cartpole_c51_config.py -s 0 | |
3 | QRDQN | QRDQN doc policy/qrdqn |
ding -m serial -c cartpole_qrdqn_config.py -s 0 | |
4 | IQN | IQN doc policy/iqn |
ding -m serial -c cartpole_iqn_config.py -s 0 | |
5 | FQF | FQF doc policy/fqf |
ding -m serial -c cartpole_fqf_config.py -s 0 | |
6 | Rainbow | Rainbow doc policy/rainbow |
ding -m serial -c cartpole_rainbow_config.py -s 0 | |
7 | SQL | SQL doc policy/sql |
ding -m serial -c cartpole_sql_config.py -s 0 | |
8 | R2D2 | R2D2 doc policy/r2d2 |
ding -m serial -c cartpole_r2d2_config.py -s 0 | |
9 | PG | PG doc policy/pg |
ding -m serial -c cartpole_pg_config.py -s 0 | |
10 | A2C | A2C doc policy/a2c |
ding -m serial -c cartpole_a2c_config.py -s 0 | |
11 | PPO/MAPPO | PPO doc policy/ppo |
python3 -u cartpole_ppo_main.py / ding -m serial_onpolicy -c cartpole_ppo_config.py -s 0 | |
12 | PPG | PPG doc policy/ppg |
python3 -u cartpole_ppg_main.py | |
13 | ACER | ACER doc policy/acer |
ding -m serial -c cartpole_acer_config.py -s 0 | |
14 | IMPALA | IMPALA doc policy/impala |
ding -m serial -c cartpole_impala_config.py -s 0 | |
15 | DDPG/PADDPG | DDPG doc policy/ddpg |
ding -m serial -c pendulum_ddpg_config.py -s 0 | |
16 | TD3 | TD3 doc policy/td3 |
python3 -u pendulum_td3_main.py / ding -m serial -c pendulum_td3_config.py -s 0 | |
17 | D4PG | D4PG doc policy/d4pg |
python3 -u pendulum_d4pg_config.py | |
18 | SAC/[MASAC] | SAC doc policy/sac |
ding -m serial -c pendulum_sac_config.py -s 0 | |
19 | PDQN | policy/pdqn | ding -m serial -c gym_hybrid_pdqn_config.py -s 0 | |
20 | MPDQN | policy/pdqn | ding -m serial -c gym_hybrid_mpdqn_config.py -s 0 | |
21 | HPPO | policy/ppo | ding -m serial_onpolicy -c gym_hybrid_hppo_config.py -s 0 | |
22 | BDQ | policy/bdq | python3 -u hopper_bdq_config.py | |
23 | MDQN | policy/mdqn | python3 -u asterix_mdqn_config.py | |
24 | QMIX | QMIX doc policy/qmix |
ding -m serial -c smac_3s5z_qmix_config.py -s 0 | |
25 | COMA | COMA doc policy/coma |
ding -m serial -c smac_3s5z_coma_config.py -s 0 | |
26 | QTran | policy/qtran | ding -m serial -c smac_3s5z_qtran_config.py -s 0 | |
27 | WQMIX | WQMIX doc policy/wqmix |
ding -m serial -c smac_3s5z_wqmix_config.py -s 0 | |
28 | CollaQ | CollaQ doc policy/collaq |
ding -m serial -c smac_3s5z_collaq_config.py -s 0 | |
29 | MADDPG | MADDPG doc policy/ddpg |
ding -m serial -c ant_maddpg_config.py -s 0 | |
30 | GAIL | GAIL doc reward_model/gail |
ding -m serial_gail -c cartpole_dqn_gail_config.py -s 0 | |
31 | SQIL | SQIL doc entry/sqil |
ding -m serial_sqil -c cartpole_sqil_config.py -s 0 | |
32 | DQFD | DQFD doc policy/dqfd |
ding -m serial_dqfd -c cartpole_dqfd_config.py -s 0 | |
33 | R2D3 | R2D3 doc R2D3中文文档 policy/r2d3 |
python3 -u pong_r2d3_r2d2expert_config.py | |
34 | Guided Cost Learning | Guided Cost Learning中文文档 reward_model/guided_cost |
python3 lunarlander_gcl_config.py | |
35 | TREX | TREX doc reward_model/trex |
python3 mujoco_trex_main.py | |
36 | Implicit Behavorial Cloning (DFO+MCMC) | policy/ibc model/template/ebm |
python3 d4rl_ibc_main.py -s 0 -c pen_human_ibc_mcmc_config.py | |
37 | BCO | entry/bco | python3 -u cartpole_bco_config.py | |
38 | HER | HER doc reward_model/her |
python3 -u bitflip_her_dqn.py | |
39 | RND | RND doc reward_model/rnd |
python3 -u cartpole_rnd_onppo_config.py | |
40 | ICM | ICM doc ICM中文文档 reward_model/icm |
python3 -u cartpole_ppo_icm_config.py | |
41 | CQL | CQL doc policy/cql |
python3 -u d4rl_cql_main.py | |
42 | TD3BC | TD3BC doc policy/td3_bc |
python3 -u d4rl_td3_bc_main.py | |
43 | Decision Transformer | policy/dt | python3 -u d4rl_dt_main.py | |
44 | MBSAC(SAC+MVE+SVG) | policy/mbpolicy/mbsac | python3 -u pendulum_mbsac_mbpo_config.py \ python3 -u pendulum_mbsac_ddppo_config.py | |
45 | STEVESAC(SAC+STEVE+SVG) | policy/mbpolicy/mbsac | python3 -u pendulum_stevesac_mbpo_config.py | |
46 | MBPO | MBPO doc world_model/mbpo |
python3 -u pendulum_sac_mbpo_config.py | |
47 | DDPPO | world_model/ddppo | python3 -u pendulum_mbsac_ddppo_config.py | |
48 | PER | worker/replay_buffer | rainbow demo |
|
49 | GAE | rl_utils/gae | ppo demo |
|
50 | ST-DIM | torch_utils/loss/contrastive_loss | ding -m serial -c cartpole_dqn_stdim_config.py -s 0 | |
51 | PLR | PLR doc data/level_replay/level_sampler |
python3 -u bigfish_plr_config.py -s 0 | |
52 | PCGrad | torch_utils/optimizer_helper/PCGrad | python3 -u multi_mnist_pcgrad_main.py -s 0 |
(Click to Collapse)
means hybrid (discrete + continuous) action space
means multi-agent RL environment
means environment which is related to exploration and sparse reward
means Imitation Learning or Supervised Learning Dataset
means environment that allows agent VS agent battle
P.S. some enviroments in Atari, such as MontezumaRevenge, are also sparse reward type
DI-engine utilizes TreeTensor as the basic data container in various components, which is ease of use and consistent across different code modules such as environment definition, data processing and DRL optimization. Here are some concrete code examples:
-
TreeTensor can easily extend all the operations of
torch.Tensor
to nested data:(Click for Details)
import treetensor.torch as ttorch # create random tensor data = ttorch.randn({'a': (3, 2), 'b': {'c': (3, )}}) # clone+detach tensor data_clone = data.clone().detach() # access tree structure like attribute a = data.a c = data.b.c # stack/cat/split stacked_data = ttorch.stack([data, data_clone], 0) cat_data = ttorch.cat([data, data_clone], 0) data, data_clone = ttorch.split(stacked_data, 1) # reshape data = data.unsqueeze(-1) data = data.squeeze(-1) flatten_data = data.view(-1) # indexing data_0 = data[0] data_1to2 = data[1:2] # execute math calculations data = data.sin() data.b.c.cos_().clamp_(-1, 1) data += data ** 2 # backward data.requires_grad_(True) loss = data.arctan().mean() loss.backward() # print shape print(data.shape) # result # <Size 0x7fbd3346ddc0> # ├── 'a' --> torch.Size([1, 3, 2]) # └── 'b' --> <Size 0x7fbd3346dd00> # └── 'c' --> torch.Size([1, 3])
-
TreeTensor can make it simple yet effective to implement classic deep reinforcement learning pipeline
(Click for Details)
import torch import treetensor.torch as ttorch B = 4 def get_item(): return { 'obs': { 'scalar': torch.randn(12), 'image': torch.randn(3, 32, 32), }, 'action': torch.randint(0, 10, size=(1,)), 'reward': torch.rand(1), 'done': False, } data = [get_item() for _ in range(B)] # execute `stack` op - def stack(data, dim): - elem = data[0] - if isinstance(elem, torch.Tensor): - return torch.stack(data, dim) - elif isinstance(elem, dict): - return {k: stack([item[k] for item in data], dim) for k in elem.keys()} - elif isinstance(elem, bool): - return torch.BoolTensor(data) - else: - raise TypeError("not support elem type: {}".format(type(elem))) - stacked_data = stack(data, dim=0) + data = [ttorch.tensor(d) for d in data] + stacked_data = ttorch.stack(data, dim=0) # validate - assert stacked_data['obs']['image'].shape == (B, 3, 32, 32) - assert stacked_data['action'].shape == (B, 1) - assert stacked_data['reward'].shape == (B, 1) - assert stacked_data['done'].shape == (B,) - assert stacked_data['done'].dtype == torch.bool + assert stacked_data.obs.image.shape == (B, 3, 32, 32) + assert stacked_data.action.shape == (B, 1) + assert stacked_data.reward.shape == (B, 1) + assert stacked_data.done.shape == (B,) + assert stacked_data.done.dtype == torch.bool
-
File an issue on Github
-
Open or participate in our forum
-
Discuss on DI-engine slack communication channel
-
Discuss on DI-engine's QQ group (700157520) or add us on WeChat
-
Contact our email ([email protected])
-
Contributes to our future plan Roadmap
We appreciate all the feedbacks and contributions to improve DI-engine, both algorithms and system designs. And CONTRIBUTING.md
offers some necessary information.
@misc{ding,
title={{DI-engine: OpenDILab} Decision Intelligence Engine},
author={DI-engine Contributors},
publisher = {GitHub},
howpublished = {\url{https://github.com/opendilab/DI-engine}},
year={2021},
}
DI-engine released under the Apache 2.0 license.