---

Updated on 2021.09.30 DI-engine-v0.2.0 (beta)

Introduction to DI-engine (beta)

DI-engine is a generalized Decision Intelligence engine. It supports most basic deep reinforcement learning (DRL) algorithms, such as DQN, PPO, SAC, and domain-specific algorithms like QMIX in multi-agent RL, GAIL in inverse RL, and RND in exploration problems. Various training pipelines and customized decision AI applications are also supported. Have fun with exploration and exploitation.

Application

• DI-star
• DI-drive

Environment

• GoBigger

System Optimization and Design

• DI-orchestrator
• DI-hpc
• DI-store

Other

• DI-engine-docs
• treevalue
• DI-treetensor (preview)

Installation

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.

• base: opendilab/ding:nightly
• atari: opendilab/ding:nightly-atari
• mujoco: opendilab/ding:nightly-mujoco
• smac: opendilab/ding:nightly-smac

Documentation

The detailed documentation are hosted on doc(中文文档).

Quick Start

3 Minutes Kickoff

3 Minutes Kickoff(colab)

3 分钟上手中文版(kaggle)

Bonus: Train RL agent in one line code:

ding -m serial -e cartpole -p dqn -s 0

Feature

Algorithm Versatility

No	Algorithm	Label	Implementation	Runnable Demo
1	DQN		policy/dqn	python3 -u cartpole_dqn_main.py / ding -m serial -c cartpole_dqn_config.py -s 0
2	C51		policy/c51	ding -m serial -c cartpole_c51_config.py -s 0
3	QRDQN		policy/qrdqn	ding -m serial -c cartpole_qrdqn_config.py -s 0
4	IQN		policy/iqn	ding -m serial -c cartpole_iqn_config.py -s 0
5	Rainbow		policy/rainbow	ding -m serial -c cartpole_rainbow_config.py -s 0
6	SQL		policy/sql	ding -m serial -c cartpole_sql_config.py -s 0
7	R2D2		policy/r2d2	ding -m serial -c cartpole_r2d2_config.py -s 0
8	A2C		policy/a2c	ding -m serial -c cartpole_a2c_config.py -s 0
9	PPO		policy/ppo	python3 -u cartpole_ppo_main.py / ding -m serial_onpolicy -c cartpole_ppo_config.py -s 0
10	PPG		policy/ppg	python3 -u cartpole_ppg_main.py
11	ACER		policy/acer	ding -m serial -c cartpole_acer_config.py -s 0
12	IMPALA		policy/impala	ding -m serial -c cartpole_impala_config.py -s 0
13	DDPG		policy/ddpg	ding -m serial -c pendulum_ddpg_config.py -s 0
14	TD3		policy/td3	python3 -u pendulum_td3_main.py / ding -m serial -c pendulum_td3_config.py -s 0
15	SAC		policy/sac	ding -m serial -c pendulum_sac_config.py -s 0
16	QMIX		policy/qmix	ding -m serial -c smac_3s5z_qmix_config.py -s 0
17	COMA		policy/coma	ding -m serial -c smac_3s5z_coma_config.py -s 0
18	QTran		policy/qtran	ding -m serial -c smac_3s5z_qtran_config.py -s 0
19	WQMIX		policy/wqmix	ding -m serial -c smac_3s5z_wqmix_config.py -s 0
20	CollaQ		policy/collaq	ding -m serial -c smac_3s5z_collaq_config.py -s 0
21	GAIL		reward_model/gail	ding -m serial_reward_model -c cartpole_dqn_config.py -s 0
22	SQIL		entry/sqil	ding -m serial_sqil -c cartpole_sqil_config.py -s 0
23	DQFD		policy/dqfd	ding -m serial_dqfd -c cartpole_dqfd_config.py -s 0
24	HER		reward_model/her	python3 -u bitflip_her_dqn.py
25	RND		reward_model/rnd	python3 -u cartpole_ppo_rnd_main.py
26	CQL		policy/cql	python3 -u d4rl_cql_main.py
27	PER		worker/replay_buffer	rainbow demo
28	GAE		rl_utils/gae	ppo demo
29	D4PG		policy/d4pg	python3 -u pendulum_d4pg_config.py

means discrete action space, which is only label in normal DRL algorithms(1-15)

means continuous action space, which is only label in normal DRL algorithms(1-15)

means distributed training (collector-learner parallel) RL algorithm

means multi-agent RL algorithm

means RL algorithm which is related to exploration and sparse reward

means Imitation Learning, including Behaviour Cloning, Inverse RL, Adversarial Structured IL

means offline RL algorithm

means other sub-direction algorithm, usually as plugin-in in the whole pipeline

P.S: The .py file in Runnable Demo can be found in dizoo

Environment Versatility

No	Environment	Label	Visualization	dizoo link
1	atari			dizoo link
2	box2d/bipedalwalker			dizoo link
3	box2d/lunarlander			dizoo link
4	classic_control/cartpole			dizoo link
5	classic_control/pendulum			dizoo link
6	competitive_rl			dizoo link
7	gfootball			dizoo link
8	minigrid			dizoo link
9	mujoco			dizoo link
10	multiagent_particle			dizoo link
11	overcooked			dizoo link
12	procgen			dizoo link
13	pybullet			dizoo link
14	smac			dizoo link
15	d4rl			dizoo link
16	league_demo			dizoo link
17	pomdp atari			dizoo link
18	bsuite			dizoo link
19	ImageNet			dizoo link
20	slime_volleyball			dizoo link
21	gym_bybrid			dizoo link
22	GoBigger			dizoo link

means discrete action space

means continuous action space

means hybrid (discrete + continuous) action space

means multi-agent RL environment

means environment which is related to exploration and sparse reward

means offline RL environment

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

Contribution

We appreciate all contributions to improve DI-engine, both algorithms and system designs. Please refer to CONTRIBUTING.md for more guides. And our roadmap can be accessed by this link.

And users can join our slack communication channel or our forum for more detailed discussion.

For future plans or milestones, please refer to our GitHub Projects.

Citation

@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},
}

License

DI-engine released under the Apache 2.0 license.