RLLTE: Long-Term Evolution Project of Reinforcement Learning
Paper | Documentation | Tutorials | Forum | Benchmarks
- Overview
- Quick Start
- Function List (Part)
- RLLTE Ecosystem
- API Documentation
- Cite the Project
- How To Contribute
- Acknowledgment
- Miscellaneous
Inspired by the long-term evolution (LTE) standard project in telecommunications, aiming to provide development components for and standards for advancing RL research and applications. Beyond delivering top-notch algorithm implementations, RLLTE also serves as a toolkit for developing algorithms.
An introduction to RLLTE.
Why RLLTE?
- 𧬠Long-term evolution for providing latest algorithms and tricks;
- ποΈ Complete ecosystem for task design, model training, evaluation, and deployment (TensorRT, CANN, ...);
- 𧱠Module-oriented design for complete decoupling of RL algorithms;
- π Optimized workflow for full hardware acceleration;
- βοΈ Support custom environments and modules;
- π₯οΈ Support multiple computing devices like GPU and NPU;
- πΎ Large number of reusable benchmarks (RLLTE Hub);
- π¨β
βοΈ Large language model-empowered copilot (RLLTE Copilot).
β οΈ Since the construction of RLLTE Hub requires massive computing power, we have to upload the training datasets and model weights gradually. Progress report can be found in Issue#30.
See the project structure below:
For more detailed descriptions of these modules, see API Documentation.
- Prerequisites
Currently, we recommend Python>=3.8, and user can create an virtual environment by
conda create -n rllte python=3.8- with pip
recommended
Open a terminal and install rllte with pip:
pip install rllte-core # basic installation
pip install rllte-core[envs] # for pre-defined environments- with git
Open a terminal and clone the repository from GitHub with git:
git clone https://github.com/RLE-Foundation/rllte.gitAfter that, run the following command to install package and dependencies:
pip install -e . # basic installation
pip install -e .[envs] # for pre-defined environmentsFor more detailed installation instruction, see Getting Started.
RLLTE provides implementations for well-recognized RL algorithms and simple interface for building applications.
Suppose we want to use DrQ-v2 to solve a task of DeepMind Control Suite, and it suffices to write a train.py like:
# import `env` and `agent` module
from rllte.env import make_dmc_env
from rllte.agent import DrQv2
if __name__ == "__main__":
device = "cuda:0"
# create env, `eval_env` is optional
env = make_dmc_env(env_id="cartpole_balance", device=device)
eval_env = make_dmc_env(env_id="cartpole_balance", device=device)
# create agent
agent = DrQv2(env=env, eval_env=eval_env, device=device, tag="drqv2_dmc_pixel")
# start training
agent.train(num_train_steps=500000, log_interval=1000)Run train.py and you will see the following output:
Similarly, if we want to train an agent on HUAWEI NPU, it suffices to replace cuda with npu:
device = "cuda:0" -> device = "npu:0"Developers only need three steps to implement an RL algorithm with RLLTE. The following example illustrates how to write an Advantage Actor-Critic (A2C) agent to solve Atari games.
- Firstly, select a prototype:
from rllte.common.prototype import OnPolicyAgent- Secondly, select necessary modules to build the agent:
from rllte.xploit.encoder import MnihCnnEncoder
from rllte.xploit.policy import OnPolicySharedActorCritic
from rllte.xploit.storage import VanillaRolloutStorage
from rllte.xplore.distribution import Categorical- Run the
.describefunction of the selected policy and you will see the following output:
OnPolicySharedActorCritic.describe()
# Output:
# ================================================================================
# Name : OnPolicySharedActorCritic
# Structure : self.encoder (shared by actor and critic), self.actor, self.critic
# Forward : obs -> self.encoder -> self.actor -> actions
# : obs -> self.encoder -> self.critic -> values
# : actions -> log_probs
# Optimizers : self.optimizers['opt'] -> (self.encoder, self.actor, self.critic)
# ================================================================================This will illustrate the structure of the policy and indicate the optimizable parts. Finally, merge these modules and write an .update function:
from torch import nn
import torch as th
class A2C(OnPolicyAgent):
def __init__(self, env, tag, seed, device, num_steps) -> None:
super().__init__(env=env, tag=tag, seed=seed, device=device, num_steps=num_steps)
# create modules
encoder = MnihCnnEncoder(observation_space=env.observation_space, feature_dim=512)
policy = OnPolicySharedActorCritic(observation_space=env.observation_space,
action_space=env.action_space,
feature_dim=512,
opt_class=th.optim.Adam,
opt_kwargs=dict(lr=2.5e-4, eps=1e-5),
init_fn="xavier_uniform"
)
storage = VanillaRolloutStorage(observation_space=env.observation_space,
action_space=env.action_space,
device=device,
storage_size=self.num_steps,
num_envs=self.num_envs,
batch_size=256
)
dist = Categorical()
# set all the modules
self.set(encoder=encoder, policy=policy, storage=storage, distribution=dist)
def update(self):
for _ in range(4):
for batch in self.storage.sample():
# evaluate the sampled actions
new_values, new_log_probs, entropy = self.policy.evaluate_actions(obs=batch.observations, actions=batch.actions)
# policy loss part
policy_loss = - (batch.adv_targ * new_log_probs).mean()
# value loss part
value_loss = 0.5 * (new_values.flatten() - batch.returns).pow(2).mean()
# update
self.policy.optimizers['opt'].zero_grad(set_to_none=True)
(value_loss * 0.5 + policy_loss - entropy * 0.01).backward()
nn.utils.clip_grad_norm_(self.policy.parameters(), 0.5)
self.policy.optimizers['opt'].step()Then train the agent by
from rllte.env import make_atari_env
if __name__ == "__main__":
device = "cuda"
env = make_atari_env("PongNoFrameskip-v4", num_envs=8, seed=0, device=device)
agent = A2C(env=env, tag="a2c_atari", seed=0, device=device, num_steps=128)
agent.train(num_train_steps=10000000)As shown in this example, only a few dozen lines of code are needed to create RL agents with RLLTE.
RLLTE allows developers to replace settled modules of implemented algorithms to make performance comparison and algorithm improvement, and both
built-in and custom modules are supported. Suppose we want to compare the effect of different encoders, it suffices to invoke the .set function:
from rllte.xploit.encoder import EspeholtResidualEncoder
encoder = EspeholtResidualEncoder(...)
agent.set(encoder=encoder)RLLTE is an extremely open framework that allows developers to try anything. For more detailed tutorials, see Tutorials.
| Type | Algo. | Box | Dis. | M.B. | M.D. | M.P. | NPU | π° | π |
|---|---|---|---|---|---|---|---|---|---|
| On-Policy | A2C | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | β |
| On-Policy | PPO | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | β |
| On-Policy | DrAC | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ |
| On-Policy | DAAC | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | β |
| On-Policy | DrDAAC | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ | βοΈ |
| On-Policy | PPG | βοΈ | βοΈ | βοΈ | β | βοΈ | βοΈ | βοΈ | β |
| Off-Policy | DQN | βοΈ | β | β | β | βοΈ | βοΈ | βοΈ | β |
| Off-Policy | DDPG | βοΈ | β | β | β | βοΈ | βοΈ | βοΈ | β |
| Off-Policy | SAC | βοΈ | β | β | β | βοΈ | βοΈ | βοΈ | β |
| Off-Policy | SAC-Discrete | β | βοΈ | β | β | βοΈ | βοΈ | βοΈ | β |
| Off-Policy | TD3 | βοΈ | β | β | β | βοΈ | βοΈ | βοΈ | β |
| Off-Policy | DrQ-v2 | βοΈ | β | β | β | β | βοΈ | βοΈ | βοΈ |
| Distributed | IMPALA | βοΈ | βοΈ | β | β | βοΈ | β | β | β |
Dis., M.B., M.D.:Discrete,MultiBinary, andMultiDiscreteaction space;M.P.: Multi processing;- π: Developing;
- π°: Support intrinsic reward shaping;
- π: Support observation augmentation.
| Type | Modules |
|---|---|
| Count-based | PseudoCounts, RND |
| Curiosity-driven | ICM, GIRM, RIDE |
| Memory-based | NGU |
| Information theory-based | RE3, RISE, REVD |
See Tutorials: Use Intrinsic Reward and Observation Augmentation for usage examples.
Explore the ecosystem of RLLTE to facilitate your project:
- Hub: Fast training APIs and reusable benchmarks.
- Evaluation: Reasonable and reliable metrics for algorithm evaluation.
- Env: Packaged environments for fast invocation.
- Deployment: Convenient APIs for model deployment.
- Pre-training: Methods of pre-training in RL.
- Copilot: Large language model-empowered copilot.
View our well-designed documentation: https://docs.rllte.dev/
Welcome to contribute to this project! Before you begin writing code, please read CONTRIBUTING.md for guide first.
If you use RLLTE in your research, please cite this project like this:
@article{yuan2023rllte,
title={RLLTE: Long-Term Evolution Project of Reinforcement Learning},
author={Mingqi Yuan and Zequn Zhang and Yang Xu and Shihao Luo and Bo Li and Xin Jin and Wenjun Zeng},
year={2023},
journal={arXiv preprint arXiv:2309.16382}
}This project is supported by The Hong Kong Polytechnic University, Eastern Institute for Advanced Study, and FLW-Foundation. EIAS HPC provides a GPU computing platform, and HUAWEI Ascend Community provides an NPU computing platform for our testing. Some code of this project is borrowed or inspired by several excellent projects, and we highly appreciate them. See ACKNOWLEDGMENT.md.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent