We provide examples of different algorithms in different domains.
We provide a set of Benchmarked RL algorithms and also simplified versions to allow to well understand the principle of the library. These algorithms can be easily modified to start a new project.
- A2C Algorithms are used as a tutorial to present different functionalities of
salina - REINFORCE shows a simple implementation of REINFORCE with
salina. - DQN proposes implementations of Q-Learning algorithms
- TD3
- DDPG
- PPO on Brax
- PPO Continuous actions
- PPO Discrete actions
- (more to come, please do a Pull request if you implement other algorithms)
Note that due to the modularity of salina, all the implementations work with any type of policies (recurrent, hierarchical, transformers,....)
Based on D4RL, we provide implementations of algorithms for off-policy learning
- BC Behavioral Cloning
See Computer Vision Algorithms
Under progress...
It is a simple script to compute the execution speed of any agent. It allows to optimize experiments by choosing the right ratio between the batch size, the time size and the number of processes