stable-baselines3 VS agents

The latest release (v3.0.0) of Upkie's software brings a functional sim-to-real reinforcement learning pipeline based on Stable Baselines3, with standard sim-to-real tricks. The pipeline trains on the Gymnasium environments distributed in upkie.envs (setup: pip install upkie) and is implemented in the PPO balancer. Here is a policy running on an Upkie:

PettingZoo 1.24.0 is now live! This release includes Python 3.11 support, updated Chess and Hanabi environment versions, and many bugfixes, documentation updates and testing expansions. We are also very excited to announce 3 tutorials using Stable-Baselines3, and a full training script using CleanRL with TensorBoard and WandB.

I am wondering why there is so few algorithms in Stable Baselines 3 (SB3, https://github.com/DLR-RM/stable-baselines3/tree/master)? I was expecting some algorithms like ICM, HIRO, DIAYN, ... Why there is no model-based, skill-chaining, hierarchical-RL, ... algorithms implemented there?

Discord is more focused, and they have a page for people who wants to contribute https://github.com/DLR-RM/stable-baselines3/blob/master/CONTRIBUTING.md

Therefore, I debugged this error to the ReplayBuffer that was imported from `SB3`. This is the problem function -

Have you ever wanted to use dm-control with stable-baselines3? Within Reinforcement learning (RL), a number of APIs are used to implement environments, with limited ability to convert between them. This makes training agents across different APIs highly difficult, and has resulted in a fractured ecosystem.

Changelog: https://github.com/DLR-RM/stable-baselines3/releases/tag/v1.8.0

Great project! One question though, is there any reason why you are not using existing RL models instead of creating your own, such as stable baselines?

im trying to follow this tutorial = https://github.com/tensorflow/agents/blob/master/docs/tutorials/9_c51_tutorial.ipynb

I followed the TensorFlow tutorial for agents and the multi armed bandit tutorial and now I'm trying to make one of the already implemented agents, from the examples, work on my own environment. Basically my environment exists of 5 actions and 5 observations. Applying one action i results in the same state i. One action contains another step of sending that action number to a different program via a socket and the answer from the program is interpreted for the reward. My environment seems to be working, I used the little test script below to test the observe and action functions. I know this is not a full proof but showed its atleast working.

Depending on what task you pick the "best" algo will vary. There are also a bunch of variations and tricks for each of those, some of which have been given new names over time. If you are working on a project I would suggest whichever one has the simplest and most extendable implementation. If you really want to compare all of them you can use libraries that have them all implemented, such as tfagents.

I'm using a simple training loop, based on this.

Hello everyone, I published my python project a month ago, it's a command line interface for training, tuning and reusing reinforcement learning algorithms in tensorflow 2.x. It's similar to stable-baselines, tf-agents, and not so many others. It seems like it's not getting enough attention despite the README, license, and everything else.

Well it seems it doesn't flatten anything, just passes OrderedDict as input dense. Not sure but apparently it's keras that makes that a list of tensors. You can dig around places like https://github.com/tensorflow/agents/blob/v0.8.0/tf_agents/networks/network.py https://github.com/tensorflow/agents/blob/v0.8.0/tf_agents/agents/dqn/dqn_agent.py https://github.com/openai/gym/blob/master/gym/spaces/dict.py if you want to be really sure.

2- You could go the reinforcement learning approach by controlling these parameters using an agent. This would mean that the parameters would have to change on the fly, which I am not sure if appropriate. If so, creating a gym environment is not so hard, which would then use something like tf.agents , rlax or any other rl framework of your liking.

Not the same style and looks somehow more complicated but i want to mention tf.agents if you don't know about it already.