cleanrl VS spinningup

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.

Here's where the problem starts: after implementing a custom environment that follows the typical gymnasium interface, and use a slightly adjusted PPO implementation from CleanRL, I cannot get the agent to learn anything at all, even though this specific implementation seems to work just fine on basic gymnasium examples. I am hoping the RL community here can help me with some useful pointers.

Try out a standard implementation with some standard parameters from here: https://github.com/vwxyzjn/cleanrl/tree/master/cleanrl

I am trying to run cleanrl on the `Pendulum-v1` environment. I did that by going here and changing the default `env-id` to ` parser.add_argument("--env-id", type=str, default="Pendulum-v1",

git clone https://github.com/vwxyzjn/cleanrl.git && cd cleanrl poetry install

Can someone please help me debug my implementation of SAC. Please let me know if you have any questions. I tried comparing my work with CleanRL and caught a couple of errors. However, my implementation does diverge a lot from theirs as I wanted to test my understanding.

Shameless self-plug: as far as implementation is concerned, I am working on a (hopefully) easier to understand Dreamer architecture under the CleanRL library, toward also re-implementing Director, Dreamer-v3, and and JAX variant for faster training.

Happy to share that CleanRL now has a new algorithm called Robust Policy Optimization — 5 lines of code change to PPO to get better performance in 57 out of 61 continuous action envs 🚀 (e.g., dm_control)

I am trying to follow a vanilla implementation of REINFORCE algorithm found in here: https://github.com/openai/spinningup/blob/master/spinup/examples/pytorch/pg_math/1_simple_pg.py

I am looking at OpenAI's implementation of SAC over here. Also, here is their code to compute the action and its log prob -

In OpenAI's spinningup's VPG-GAE implementation , the authors update the value function V(s_t) multiple times at every epoch using the same batch of observations. Copying their code (line 237 onwards in link above):

https://spinningup.openai.com/ is a good resource to start understanding the algorithms hands on by looking at their implementations.

so PPO can be used to produce deterministic outputs, as you mentioned the two values of the actor network can be those two deterministic values, no need to model a mean and variance, but if you want to get a normal distribution, spinning up intro RL check the section where they explain policies, and later the code where the implementation is, is in pytorch but they have a tensorflow as well, you can see there is comething called gaussian actor, in how they computed the log_prob and then get the mean and var to generate a normal.

Take a look at the SpinningUp repo