[Python] How to set up a Conda environment for using Gym-based CityLearn (static buildings)...

I’m trying to build a Conda environment that supports CityLearn (v1.4.4) and Stable-Baselines3, both of which are based on the Gym API. My goal is to train reinforcement learning agents (e.g., PPO, SAC) from SB3 in a Gym-compatible static building setup in CityLearn.

However, I’m not sure about the exact package versions required to avoid compatibility issues, especially between:

citylearn

stable-baselines3

gym

numpy, torch, etc.

My Use Case: Use CityLearnEnv from CityLearn v1.4.4 (Gym-compatible) as the environment.

Apply model-free RL algorithms (like PPO or SAC) from Stable-Baselines3.

Focus on static building energy optimization (cost and battery control).

Use a Conda environment to manage dependencies cleanly.

My Questions: What is the correct way to set up a Conda environment for this use case?

What versions of gym, stable-baselines3, torch, etc., are compatible with citylearn==1.4.4?

Are there any known issues or tips for running Stable-Baselines3 in CityLearn (e.g., observation shape mismatch, multi-agent support)?

Do I need to apply any wrappers around the CityLearn environment to make it fully compatible with SB3’s training interface?

System Info: CityLearn version: 1.4.4

Platform: Conda environment (Windows/Linux)

Python version: (please specify yours)

Preferred SB3 algorithms: PPO, SAC, DDPG

Goal: Cost and battery optimization in static building environments using model-free RL.

Any help with the correct Conda environment setup or example environment.yml file would be greatly appreciated!

Continue reading...