Hello there, I was wondering if there is a way to accelerate this environment, because it takes way too much time to compute a single step (>x1000 compared to atari for example) … Please, when doing RL competition, provide a fast C++ environment.
Hi @joseph_amigo, thanks for your suggestion about using a faster language to write the environment.
The current developers of CityLearn are not computer science nor software programming experts as they are building energy and science experts hence, we agree that the environment can benefit from some optimization for faster runtime.
If this is an area either you or anyone else has experience in and wants to contribute to, we are happy to look into it.
I couldn’t find exactly the bottleneck in your code, but there is one. I think it’s in the use of pandas. I suspect several operations to happen on the all dataset every time instead of the current timestep.