search

amidos2006 / gym-pcgrl

A package for "Procedural Content Generation via Reinforcement Learning" OpenAI Gym interface.
MIT License
113 stars 27 forks source link

Making the project work in 2022 #14

Open visuallization opened 1 year ago

visuallization commented 1 year ago

Hey there,

Really cool & interesting project. I am currently thinking about writing my master thesis about pcgrl.

Anyways, I also wanted to mention that I had some issues in making the project run (especially the jupyter notebook file: inference.pynb) in 2022 as a lot of required packages have changed over the last years and are not compatible anymore with the project.

So I added/updated the setup.py with this:

install_requires=['gym==0.21.0', 'numpy>=1.17', 'pillow', 'tensorflow==1.15', 'pyglet', 'pyparsing==2.4.7', 'protobuf==3.20.2', 'stable_baselines==2.10.0', 'IPython'],
py_modules =[],

And switched to python 3.7.0 to make the project work again.

Should I create a PR with the changes @amidos2006 ?

Cheers!

amidos2006 commented 1 year ago

The setup.py is supposed to be for the actual framework and not for the trained networks nor training them (not for inference.pynb). It would be amazing if I separate the pcgrl repo from the trained networks or have a different folder for training and inference with requirements.txt. Feel free to do that without changing the setup.py as I want to make sure setup.py is not dependant on any specific machine learning method and can be used easily with anything. The only change is if the gym environment needs 0.21.0 specifically and can't work with anything newer. I am not sure it needs that as I didn't use anything special from the gym framework and I think it can work with the latest :)

Also, there is some stuff that you added that I don't think my system uses in training or inference such as protobuf and pyparsing.