How do I use rllib for the examples provided?

[ryash072007]

SO, I found out that sample-factory is not supported on Windows OS. And rllib is the only backend that successfully installed on my pc. So, how can I use rllib to run the examples provided and make my own RL environments with it.

[edbeeching]

To use with rllib:

pip install godot-rl[rllib]

To train, with the BallChase example (for example):

gdrl --trainer=rllib --env_path=examples/godot_rl_BallChase/bin/BallChase.exe

rllib parameters can be modified by changing / copying the ppo_test.yaml file

Windows support and docs is a bit minimal at the moment, so any feedback and contributions towards this would be great.

[edbeeching]

Ah I just found a bug with the rllib wrapper, I just pushed a change. Please reinstall with:

pip install godot-rl[rllib]

[ryash072007]

OMG Thank you. I just managed to fix my issues with using sb3 and you replied. Thanks. Which one would be better btw? sb3 or rllib?

[edbeeching]

It depends on your problem, I think sb3 is simpler and easier to get started with, but rllib has more complex rl algorithms for more challenging environments. I would start with sb3, the default parameters should work well on most problems. Only a few example envs (Ball Chase and FlyBy) are support with sb3, whereas they should all work with rllib.

[ryash072007]

Where should the ppo_test.yaml be kept? in which folder because I get this error which says it can not locate it.

[ryash072007]

Also, is there anyway I can make the training process run infinitely until I stop it in sb3?

[ryash072007]

Also, is there anyway I can make the training process run infinitely until I stop it in sb3?

I figured how to this.

[ryash072007]

Where should the ppo_test.yaml be kept? in which folder because I get this error which says it can not locate it.

Can you help with this?

[edbeeching]

gdrl.interactive will run 1000 steps and you can load an example in the editor to see random actions being executed.

The ppo_test.yaml can be kept in the directory where you run your training. Alternatively, you can provide a path with --config_file

[ryash072007]

Error: No available node types can fulfill resource request {'CPU': 1.0, 'GPU': 1.0}. Add suitable node types to this cluster to resolve this issue. I keep getting this error and the training fails. Can you provide some information pertaining to this?

[edbeeching]

Ah, this is related to rllib. What are the specs of your machine, do you have an NVidia GPU? If not try setting the following in the yaml file:

num_gpus: 0

Training will be slower but from simple problems it should be fast enough.

[edbeeching]

FYI I updated the custom env doc a bit: https://github.com/edbeeching/godot_rl_agents/blob/main/docs/CUSTOM_ENV.md

I think it is still not so clear, would a video where I go through have to add an agent controller to a game be useful? I could try and look at that in the new year.

[ryash072007]

FYI I updated the custom env doc a bit: https://github.com/edbeeching/godot_rl_agents/blob/main/docs/CUSTOM_ENV.md

I think it is still not so clear, would a video where I go through have to add an agent controller to a game be useful? I could try and look at that in the new year.

yes please, that would be very helpful