Management refactor - Githubissues

RedTachyon / cpr_reputation

0 stars 0 forks source link

Management refactor #51

Closed bengreenberg5 closed 3 years ago

bengreenberg5 commented 3 years ago

change the config standard from INI to YAML
add new hyperpameters (see open_map.yaml)
specify run name, config path, and # iterations in CLI args

quinn-dougherty commented 3 years ago

must save checkpoints
then fix record_video.py

quinn-dougherty commented 3 years ago

TODO

add environment.render method (it's much like board.render)
wandb integration (with gym.Monitor)
- include config in wandb save.

RedTachyon commented 3 years ago

hyperparameter_defaults = dict(
    gamma=0.99,
    eta=0.0,
    seed=0
)

wandb.init(project="<project-name>",
           name="<run-name>",
           entity="marl-cpr",
           sync_tensorboard=True,
           monitor_gym=True,
           config=hyperparameter_defaults)

config = wandb.config

bengreenberg5 commented 3 years ago

python learn.py --name single-agent-test --config configs/single_agent.yaml --iters 500 --checkpoint_freq 10

python record_video_ray.py --config configs/single_agent.yaml --checkpoint_path ~/ray_results/PPO/.../checkpoint-500/checkpoint-500

RedTachyon commented 3 years ago

Does the video recording component work? In that case I think this can be merged, if nothing else then to make it a bit more tidy w.r.t. branches