bombora
bombora copied to clipboard
My experimentations with Reinforcement Learning in Pytorch
Bombora
My ongoing experiments in RL, originally based on ikostrikov/pytorch-a3c.
Usage
OMP_NUM_THREADS=1 python main.py --env-name "Pong-v0" --algo a3c --num-processes 4
Note: Pytorch is still in beta and non recent version might have some problems.
Dashboard
Logs are stored in dblogs
directory, to watch agents learn run:
python dashboard.py --dbdir dblogs --env 'Pong-v0'
This will start a visdom server and show the last 2 run logs in Pong-v0 env.


Dependencies
- pytorch
- torchvision
- gym
- sqlite3
- visdom
Credits
Cheers to Pytorch and authors of the follwing repos: