My ongoing experiments in RL, originally based on ikostrikov/pytorch-a3c.

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.

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.

video

• pytorch
• torchvision
• gym
• sqlite3
• visdom

Cheers to Pytorch and authors of the follwing repos:

• ikostrikov/pytorch-a3c
• steveKapturowski/tensorflow-rl