AI Learning How to Play Pong in RealTime – PyTorch

floydlabs.com

63 points·diegoalejogm··12 comments

PyTorch code: https://github.com/diegoalejogm/deep-q-learning

I'll add final results and performance videos soon!

Neat! Of note: if I'm reading correctly, this is learning to play the video game pong with no prior knowledge using the screen pixels alone. Pretty cool.

Yeah! I couldn't have described it better than you just did!

Feel free to take a look at the code and star the code if you like it. I'll keep upgrading it and adding more Reinforcement Learning variations

Forgot to mention that it is learning from raw pixels!! No other info is passed to the agent, except the screen frames :)

Also, you can check the FloydExecution in: https://www.floydhub.com/diegoalejogm/projects/atari/29

Also, it will be training for several hours more (I approximate 20), so make sure to keep updated. I'll post videos on the current execution performance.

The help link doesn't work for me, did the github repo change?

No, it hasn't. Here: http://github.com/diegoalejogm/deep-q-learning

Try again and let me know if this time it works :)

Ah, I'm new to tensorboard. The help link in tensorboard was broken. But that link works :) very cool!

No worries! Here's also a link to the current progress https://www.floydhub.com/diegoalejogm/projects/atari/29 in case you wanna check the machine usage too

Just so you know, machine stats feature is not ramped to all the users yet, so most of them won't be able to see the fancy UI you were seeing. We will push it out to all users late next week ;)

Good to know :) Thanks for the info, I'll have that in mind for the following posts!

Reply on news.ycombinator.com