I’m exploring following along an RL tutorial but my goal is to implement it all in swift.
Here is a swift notebook that implements Q-learning using an open-ai’s Gym.
Interestingly I found that using S4TF for this simple problem radically slowed down my approach. Here is the same algorithm but implemented using a S4TF tensor as the table store.
Hopefully you might find it interesting to compare the python implementation to the swift one.
Week 3 involves solving Pong, so I’ll need to use some swift fastai to do that