Those in the DeepRacer league must write a reward function, which affects the performance of their model, and the time efficiency of the training process.
Therein lies the fun of DeepRacer, old sport!
Okay — so I took a bunch of pictures of my model training. In retrospect, I definitely should have just taken a screenshot, but I really just took those pictures to send to my mom. As of right now, I am unable to access my AWS account (no, I don’t want to talk about it), so this is the best you’re going to get for now. Until the issue is resolved and I’m able to open up my AWS console again, screenshots are in order!
Also, pro-tip: I wouldn’t do anything machine learning related on a MacBook if you can help it. I couldn’t, so I brought my MacBook to the event. You’re better off with a machine with a lot of RAM. From Gokkulnath T S’ article on personal deep learning machines,
“If you are a researcher/Student/Hobbyist Consider for a Dual GPU Build. If you plan to run huge models and participate in insane contests like ImageNet which require heavy computation consider for a Multi GPU Build.”
As you can see between the first and second photo above, reinforcement learning is not linear in its learning and growth. The model is going to make mistakes, but, much like humans, it learns from its mistakes.
Also, on the right-hand side you can see a track simulation. I was able to see how my model performed.
Within each episode iteration, my model was learning through the reward function I set up.
For those of you who follow my GitHub, Twitter, Instagram, or pretty much anything else, you know my deal. I’m pretty much always @mackied0g. On this track (the smallest, just for efficiency purposes), I ended up in 928th place out of 1817 people.
I never dreamed about being excited to be in 928th place, but this is pretty thrilling for me, considering it was my first stab at reinforcement learning.
While I hate to answer a question with a question (not really, I love it), where can’t this technology be applied? We can use reinforcement learning to perform surgeries with artificial intelligence, to provide business forecasts, to help us out with our fantasy football lineup, to so much more.