1
0 Comments

Game Engine Adventures: AI Playing 2048 Part III [Week 8]

The training for 2048 agent is done. The agent played around 215,000 rounds of 2048. The final version can win 4% (+/-3%) of rounds. I was hoping it would do much better. You can see a winning game in the first row, third column in video below.

I've noticed that the agent sometimes chooses to combine any tiles it can as oppose to work toward 2048. This is probably because there is a reward for increasing the overall score of the game. I will experiment with simplifying the reward function to just provide a reward each time a new value tile is formed. The first time the agent gets a tile of 16, it will get 16 reward, etc. Let's see the results next week.

Trending on Indie Hackers
How I grew a side project to 100k Unique Visitors in 7 days with 0 audience 49 comments Competing with Product Hunt: a month later 33 comments Why do you hate marketing? 28 comments My Top 20 Free Tools That I Use Everyday as an Indie Hacker 15 comments $15k revenues in <4 months as a solopreneur 14 comments Use Your Product 13 comments