Pierre G. – Medium

Pinned

One month challenge: creating a self-learning game AI

Why this challenge? I was fascinated when in 2015, AlphaGo, a self-learning AI, beat for the first time a professional human player at the game of Go, or when, a few years later, an other AI by DeepMind reached top human player level at the complex real time strategy game Starcraft II. Having…

Self Improvement

2 min read

Jul 6, 2021

Day 30: Creating a self-learning AI

Read the story from the start A drop in motivation My plan last week was to make the challenge a round 5 weeks: spend one week on Connect 4, and then, regardless of the results, use the last week to give my best try at Photosynthesis (even if it fails). I ended up getting…

Self Improvement

3 min read

Jul 1, 2021

Day 25: Creating a self-learning AI

Read the story from the start Working on two fronts I started working on Connect 4 at the beginning of the week with two objectives in mind: Test how the training pipeline I built for Tic-Tac-Toe can be extended to train an AI for another game. Tackle speed performance issues. My training iterations are…

Self Improvement

5 min read

Jun 29, 2021

Day 23: Creating a self-learning AI

Read the story from the start Cloturing Tic-Tac-Toe On Thursday, I finally got pretty good results with my self-learned Tic-Tac-Toe AI, beating both my basic hand-written AI and my old Deep Q-Learning AI. I decided to spend one more day on it though, for two reasons: Despite getting much better results, there…

Self Improvement

5 min read

Jun 24, 2021

Day 19: Creating a self-learning AI

Read the story from the start We did it! I left yesterday thinking that the simplicity of our model may after-all be limiting our performance (I had assumed otherwise because I had gotten better results using the same model with my old AI). It turned out that it was definitely the case. After…

Self Improvement

1 min read

Jun 24, 2021

Day 18: Creating a self-learning AI

Read the story from the start This blog is officially online Wait a moment, wasn't that the case already? Well, yes and no: obviously it was already online, but I hadn't sent the link to anyone but a very few people, so the possibility of having other people read it was still mostly theoretical. Yesterday…

Self Improvement

4 min read

Jun 23, 2021

Day 17: Creating a self-learning AI

Read the story from the start Reevaluating my objectives This week was a rollercoaster! I had a few big breakthroughs, and also quite a few setbacks. …

Self Improvement

11 min read

Jun 16, 2021

Day 10: Creating a self-learning AI

Read the story from the start Taking a break from the blog As you may have noticed, I have taken a few days of break from writing here (contrary to my initial objective to write daily updates of my progress). Being new to writing, it turns out that it is taking me much more energy than…

Self Improvement

7 min read

Jun 10, 2021

Day 4: Creating a self-learning AI

Read the story from the start Debugging, debugging, debugging Today I started off by writing the code to generate player agents. I can now generate three kinds of player agents: A basic player agent that executes a very simple hand-written strategy (It checks whether or not it can get 3 in a row…

Self Improvement

2 min read

Jun 9, 2021

Day 3: Creating a self-learning AI

Read the story from the start No pressure As planned, today I tackled the task of implementing the MCTS algorithm, using this cheat sheet, this blog post, and the AlphaGo Zero paper as references. Something interesting happened: I wanted to finish the MCTS implementation today, and was hoping to play-test my old…

Self Improvement

1 min read