Halite

Halite was an AI programming competition hosted by Two Sigma.
Players compete by programming a bot that controls ships to strategically beat the opponent.

Halite

Side Project
Oct. 2017 – Jan. 2018

• Beat 99% of the competition: Out of 5832 contestants, my bot was ranked #54.
• Implemented a complete collision-less path planning algorithm from scratch.
• Utilized NEAT, an Evolutionary Algorithm framework, to evolve a population of bots through self-play.

Let Them Fight

For this competition, I wrote a self-learning AI capable of beating over 99% of the other contestants.
Out of 5832 contestants, my bot was ranked #54.
Out of 153 contestants who used Machine Learning, my bot was ranked #2.

One thing to note is that my bot is deliberately not purely ML. It combines both hand-written code for path planning and ML for strategy. I strongly believe that the best result is at the intersection between machine intelligence and human intelligence.

Machine Learning

My main goal for this competition was not necessarily to win the competition, but to learn more about Machine Learning. As a result, I would say the project was a great success. I’ll always prefer doing project-based learning over following some online tutorial!

My bot code is publicly available, and it’s a good indicator of my usual coding style. I highly recommend reading my postmortem if you’re interested.