suchmarines

Internal EPAM competition bot based on machine learning

Game

Turn based Space strategy.
Your goal is to capture all planets on the map. This could be done with drones.
Drones is your primary weapon. They could be used to capture neutral planets and attack enemies.
Each planet contains drones factories (planet regeneration level).

Idea

Based on I am a legend: Hacking Hearthstone with machine learning
Could we make a bot based on machine learning for a game, where is pretty difficult to analyze all possible combinations?

Timeline

08 aug - Dirty MVP created. Machine learning could play! Yey!
19 aug - bot passed semifinals. Heh. I don't know why. Was it a small DB or algo issue, but bot decided to defend during all semifinal games. And passed them =)
Here is a screenshot of a final game (2nd place):
23 aug - final. Bot still wanted to play a defend game. Heh. But it was so boring. We've implemented slightly different strategy. So, finally, our bot made first aggressive moves with hardcoded algorithm. And then switched to machine learning. We lost that game ;).

Summary

There was a guess. Can we use machine learning without any experience with it?
The answer is: "Yes, we can!". And that's great!

Authors