Push Battle MCTS

Inspiration

We were inspired by the provided ideas at the Datathon as well as the relevant workshop for our challenge.

What it does

Our model is a simple implementation of the Monte Carlo Tree Search search technique to create a bot that can predict an optimal outcome in a non-deterministic game without having to pre-compute all possible game conditions.

How we built it

We built the model entirely in Python in such a way that it can interface with the Datathon's Flask server and provide a valid move within the time constraint.

Challenges we ran into

We found it difficult to figure out how we could create a model that would persist over multiple runs and as our group was new to the idea of the Monte Carlo Tree Search technique we had difficulty fully grasping the concept and theory behind it at first.

Accomplishments that we're proud of

We are proud of creating a model that can actually function and interface with the given API's correctly and is able to at the very least correctly predict moves when the game is in a terminal state.

What we learned

We learned a lot about the MCTS method and how it works at a very low level because we had to implement a completely new version of the method to function with our given constraints. We also learned a lot about python web servers and tree representations.