Connect 4i

Inspiration

We're really bad at Connect Four.

What it does

It plays Connect Four using eight randomly generated neural networks and applies a genetic algorithm by cross breeding the best players in each generation to produce the most "skilled" player.

How we built it

Connect 4i was built using solely python. The only dependencies used is the numpy library for mathematical computation. Each neural network corresponds to a flattened version of the state of the Connect Four board and seven outputs that correspond to the columns. The eight players in each generation were put into a tournament style competition where the top two winners in the bracket were crossbred and mutated to create the next generation of "improved" players.

Challenges we ran into

We didn't have the processing power needed to train the model enough to create an expert player within 24 hours. We were limited to using MacBook Airs and Amazon Web Services to host the training.

Accomplishments that we're proud of

It probably works. We're still training it as we are writing this.

What we learned

We learned how to create simple neural networks and apply genetic algorithms to optimize the neural networks.

What's next for Connect 4i

In the future, we can improve the genetic algorithm that would produce better bots as well as create more robust neural networks with more processing power.

Built With

• numpy
• python