Kell's chess predictor

An AI-based Chess Analytics tool where you can pair Players against each other and get game predictions based on an Analysis of older games.

Comment

Inspiration

Genuinely chess is the only sport I actively follow and since I have some understanding of how it works and what is important as well as it having a good amount of public data to train on I just picked chess. Furthermore chess is one of the more predictable sports there are.

What it does

Basically you just pair two players against each other and get predicted winning chances for each one of them. For this it uses a database of collected and preprocessed player data.

More specific Information is in the repo Readme.

How we built it

The backend is just Python, using Keras with Tensorflow for the prediction model. The UI is written in HTML and served with Flask.

First I planned and documented the Project using the readme of the github repo. Then I started by downloading lots of game and elo data and preprocessing and packaging them into player profiles with Python. Then writing and training the model before testing with new datasets.

Challenges we ran into

Im not used to Keras and I can be bad with time management. Also git and too big files.

Accomplishments that we're proud of

The Model. Working with such big datasets.

What we learned

A LOT. Learned a lot about basically every part of this project. Especially project management.

What's next for Kell's chess predictor

  • Scale Player Database
  • More & Better information on Players
  • Automatic Updating of player data
  • Include more Parameters for the NN
  • Make this mess more efficient. Like WAY more efficient

Built With

Share this project:

Updates