Intelecy News

Halite by Two Sigma. Kaggles multi-agent programming challenge

Written by Tom Van de Wiele | October 1, 2020 9:16:00 AM Z

Source: https://www.kaggle.com/c/halite/leaderboard?dialog=episodes-episode-3576456

At Intelecy, we are passionate about machine learning and we want to leverage the power of that technology to become the best in the world in solving predictive maintenance and process optimization challenges.

We also believe that in order to achieve our mission, we need to keep an open mind and measure our skills and technology against the best possible competitors.

In an earlier effort, we focused on predicting COVID-19 related deaths on a country-by-country basis. More recently, we decided to compete in a multi-agent programming challenge, organised by the most popular data science platform Kaggle.

Kaggle is an online community of data scientists and machine learning practitioners. It is most famous for hosting competitions to solve data science challenges. The competition we joined this time is special in the sense that it required participants to submit code rather than predictions.

The aim of the challenge was to collect as many resources as possible in a dynamic environment where there was ferocious competition with other players. Of the 1,139 participants, our submission came out as the overall winner (see graph below)!

Please refer to the official competition page if you want to learn more about this world-class challenge. On the top of the page you can see what a game looks like (our agent is the purple agent)

Our approach combined handcrafted rules with Bayesian optimization to balance the priority of different tasks. You can read more about this in an extensive technical discussion if you want to learn the details of our implementation.