Black Box Challenge Machine Learning Competition

Take part in an unusual machine learning competition — program an agent (in Python) that can play a game with unknown rules.

By Georgy Cheremovskiy, PhysTech.


We’d like to invite you to participate in an unusual machine learning competition — Black Box Challenge (

The conception is simple — one need to program an agent that can play a game with unknown rules.

At each time step agent is given an environment state vector and has a few possible actions. The rewards may be delayed and have stochastic nature — the same actions can lead to different rewards.

The competition is created with support of (one of the largest Russian Internet companies) and Data-Centric Alliance (DCA is a Russian company specializing on big data and high-load systems), and the winners will be rewarded with pleasant prizes:

$4400 (300,000 rubles) — for the 1st place
$2550 (175,000 rubles) — for the 2nd place
$1820 (125,000 rubles) — for the 3rd place
Microsoft Xbox One — for the 4th, 5th, 6th, 7th and 8th places

You can read about the problem in detail at the site.(

It’s easy to create your first bot — you need to have basic knowledge of Python (if you’re into R, Python isn’t hard to master) Sign up at and submit your solutions!