Getting started

The arena scores Othello agents that have to work out the rules of an environment by playing it. An environment is a board (size and obstacles), a win condition (most or fewest discs wins) and a capture rule (how a move flips discs). Your agent is told only the board.

The four steps

1. Get the kit. Download it once your account is approved. You edit one file, my_agent.py; the agent API says what it receives and returns.
2. Check it locally. make check runs the same gate the server runs, and make play-local plays practice environments with their rules shown.
3. Upload it. On Submit: a single agent.py, or a .zip with agent.py at its root. Choose CPU or GPU. To install packages, put a requirements.txt at the root of the zip (make package does this). The screen answers at once and a smoke test follows; a passing upload becomes your active agent.
4. Get ranked. From your submission's page you can ask for a scored run, which plays your agent against three built-in agents, shows your own curve and ranks nobody. The leaderboard comes from rounds the professor runs over the whole field: on every environment every two agents meet fresh for a pairing of 200 games, both learning the rules as they play and both told every result, and your standing is your mean score over those pairings. Four built-in agents that know the rules play every entrant too; they appear as reference rows with no rank and never count in an entrant's standing.

Play an agent

The home page lets you play against an agent in the browser once you are signed in: one of the nine built-ins (Random, Greedy, Positional, Minimax 3, 5 and 10, MCTS 10, 30 and 100), or any student agent whose owner opened it for play (a switch on the submission page). Greedy and the Minimax agents play the board's win condition, and the MCTS agents already do; the slower ones think for seconds a move, and the home page says how long each takes on the standard board. The standard 8×8 board is the default and every practice board can be chosen. Your recent games are listed under the board.

Korean interface

The language link at the top of every page switches menus, buttons and short notes to Korean for this browser, and back to English. The documents stay in English.

Where to look next

• Competition rules: the environments, what is scored, the limits, and what is not allowed.
• Agent API and starter kit: the interface and the kit's commands.
• Benchmark specification: the benchmark and where it departs from the paper.
• Environments: every practice environment with its rules.
• Replays: watch games, including your own, a pairing at a time.