Dear lichess.org community,
my team and I created an AI called CrazyAra for the chess variant Crazyhouse as our semester university project.
The neural network is mainly based on the techniques described in Alpha-Go/Alpha-Zero paper, with the main exception that it was only trained supervised on a single GPU by now.
CrazyAra learned the game of Crazyhouse, by processing all games you've played from January 2016 to June 2018 (database.lichess.org/).
We've selected all games including Bullet-Games with a minimum elo >= 2000 for both players, resulting in 56,9537 training games total.
A paper about the implementation details will be made available soon.
You can challenge the bot in blitz-time control format in the Crazyhouse variant.
@CrazyAra
The bot will be online from September 7, 20:00 hrs to September 9, 11:30 hrs (Central European Time) for now.
We're excited to see how CrazyAra will perform against it's supervisors.
Greetings and good luck in your games,
Your CrazyAra-Team
my team and I created an AI called CrazyAra for the chess variant Crazyhouse as our semester university project.
The neural network is mainly based on the techniques described in Alpha-Go/Alpha-Zero paper, with the main exception that it was only trained supervised on a single GPU by now.
CrazyAra learned the game of Crazyhouse, by processing all games you've played from January 2016 to June 2018 (database.lichess.org/).
We've selected all games including Bullet-Games with a minimum elo >= 2000 for both players, resulting in 56,9537 training games total.
A paper about the implementation details will be made available soon.
You can challenge the bot in blitz-time control format in the Crazyhouse variant.
@CrazyAra
The bot will be online from September 7, 20:00 hrs to September 9, 11:30 hrs (Central European Time) for now.
We're excited to see how CrazyAra will perform against it's supervisors.
Greetings and good luck in your games,
Your CrazyAra-Team