mjai.app is a platform for mahjong AI competition. This repository contains the implementation of mahjong game simulator for evaluating submission files on mjai.app.
mjai
package is available at the Python Package Index (PyPI).
You can install it with pip
command.
$ pip install mjai
You can simulate a mahjong game by specifying submission files as shown in the code below.
The Simulator runs Docker internally. The docker
command must be installed and be available to run with user privileges.
import mjai
submissions = [
"examples/shanten.zip",
"examples/tsumogiri.zip",
"examples/tsumogiri.zip",
"examples/invalidbot2.zip",
]
mjai.Simulator(submissions, logs_dir="./logs").run()
This package also provides a base class for developing bots that communicate via the mjai protocol.
from mjai import Bot
class RulebaseBot(Bot):
def think(self) -> str:
if self.can_tsumo_agari:
return self.action_tsumo_agari()
elif self.can_ron_agari:
return self.action_ron_agari()
elif self.can_riichi:
return self.action_riichi()
...
if __name__ == "__main__":
RulebaseBot(player_id=int(sys.argv[1])).start()
Submission files are deployed in a Docker container and run as a Docker container.
This repository contains Dockerfile and other resources that will be used to create the Docker container.
The docker image is pushed to Docker Hub (docker.io/smly/mjai-client:v3
).
Please prepare a program that outputs the appropriate mjai protocol message to the standard output when given input in the mjai protocol format from standard input. Name this program "bot.py" and pack it into a zip file. The zip file should contain bot.py directly under the root directory.
bot.py must be a Python script, but it is also possible to include precompiled libraries if they are executable. The program will be executed in a linux/amd64
environment. The submission file must be 1000 MB or less.
bot.py takes a player ID as its first argument. Player ID must be 0, 1, 2, or 3. Player ID 0 represents the chicha 起家, and subsequent numbers represent the shimocha 下家, toimen 対面 or kamicha 上家 of the chicha 起家. See example code for details.
When the mjai.Simulator
instance creates an environment to run the submission file in docker, it specifies the --platform linux/amd64
option. If you want to run on a different architecture, you will have to emulate and run the container, which will be much slower. If you are debugging on an architecture other than linux/amd64
, you can avoid timeout errors by relaxing the timeout limit. Specify the timeout
argument as follows. The timeout
is set to 2.0 by default.
Simulator(submissions, logs_dir="./logs", timeout=10.0).run()
If the simulation is interrupted, the docker container may not terminate successfully. If past docker containers remain, the new docker container may fail to launch due to duplicate HTTP ports. You can remove all docker containers that have the smly/mjai-client
image as an ancestor in a batch as follows:
% for cid in `docker ps -a --filter ancestor=smly/mjai-client:v3 --format "{{.ID}}"`; do docker rm -f $cid; done
Our Mjai protocol is basically based on the Gimite's original Mjai Protocol, but customized based on Mortal's Mjai Engine implementation. The following points are customized:
- Messages are sent and received by standard input and standard output.
- The game server sends messages collectively up to the event that the player can act on.
- When the player does not send the appropriate event message, it is treated as a chombo and a mangan-sized penalty is paid.
- If the player does not send a message within 2 seconds, it is treated as a chombo and a mangan-sized penalty is paid.
- First round (kyoku) is not necessarily East 1.
If an error occurs in any one of the 4 players, the game is treated as a abortive draw (ryukyoku; 流局). At this time, all 4 players' containers are terminated. The players' containers are restarted and the game is resumed from the next round in which an error occurs. Therefore, the first round received by the player may not be East 1.
The following are messages sent and received as seen from player 0.
<-
denotes events for the player.
->
denotes events from the player.
# Game resumed from S1-1
0 <- [{"type":"start_game","names":["0","1","2","3"],"id":0}]
0 -> {"type":"none"}
# NOTE: No ryukyoku events are sent from the game server.
0 <- [{"type":"end_kyoku"}]
0 -> {"type":"none"}
# S2-2 first tsumo tile
0 <- [{"type":"start_kyoku","bakaze":"S","dora_marker":"1p","kyoku":2,"honba":2,"kyotaku":0,"oya":1,"scores":[800,61100,11300,26800],"tehais":[["4p","4s","P","3p","1p","5s","2m","F","1m","7s","9m","6m","9s"],["?","?","?","?","?","?","?","?","?","?","?","?","?"],["?","?","?","?","?","?","?","?","?","?","?","?","?"],["?","?","?","?","?","?","?","?","?","?","?","?","?"]]},{"type":"tsumo","actor":1,"pai":"?"},{"type":"dahai","actor":1,"pai":"F","tsumogiri":false},{"type":"tsumo","actor":2,"pai":"?"},{"type":"dahai","actor":2,"pai":"3m","tsumogiri":true},{"type":"tsumo","actor":3,"pai":"?"},{"type":"dahai","actor":3,"pai":"1m","tsumogiri":true},{"type":"tsumo","actor":0,"pai":"3s"}]
0 -> {"type":"dahai","pai":"3s","actor":0,"tsumogiri":true}
When the player has already made a tenpai but the hand is furiten. Since the player cannot Ron, even if the waiting tile is discarded by the opponent, no action is possible.
For example, let's say you have 2333678m 678s 678p
. The waiting tile is 14m
and 1m
has already been discarded.
Since the hand is a Furiten, even if the other player discards 1m
, the player cannot Ron.
3 <- [{"type":"dahai","actor":3,"pai":"P","tsumogiri":true},{"type":"tsumo","actor":0,"pai":"?"},{"type":"dahai","actor":0,"pai":"2p","tsumogiri":true},{"type":"tsumo","actor":1,"pai":"?"},{"type":"dahai","actor":1,"pai":"4m","tsumogiri":true},{"type":"tsumo","actor":2,"pai":"?"},{"type":"dahai","actor":2,"pai":"6m","tsumogiri":true}]
In this case, immediately after actor 2 discards 6m, input is given to actor 3. actor 3 needs to decide whether to call chi
on 6m.
In the case of an ankan, the dora event comes first, followed by the tsumo event.
3 -> {"type": "ankan", "actor": 3, "consumed": ["6s", "6s", "6s", "6s"]}
3 <- [{"type":"ankan","actor":3,"consumed":["6s","6s","6s","6s"]},{"type":"dora","dora_marker":"6p"},{"type":"tsumo","actor":3,"pai":"7p"}]
3 -> {"type":dahai","actor":3,"pai":"7p","tsumogiri":true}
The procedures executed by Simulator can be checked and debugged one by one as follows:
# pull latest docker image
% docker pull docker.io/smly/mjai-client:v3
# launch
% CONTAINER_ID=`docker run -d --rm -p 28080:3000 --mount "type=bind,src=/Users/smly/gitws/mjai.app/examples/rulebase.zip,dst=/bot.zip,readonly" smly/mjai-client:v3 sleep infinity`
# install bot program
% docker exec ${CONTAINER_ID} unzip -q /bot.zip
# debug
% docker exec -it ${CONTAINER_ID} /workspace/.pyenv/shims/python -u bot.py 0
[{"type":"start_game","id":0}] <-- Input
{"type":"none"} <-- Output
[{"type":"start_kyoku","bakaze":"E","dora_marker":"2s","kyoku":1,"honba":0,"kyotaku":0,"oya":0,"scores":[25000,25000,25000,25000],"tehais":[["E","6p","9m","8m","C","2s","7m","S","6m","1m","S","3s","8m"],["?","?","?","?","?","?","?","?","?","?","?","?","?"],["?","?","?","?","?","?","?","?","?","?","?","?","?"],["?","?","?","?","?","?","?","?","?","?","?","?","?"]]},{"type":"tsumo","actor":0,"pai":"1m"}] <-- Input
{"type": "dahai", "actor": 0, "pai": "C", "tsumogiri": false} <-- Output
[{"type":"dahai","actor":0,"pai":"C","tsumogiri":false},{"type":"tsumo","actor":1,"pai":"?"},{"type":"dahai","actor":1,"pai":"3m","tsumogiri":false},{"type":"tsumo","actor":2,"pai":"?"},{"type":"dahai","actor":2,"pai":"1m","tsumogiri":false}] <-- Input
{"type": "none"} <-- Output
[{"type":"tsumo","actor":3,"pai":"?"},{"type":"dahai","actor":3,"pai":"1m","tsumogiri":false}] <-- Input
{"type": "none"} <-- Output
[{"type":"tsumo","actor":0,"pai":"C"}] <-- Input
{"type": "dahai", "actor": 0, "pai": "C", "tsumogiri": true} <-- Output
# install http server interface of mjai protocol
% docker cp python/mjai/http_server/server.py ${CONTAINER_ID}:/workspace/00__server__.py
# http server mode. `0` is the player index.
% docker exec -it ${CONTAINER_ID} /workspace/.pyenv/shims/python 00__server__.py 0
% curl http://localhost:28080/
OK
% curl -X POST -d '[{"type":"start_game","id":0}]' http://localhost:28080/
{"type":"none"}
% cat > request.json
[{"type":"start_kyoku","bakaze":"E","dora_marker":"2s","kyoku":1,"honba":0,"kyotaku":0,"oya":0,"scores":[25000,25000,25000,25000],"tehais":[["E","6p","9m","8m","C","2s","7m","S","6m","1m","S","3s","8m"],["?","?","?","?","?","?","?","?","?","?","?","?","?"],["?","?","?","?","?","?","?","?","?","?","?","?","?"],["?","?","?","?","?","?","?","?","?","?","?","?","?"]]},{"type":"tsumo","actor":0,"pai":"1m"}]
% curl -X POST -d '@request.json' http://localhost:28080/
{"type":"dahai","actor":0,"pai":"6p","tsumogiri":false}
Confirmed working with rustc 1.75.0 (82e1608df 2023-12-21).
$ pip install maturin # install build tool
$ maturin build --release --locked --target aarch64-apple-darwin --out dist
The code in ./src
directory is Mortal's libriichi with minor updates. Mortal is distributed under the AGPL-3.0 and is copyrighted by Equim.