2.2 KiB
Simulations of Hanabi strategies
Hanabi is a cooperative card game of incomplete information. Despite relatively simple rules, the space of Hanabi strategies is quite interesting.
This repository provides a framework for implementing Hanabi strategies. It also explores some implementations, based on ideas from this paper.
In particular, it contains a variant of their "information strategy", with some improvements. This strategy achieves the best results I am aware of for n > 2 (see below).
Please contact me if:
- You know of other interesting/good strategy ideas!
- Have questions about the framework or existing strategies
Some similar projects I am aware of:
- https://github.com/rjtobin/HanSim (written for the paper mentioned above)
- https://github.com/Quuxplusone/Hanabi
Setup
Install rust/rustc and cargo, and change the options in main.rs appropriately.
cargo run -- -h
Usage: target/debug/rust_hanabi [options]
Options:
-l, --loglevel LOGLEVEL
Log level, one of 'trace', 'debug', 'info', 'warn',
and 'error'
-n, --ntrials NTRIALS
Number of games to simulate (default 1)
-t, --nthreads NTHREADS
Number of threads to use for simulation (default 1)
-s, --seed SEED Seed for PRNG (default random)
-p, --nplayers NPLAYERS
Number of players
-g, --strategy STRATEGY
Which strategy to use. One of 'random', 'cheat', and
'info'
-h, --help Print this help menu
For example,
cargo run -- -n 10000 -s 0 -t 2 -p 5 -g cheat
Or, if the simulation is slow (as the info strategy is),
cargo build --release
time ./target/release/rust_hanabi -n 10000 -s 0 -t 2 -p 5 -g info
Results
Currently, on seeds 0-9999, we have:
| 2p | 3p | 4p | 5p |
----------|---------|---------|---------|---------| cheating | 24.8600 | 24.9781 | 24.9715 | 24.9583 | info | 18.5959 | 23.8846 | 24.7753 | 24.8719 |