README.md

Hanabi project

This project is for educational purposes: its current status should represent the student's starting point.

Hanabi module

A python3 module that knows Hanabi:

• card deck and dealer,
• CLI (command line interface), lets you play and verifies that your moves are allowed (currently limited to 2 players),
• GUI (to do),
• AI (currently only one, and it is cheating).

Tasks

There are many possible tasks:

• read the current state of the module,

  • complete its documentation when needed,
  • share with everyone this improvement,

• improve the CheaterAI -- answer the question (still an open question afaik) what is the best possible score, statistically speaking.

• add AIs. Some suggestions:

  • RandomAI (plays randomly)
  • DirectAI (plays whatever is hinted)
  • BGAAI (plays Board Game Arena's standard)
  • HansimAI (see below)
  • train a machine learning (I'm not sure if this will give anything interesting without powerful CPU/GPU resources)
  • design your own, from scratch or by improving another

Keep track of scores for all these games/AI. We will want to compare: different AIs on a same deck, or a given AI over a 1000 decks. We will need to see why a certain AI fails on a certain game.

• make it workable for up to 5 players.

• make it workable from two separate screens (network?)

During the project:

• make sure you understand the "replay" mode
• add tests (UnitTest or whole tests)
• keep notes on your questions, decisions, discussions (github's wiki)

You may also design a GUI, but be warned that this is a very time-consuming task. I like PySide2. Tkinter is more portable but harder to learn imho.

Installation

git clone https://github.com/JDGaraudEnsta/hanabi
cd hanabi
make        # pip installs it in the default directory ~/.local
hanabi
# and now you may play

If hanabi doesn't start (bash: hanabi: command not found), add this to your ~/.bashrc:

export PATH=$HOME/.local/bin:$PATH

Bibliography

Other Hanabi projects

• A C++ bot: some strategies and success rates
• HanSim: the Hat guessing strategy
• HanSim: source code
• boardgame arena
• hanabi conventions (hanabi-live), and references therein.

AI

• deepmind: Atari
• deepmind: SC2
• deepmind: Hanabi
• todo: find non-deepmind references?

Misc (coding principles, project, ...)

• keep it simple
• rule of least surprise, catbaz
• rubber duck debugging
• markdown (overview), markdown (in details)
• BGA state machine