pandemic-ai

The project aims to create self learning artificial intelligence capable to beat the board game Pandemic. As i know nothing about neural network and deep learning, I will start from scratch : no library will be used to manage neural network.

Future releases

• proper logging
• comment code
• provide use instruction
• update README with mardown syntax
• check MCTS algorithm implementation
• write JMH benchmark for critical methods
• optimize deck creation/duplication
• optimize gameState duplication
• junit tests for Pandemic.getMoves method
• provide a more optimized Pandemic.getMoves method

Changelog 2019.06.28

• Junit tests cover all pandemic package except material. JUnit tests to be added before any refactoring of the material package
• Deck, PlayerDeck, PropagationDeck refactored
• Benchmark deck critical operations (draw, duplicate)
• add equivalent method to test logical equality between objects

2019.06.20

• Deprecated observer pattern migrated to PropertyChangeListener
• JUnit tests in progress

2019.06.18

• Refactored project architecture
• JUnit tests in progress
• Add maven build process
• Improved algorithm to find all valuable combinations of 4 actions
• JMH benchmark

The steps I Building a decent Pandemic Board game

1. Initializing the game (create architecture classes, useful methods) is its simplest way : no event card, no card limit, no character player
2. Add log4j to track
3. build the game mechanics : what the game do , effect of each action etc.
4. Create the game engine : waiting for input to perform player actions and automatically performing game actions
5. Add eventhandling to throw event when : player hand is full or when it must make an action -> DONE
6. Build an interface to play pandemic and check everything runs smoothly

II Build a Monte Carlo Tree Search based AI and check if it works

1. Undertsanding MCTS algorithm https://int8.io/monte-carlo-tree-search-beginners-guide/
2. Implementing the algorithm
3. Testing & debugging
4. Optimization

III Build a neural network learning from the Monte Carlo Tree Search, Alphagozero's way

IV Standardize AI entry points and API to make it "game" independant

V Add events and characters role