Skip to content

AI that solves the card game "Taki" using Q Learning

Notifications You must be signed in to change notification settings

dondish/taki-ai

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

12 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Taki AI

This is an experiment of teaching an AI to play Taki, a game which I own no rights to.

This project uses Keras and Tensorflow for Neural Networks.

The game does not use the 3+, 3+ breaker cards and the king card (new Taki cards).

This project taught me a lot about reinforcement learning and AI and I hope to continue making silly robots that learn!

Algorithm

Uses DQN with experience replay.

The heuristic is minus the amount of cards a player has when he finishes his turn, and the sum of the enemy cards on win.

This proved to improve the win rate by a substantial amount, also teaching the AI to use 2+ against enemy players.

Card Vector

Cards are a vector with dimension of the number of cards, with one at the respective index.

A deck is a vector sum of card vectors.

Action Space

The action space takes the card scalar if it's the play card action, or draw card scalar, or close taki scalar for other instances.

State Space

A concatenation of the following vectors / scalars:

  • The hand of the player
  • The discard pile
  • The state
  • The amount of 2+ stacked
  • The card shown on the table

Using the project

train.py trains the AI.

main.py Let's you play against opponents. You can customize the agents which you play against.

Random agents are pretty easy to beat since they are likely to draw cards when they are low on cards.

AI Agents need to be given a model to use or else they will use the random policy.

game.py defines all the classes needed to run the game.

agents a folder that includes playable agents.

agents/dqn.py the DQN agent.

agents/human.py the human agent.

agents/random.py an agent with a random policy.

models a folder that includes precomputed models.

License

Copyright (c) 2020 Oded "Dondish" Shapira

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

About

AI that solves the card game "Taki" using Q Learning

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages