Concha Labra, April 2014

This folder contains blackjack.py, a program that can be used as:

• dealer: for user to play against it interactively
• player: to learn optimal strategy for a given rules set

The game is played according to some basic rules. These rules, among others, can be changed in settings.py:

• There are one player and one dealer.
• The dealer hit until his hand value is 17 or greater.
• The player starts with 100 chips and bet 1 chip each hand.

python blackjack.py --help

usage: blackjack.py [-h] [-r role] [-p policy]

Bot to play blackjack. You can use the bot as dealer (to play against it for fun) or as player (to learn the best strategy for a given set of rule).

optional arguments: -h, --help show this help message and exit -r role, --role role bot role. Must be one of [player, dealer] -p policy, --policy policy strategy file. Default = policy.conf

python blackjack.py -r dealer

To play with the program interactively

python blackjack.py -r player -p policy.conf

To learn best strategy for given rules. Results are written to dir set in settings.learn_results dir

The program in his current state is the base for develop a machine learning system able to generate optimal policies to play blackjack for a set of given rules using techniques such as Markov Decision Processes

• README.txt - this file
• blackjack.py - main program
• game.py - the game loop
• player.py - opponents
• deck.py - cards, draw()
• policy.py - used in learn mode to read policy from file
• policy.conf - policy file
• settings.py - configuration: game rules, directory for learn results, etc.
• learn_results/ - directory for result files
• tests - tests

• PEP 8 - Style Guide is used for naming conventions
• Code tries to remove as many dependencies as possible, making some trade-offs as:
  • not use python's abc module to create abstract classes explicitly. Use it would require python version >= 2.6
  • not use numpy for certain data manipulation