mcoric96/blackjack-reinforcement-learning

Learning how to play Blackjack using tabular reinforcement learning algorithms - Monte Carlo first visit - Q learning

Blackjack-Reinforcement-learning

Artificial intelligence is no match for natural stupidity

― Albert Einstein

Game version

This is simplified version of Blackjack game, suitable for tabular RL methods without card counting and shuffle tracking methods
Allowed actions:

• hit
• stand

Game is represented as stationary environment:

• no multiple decks
• every hand is dealt from full deck
• cards are not put aside after one action

Algorithms

• First visit Monte carlo method
• Q learning method

Possible improvements

• add new algorithms
• add non-stationary environment and enable card counting

Resource

Reinforcement Learning: An Introduction

by Andrew Barto and Richard S. Sutton