Blackjack
Play the dealer alongside three reinforcement-learning agents.
Funds$1000
Bet$0
Dealer
–0cards left
AI agents
Q-Learning– wins
Sarsa– wins
TD Learning– wins
You
–– winsHow the agents learn
Q-Learning
Scores each move by its reward plus the best move that could follow. Tries new moves, then sticks with what works.
Sarsa
Learns from the move it actually takes next, which makes it more cautious than Q-Learning.
TD Learning
Updates its estimates after every step, from the gap between the reward it expected and the one it got.
Win rates count every visitor's hands. Ties don't count.