Blackjack
MacOS/Linux Command Line Interface written in Swift for Blackjack Monte Carlo Simulations and Testing.
How to run
Clone the repo and double click on the unix executable under Product/Blackjack to run the CLI on Linux and Mac.
I recommend downloading Xcode 11+ and Swift 5+ to browse and run the code.
Design
Rules
This game makes several simplifications for now. It assumes a continuous shuffle machine rather than a fixed set of decks. Dealer hits on a hard 17. No splitting, no doubling, no surrendering. Only original bets are lost on dealer blackjack, although the dealer wins immediately.
Features
Players start out with 1000 coins. Every hand dealt, they can receive a hint, stand, or hit. The hint is currently implemented as a expectation maximization hand lookup table, where every possible hand is assigned an optimal move (either hit or stand in this case). The player hits until they stand or bust, the dealer draws until they reach 17 or bust, and then the winner takes the prize. The game continues until the player is out of coins or quits the game voluntarily.
Code Design
main.swift is the entry point of the application. It creates a new Game object to play().
A Game contains the state machine for the blackjack game, such as coins remaining,
and the play() method runs an indefinite loop until the player exits the game or loses.
The Game reads user input using the Input enumeration and its static method getCurrentLine(), which reads from stdin.
Both the .player and the .dealer receive a Hand.
A Hand contains initially two Card's, and can hit() to receive more or stand() to remain at the current sum.
The sum returns either the soft value or the hard value, depending on which is greater and not over 21.
The soft value is determined by calling reduce(initialResult:,nextPartialResult:) on the soft value of every Card that makes up the Hand, which is 1 for Ace and face value for everything else.
The hard value is determined by reducing the hard values of each card, which is 11 for ace and face value for everything else.
A Card is an enumeration of either .ace or .notAce with associated values containing the name of the card, as well as the 1) value of the card for .notAce, or 2) softValue and hardValue for .ace.
The Card .init is noteworthy - it simply calls Swift's built-in Array method randomElement(), which in turn calls the Collection protocol's randomElement(using:).
The random card will be chosen from a uniform distribution - in this case, every card has an equal chance of being chosen since every card appears four times in a deck.
See https://github.com/apple/swift-evolution/blob/master/proposals/0202-random-unification.md for the Swift Evolution proposal.
See https://github.com/apple/swift/blob/master/stdlib/public/core/Collection.swift for randomElement's implementation with the default generator.
If the dealer's hand isBlackjack, the dealer automatically wins the hand.
Else, the player can hit() until they bust or stand(), with the ability to receive hints every draw.
Receiving a Hint involves calling bestMove() given both the dealer and player hands,
and returning the move with the highest expected value using a priori blackjack knowledge.
For example, if the dealer's hand currently sums to less than 9, and the player has an ace and a seven, the best move is to Move.hit.
Now that the player's hand is finalized, the dealer draws until they reach 17 or bust.
The Game instance's gameOutcome() is then decided, printed to stdout, and the coins are exchanged in the case of no tie.
The loop starts back from the top, with each Player dealt a hand.
Todo
The blackjack game needs to be automated for Monte Carlo simulations. This goes hand in hand with increasing game rule options. For example, the user could run 10000 iterations on a game with 2 decks vs 4 decks and measure the difference in outcomes. Test cases will need to be added along with automation to check for edge cases like having multiple aces. The hints should be customizable by strategy, and perhaps even machine-learned via thousands of iterations with various parameters. Then, this simulation will be the perfect tool for any gambler wanting to brush up on their strategy before heading to the casino.