A simple Rock-Paper-Scissors game using an AI N-gram for pattern detection.
This game uses a simple AI to try and predict your next move. The AI stores multiple N-grams - it keeps count of various possible sequences of turns (including both yours and the AI's own play) and what you played next.
The AI keeps several such counts for different lengths of sequences (e.g. pairs, triplets and so on). The probabilities that they generate are then combined with linear weights to make a final prediction. After a play is made, the weights are updated based on the actual outcome, to maximize the cross-entropy of the predicted and real distributions.
This allows the AI to pick up on simple repeated patterns in your play. If played completely randomly, rock-paper-scissors should lead to a ~33% win rate for each player. After playing enough turns (about 40 usually are enough), check your score at the top of the page and see what happened. If your play is not truly random, the AI will have exploited that to win more games than you.
To clear the AI's memory and start anew, simply refresh the page.
This repository was created and developed using Bun. It can be initialized for development easily by running bun install.
Use bun run build-watch to live compile the Typescript source for development. It will be compiled in the static folder together with the HTML/CSS
files where they can be previewed together.
Use bun run build to build the final dist folder and bun run publish to upload it to the gh-pages branch.