We developed a game-playing AI for a novel board game called Milestone, developed by a Cornell Math Ph.D. student Mark Schachner. We've implemented this AI two-fold: as a tree-search-based, hyper-heuristic AI that is improved genetically; and as a NN trained on the resulting data.

To launch our Milestone AI, simply launch the rust process. We recommend cargo run --release (with rust installed..).

• Creating a board from scratch:
  • "b:b/bb/bbb/bbbb/3/4/3/4/3/wwww/www/ww/w"
    • b -> black, b -> white, x -> number of blank squares, / means next line, b: means black's turn (w: for white's turn)

• Passing any arguments to this program (ie cargo run genetic will cause the program to launch in genetic mode)
• In addition, the following env vars can be set:
  • LAUNCH_ARG: corresponds to the number input at the beginning
  • PER_NUM_BATCHES, TOTAL_NUM_BATCHES, NUM_AGENTS_RETAINED, NUM_CHILDREN_PER_RETAINED_AGENT, MAX_PERTURB_AMT, PERTURB_DECR, NUM_AGENTS, NUM_MATCHES, AGENT_DEPTH
  • PLAY_AFTER to toggle if a game is launched following the completion of the genetic process (currently the program only checks if this var exists, not its value)
  • DATABASE_URL for the database URL

This process can be launched via ./target/release/milestone (after compiling), which can be done periodically using a cron job. For example, we can relaunch this process every two hours, and the recovery database will ensure that process is maintained. Logs will persist in the logs/ directory and stderr can be piped to a file to debug any other issues with the process.

• Download VsCode Sqlite Extension (SQLite by alexcvzz)