LeonaRaging/connectx — reverse-engineered prompt

Build me a Python project for a strong Kaggle ConnectX player that uses different strategies depending on the stage of the game. I want it to use an opening book for the first several moves, then switch to an AlphaZero style mid game agent with Monte Carlo tree search, and then use a fast alpha beta minimax solver for the end game. The goal is a competitive agent that plays Connect Four really well and can be used as a Kaggle submission.

Please make it easy to run locally so I can play against the AI from the terminal, and include simple test scripts for one game and many games so I can see how it performs. I also want benchmark scripts to compare move generation and scoring speed, plus a way to generate the opening book. Keep the implementation focused on performance, including efficient board handling and caching of repeated positions where it helps.

If anything about the Kaggle environment is unclear, look up the current docs online and wire it up so it works cleanly.