Žolík: search-based thinking bots

old neural-net trainer runs

A search-based bot looks ahead: at a draw or a discard it deals the cards it cannot see again among the stock and the other hands, tries each candidate move on a copy of the round, plays the round out, and keeps the move with the best average outcome (the number after search: is how many such guesses it makes per choice). Each run plays the same fresh deals twice, once in each seat, against the base algorithm (the game's own medium bot) or against hard. Player and opponent are average place scores (+1 winner … -1 last); difference = player minus opponent, ± is its 95% margin. A difference is shown coloured only on a finished run where it is larger than its margin: below that it cannot be told from chance. Green means the search bot is ahead, red behind.