The computer plays its official move at the desired training strength (a problem for another day), then the human picks its move. Nearly immediately, the computer lists any better moves and the point difference compared to the human's move, this time computed with the computer's full strength (MultiPV?). The goal is for the human to learn patterns about what are good moves, to improve one's play. It's not necessary to always play the best move (that could be an "ugly" computer move), just to learn to play better ones: hopefully you see a better move that "makes sense".
Next, as additional training, the computer takes back your move and its official move, and we repeat the same exercise for all other possible moves the computer could have made. The goal is for the human to learn to take advantage of poor moves by the opponent.
After going through all the moves, the computer returns to its official move and your response, and the whole process repeats.
1 comment :
your first suggestion is how lots of people use gnubg to learn backgammon. i like the rest of your suggestion and wonder why it's not built into gnubg already.
Post a Comment