--- In [email protected], "John Hart" <[EMAIL PROTECTED]> wrote: > > Seems like a good opportunity for Quackle to go beyond > just making the right play and actually TEACH the user > WHY certain plays were considered or discarded. > > Let's assume that the 5 tiles left in the bag cannot be > any combination of letters that produce 64-point QI or > ZA plays, or other high-scoring prospects. Then, the > 60-point bingo would almost certainly lose the game > (still trailing by 30, no way to catch up). An expert > player knows that cultivating the rack with an eye > toward bingo-ing OUT is the only way to win; Quackle > should understand that, too. > > It would be nice to have an option (maybe a checkbox) to > always include the highest-scoring play, even if it's not > a candidate that Generate Choices would select. Thinking > back to Maven, there was a 'Compare Moves' feature that > offered an opinion (sometimes useful, sometimes not) on > how one move stacked up against another. Not that Maven > was perfect; the very first time I ever saw it, it failed > spectacularly to find a winning move at the end of a game. > More on that in a footnote. > > For Quackle in the here-and-now, including the highest- > scoring play, even when it doesn't meet the criteria for > a "best" play, could be very instructive, especially when > doing post-mortem. If the game-losing 60-point bingo is > included in the Choices list, and ends up dead last in the > Sim, with a win %-age of 0, sobeit. > > John Hart > > > When Maven failed to find a winning endgame play: Circa > 1994, a friend bought Mac-Maven and we used it to do a > post-mortem on a game we played. We wrote down all our > racks, recorded all moves played, and set about reviewing > the game. I was behind by about 25 points with X, blank, > and 5 vowels on my rack; opp had just emptied the bag. > All the vowels were of the A-E-O type; no I or U. There > was a place to play either Xi or Xu (X on TLS) for 50, > winning the game, which was exactly what I had done. > Maven listed a slew of 17- and 18-point plays, all of > which were AX, EX, or OX, all keeping the blank on the > rack. All of those plays would have lost, because my > opponent was going to go out on her next turn. > > I still remember being shocked that an easy 50-point > play like Xi was not listed. "Isn't there a way to tell > it to list the highest-scoring play?" I asked. There was > not. No way to override the program's opinion about rack > leave, etc. That was a serious flaw, which I assume has > been fixed in the intervening years (someone who saw Maven > play in Toronto, please fill me in).
Maven missed an obvious move in an *endgame*? No way! Something about your recollection is off. A simple thing like having 1 tile in the bag would make a huge difference. Or if you forgot to mark a blank while analyzing the game. Regarding the number of plays generated: Maven gave you the ability to add another move to the Kibitzer. You must have missed that feature. The Compare Move feature only explained static evaluation. Static evaluation might not think the 60-point bingo was a bad move. You have to look at winning percentage for that. Or create the ability to explain why a simulation prefers a play. I made some attempts to prototype that feature, but never released them in a product. BTW, Maven would certainly not be improved if users had the ability to tweak its rack evaluations. Where do users get such notions... :-) Best Regards, Sapphire Brand
