| |
BGonline.org Forums
How To Use The Bots Most Effectively
Posted By: playBunny In Response To: How To Use The Bots Most Effectively (Seth)
Date: Thursday, 30 July 2009, at 12:58 a.m.
(And I might as well repost the response that I gave to your original post)
Should I lower the checkerplay of gnu to intermediate, advanced ,expert to realistically reflect the people that I play?
Bots can't realistically reflect people's play unless the people in question are experts. (And preferably bot-trained experts!)
Even 0-ply is immensely strong and it forms the basis of all bot play. The levels below Expert (which is unadulterated 0-ply) are implemented by adding varying degrees of "noise" to the evaluations. That changes the order in which it lists the candidate moves so that when it chooses the top one, it's not necessarily the one that's the best move. The bot's play is thus basically 0-ply but with random mistakes. The bot will play like a champ but make wierd, incongruous mistakes while doing so. For instance it'll know exactly when and how to build a prime and will do something advanced like slotting the front but then might "forget" to cover it, if the noise just happens to edge that move's evaluation out of the running. The greater the noise the more wierd the errors.
Human players tend to have consistency in their mistakes. One might be scared of leaving blots while another has a tendency to play with reckless optimism. Some run too early from an advanced anchor, some will destroy their home table rather than risk a blot in the outer table, some will play "frog in hot water" sitting on the ace point hoping to save the gammon until, oops, they've lost a backgammon. The bots don't know anything about human play.
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.