| |
BGonline.org Forums
Questions about variant engines - BGBlitz 3.2.1 is published
Posted By: MK In Response To: Questions about variant engines - BGBlitz 3.2.1 is published (Frank Berger)
Date: Monday, 3 February 2025, at 9:58 p.m.
I had asked about all your 7 variants but you addressed only 2. Okay, never mind Crawford and Jacoby but the others deserve answers.
Frank said: About 20 years ago I trained a Nackgammon-AI ..... I think it is plausible that they converge.
I had alrady said you could skip Nackgammon since it's just a starting position variant, played with the same engine in other bots also. All such positions can be set up manually and played starting from there. All the bots do is save the user the hassle of setting up the position, only for Nackgammon for whatever special reason for a lame variant in my opinion. There must be unlimited number of starting positions to lead to unusual, more difficult decisions, ironically more so for the bots than for humans. ;)
Frank said: Surely for e.g. Snake it would undoubtly be better if an AI was trained especially for that, but having limited time and the few requests I wont do that.
If you can't do it right, why bother doing it half-ass, which amount to you BG-Bzzzt bot faking it and misleading the player to think he is playing a real variant?
Also said by email: Why should the equity of the very same position be different in e.g. Backgammon and Nackgammon (for Egyptian Rule this is not true)?
Again, skip Nackgammon. Thanks for admitting that your Egyptian Rule variant is a total fake.
The same applies to your Longgammon maybe to a lesser degree. The rules being same doesn't mean that the equities and strategies are the same. Just eliminating backgammons doesn't turn cubeless Gamblegammon into your Tables variant. Backgammonless equities are different enough to deserve a different engine.
Other bots are doing this wrongly also, including by not allowing doubles as opening rolls and omitting other less important rules during bearing off, etc.
Frank said: I don’t believe that any existing bot plays correctly (that would mean BG is solved), just that a few play better than any human.
Maybe they play "better than most humans" but surely not "better than any human".
GnuBG implements three Hyper-Gammon variants using claimedly solved cubeless nets, still applying jackoffski folrmulas to extrapolate cubeful equities. They are also just starting position variants, as if with fewer checker left on the board in late games. Thus they could be played by the same Gamblegammon nets just as well even if not (questionably) "perfectly".
Misfortunately, no bot implemented my HypestGammon which Tim Chow had claimed to have solved cubefully. I will write about it and CancelGammon separately. Maybe you or the GnuBg team will take on the challenges of implementing them.
MK
PS: I often criticize your BG-Bzzzt bot harshly, at the risk of hurting your feelings but you should still appreciate it since quite apparently I'm only one who pays any attention to your bot which (unjustly) almost never gets even distantly mentioned anywhere...
| |
BGonline.org Forums is maintained by Stick with WebBBS 5.12.