|
Post by Andrew Rose on Aug 13, 2014 11:58:44 GMT -8
Some of the games that are frequently played are quite biased. Sadly, this includes some of the games that are most interesting to watch (9B-TTT, Hex and even C4). I don't think I've seen any GDL that uses the " Pie Rule" to resolve this - although many would be amenable to it. I plan to code up a Hex w/Pie variant at some point (unless anybody beats me to it) but I just wanted to get this idea in the mind of the GDL authors out there.
|
|
|
Post by Steve Draper on Aug 13, 2014 12:24:27 GMT -8
Interesting idea - I look forward to seeing your Hex variant.
|
|
|
Post by Andrew Rose on Aug 13, 2014 14:07:12 GMT -8
Mostly, I look forwarding to *seeing* any Hex variant. That's my first task - to code up a visualization for the Tiltyard (see email conversation with Sam about <canvas>). After that, I'll try to do the GDL for the pie-rule variant.
--
Hex (normal) is a provable 1st player win. Somebody clever proved that you can't draw (which seems kind of obvious, but that isn't a rigorous proof). So (since it's non-simultaneous, complete information, always terminating, etc.) it must be a strong win either for the first player or the second player. After that, note that playing a piece can never harm your position - i.e. there's no position in which you'd like to no-op. Therefore, it must be a strong win for the first player.
Hex w/Pie is therefore a provable 2nd player win.
However, since the proof isn't a constructive one, and since state of the art players can't see far enough ahead anyway(*), it's still an interesting game to play and the pie rule *should* make it more balanced.
(*) Specialised programs have solved Hex for all opening moves on boards up to 9x9 (which is the size played in the Stanford repository), but there aren't any GGP players that can get anywhere near.
|
|
|
Post by Andrew Rose on Aug 16, 2014 12:19:29 GMT -8
For the Hex, the rules are fairly complicated, there's no visualization yet and it isn't on Tiltyard.
For 9-board tic-tac-toe, the rules are fairly straightforward, there's an existing visualization, it's on Tiltyard and it suffers at least as much (if not more) from 1st-player bias.
Therefore, I've submitted a Pie-enhanced version of 9BTTT to Sam for adding to the Tiltyard. Once I've seen the effect in play testing, I might take another look at Hex.
|
|
|
Post by Andrew Rose on Aug 18, 2014 0:30:44 GMT -8
Very early 9BTTT w/Pie results look promising. GreenShell managed to overcome ggtest1 as both first and second player. Previously, ggtest1 has managed to beat some of the good players when playing first at 9BTTT.
|
|
|
Post by Andrew Rose on Aug 18, 2014 9:47:37 GMT -8
|
|
|
Post by Andrew Rose on Aug 19, 2014 10:35:13 GMT -8
Okay, so in my little game-writing streak, I have adapted and/or written visualizations for the following. - 9BTTT w/Pie (GDL & visualization adaptation of the base.nineboardtictactoe). See a sample match on Tiltyard. - Hex (verbatim copy of stanford.hex rules with a new Tiltyard-compliant visualization). See a sample match on Tiltyard. - Hex w/Pie (GDL & visualization adaptation of the above). Not yet played on Tiltyard. The visualization for Hex is quite large. If it looks all wonky, either make your browser window wider or shrink the web page (Ctrl+- in most browers).
|
|
qfwfq
New Member
Posts: 29
|
Post by qfwfq on Aug 19, 2014 12:10:38 GMT -8
Okay, so in my little game-writing streak, I have adapted and/or written visualizations for the following. - 9BTTT w/Pie (GDL & visualization adaptation of the base.nineboardtictactoe). See a sample match on Tiltyard. - Hex (verbatim copy of stanford.hex rules with a new Tiltyard-compliant visualization). See a sample match on Tiltyard. - Hex w/Pie (GDL & visualization adaptation of the above). Not yet played on Tiltyard. The visualization for Hex is quite large. If it looks all wonky, either make your browser window wider or shrink the web page (Ctrl+- in most browers). Thank you for the delicious pies great work!
|
|
rxe
Junior Member
Posts: 61
|
Post by rxe on Aug 19, 2014 14:37:28 GMT -8
... - 9BTTT w/Pie (GDL & visualization adaptation of the base.nineboardtictactoe). See a sample match on Tiltyard. ... Thanks and very pretty! The 3rd move after accepting the pie in the above game, does looks a tad suspect. edit; guessing it needs something like so: (<= (next (currentBoard ?k ?l)) (true (currentBoard ?k ?l)) (true canSwap) (does blue noop))
|
|
|
Post by Andrew Rose on Aug 21, 2014 0:01:39 GMT -8
Thanks Richard. Agree that there's a real issue you've identified there. However, I'm not sure you proposed fix is quite right. (The currentBoard isn't lost on transition into the second turn, it's lost on the transition into the third turn iff 'swap' is played.) I think the correct additional rule is...
(<= (next (currentBoard ?k ?l)) (true (currentBoard ?k ?l)) (does blue swap))
Or to paraphrase... "If blue plays 'swap', the 'currentBoard' doesn't change."
I have tested this change and it seems to work. Fix submitted to Sam to updated the Tiltyard.
|
|
rxe
Junior Member
Posts: 61
|
Post by rxe on Aug 28, 2014 7:29:45 GMT -8
Thanks Richard. Agree that there's a real issue you've identified there. However, I'm not sure you proposed fix is quite right. (The currentBoard isn't lost on transition into the second turn, it's lost on the transition into the third turn iff 'swap' is played.) I think the correct additional rule is... (<= (next (currentBoard ?k ?l)) (true (currentBoard ?k ?l)) (does blue swap)) Or to paraphrase... "If blue plays 'swap', the 'currentBoard' doesn't change." I have tested this change and it seems to work. Fix submitted to Sam to updated the Tiltyard. Ah yup, your change makes sense and what I was getting at, with mine being a move too early. I should of probably tested at least somewhat before posting. Thanks again Andrew!
|
|