Author |
Topic: 2011 Arimaa Challenge (Read 9711 times) |
|
Hippo
Forum Guru
Arimaa player #4450
Gender: 
Posts: 883
|
 |
Re: 2011 Arimaa Challenge
« Reply #45 on: Mar 21st, 2011, 5:26am » |
 Quote  Modify
|
And now marwin leads by 1,5 point.
|
|
 IP Logged |
|
|
|
Arimabuff
Forum Guru
Arimaa player #2764
Gender:
Posts: 589
|
|
Re: 2011 Arimaa Challenge
« Reply #46 on: Mar 21st, 2011, 10:31am » |
Quote Modify
|
on Mar 21st, 2011, 5:26am, Hippo wrote:| And now marwin leads by 1,5 point. |
|
It's strange how things turn out.
|
|
IP Logged |
|
|
|
UruramTururam
Forum Guru
Arimaa player #2537
Gender:
Posts: 319
|
|
Re: 2011 Arimaa Challenge
« Reply #47 on: Mar 21st, 2011, 11:35am » |
Quote Modify
|
Hm, that means that my discovery of time stealing opening against Marvin may be not a totally useless one. 
|
|
IP Logged |
Caffa et bucella per attactionem corporum venit ad stomachum meum.
BGG Arimaa badges - get your own one!
|
|
|
mistre
Forum Guru
Gender:
Posts: 553
|
|
Re: 2011 Arimaa Challenge
« Reply #48 on: Mar 23rd, 2011, 7:26am » |
Quote Modify
|
Marwin's lead is cut to 1. If one of piv, sconibulus, or camelback can beat Marwin, then it will be tied.
What happens if its tied? Also, has the 2nd place bot in the computer championship ever won out in the screening?
|
|
IP Logged |
|
|
|
UruramTururam
Forum Guru
Arimaa player #2537
Gender:
Posts: 319
|
|
Re: 2011 Arimaa Challenge
« Reply #49 on: Mar 23rd, 2011, 7:34am » |
Quote Modify
|
If tied - the champion wins.
And no, this time could be the first one that the computer champion and the challenge competitor are different programs.
|
|
IP Logged |
Caffa et bucella per attactionem corporum venit ad stomachum meum.
BGG Arimaa badges - get your own one!
|
|
|
aaaa
Forum Guru
Arimaa player #958
Posts: 768
|
|
Re: 2011 Arimaa Challenge
« Reply #50 on: Mar 25th, 2011, 11:40am » |
Quote Modify
|
Omar, could you change the layout of the bot scores such that a player gets a separate row for each color? That way, one could spot more quickly whether and with which counterpart a screening game is paired.
|
|
IP Logged |
|
|
|
aaaa
Forum Guru
Arimaa player #958
Posts: 768
|
|
Re: 2011 Arimaa Challenge
« Reply #52 on: Mar 28th, 2011, 4:07pm » |
Quote Modify
|
I rather had something like the following in mind:
Code:<html>
<body>
<center>
<table border=1>
<tr>
<th rowspan="2">Player</th>
<th rowspan="2">Color</th>
<th colspan="2">bot_sharp</th>
<th colspan="2">bot_marwin</th>
</tr>
<tr>
<th>Games</th>
<th>Points</th>
<th>Games</th>
<th>Points</th>
</tr>
<td rowspan="3">rbarreira</p>
<th>Gold</th>
<td><font color=red>*</font> <a target=_blank href="/arimaa/gameroom/comments.cgi?gid=175606">r 22</a></td>
<td>1</td>
<td><font color=red>*</font> <a target=_blank href="/arimaa/gameroom/comments.cgi?gid=175632">r 20</a></td>
<td>1</td>
</tr>
<tr>
<th>Silver</th>
<td><font color=red></font> <a target=_blank href="/arimaa/gameroom/comments.cgi?gid="> </a></td>
<td></td>
<td><font color=red>*</font> <a target=_blank href="/arimaa/gameroom/comments.cgi?gid=175920">g 26</a></td>
<td><s>1</s></td>
</tr>
<tr>
<th>Both</th>
<td></td>
<td>1</td>
<td></td>
<td>1</td>
</tr>
</table>
</center>
</body>
</html>
|
|
|
|
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2011 Arimaa Challenge
« Reply #53 on: Mar 29th, 2011, 12:29am » |
Quote Modify
|
With time nearly out on the screening, Marwin has a 0.5 point lead, but in all four unfinished pairs (ngerhart, camelback, piv, Sconibulus) sharp has a win while marwin has yet to be played. It would be a shame if the screening finished in this state, given that marwin has a fair chance of losing one of those four if all pairs are completed. Can any of you find time to finish up these pairs in the next couple of days?
[EDIT]
Oh, I see rbarriera also has an unfinished pair, with mawin having a win and sharp uplayed. That makes it slightly less lopsided, but it would still be great to see these pairs finished out.
|
| « Last Edit: Mar 29th, 2011, 12:46am by Fritzlein » |
IP Logged |
|
|
|
rbarreira
Forum Guru
Arimaa player #1621
Gender:
Posts: 605
|
|
Re: 2011 Arimaa Challenge
« Reply #54 on: Mar 29th, 2011, 2:25am » |
Quote Modify
|
I plan to play my game, although I'm quite unlikely to beat sharp... Maybe someone should post game comments asking those other players if they can play another game.
|
| « Last Edit: Mar 29th, 2011, 2:26am by rbarreira » |
IP Logged |
|
|
|
Fritzlein
Forum Guru
Arimaa player #706
Gender:
Posts: 5928
|
|
Re: 2011 Arimaa Challenge
« Reply #55 on: Mar 30th, 2011, 2:58pm » |
Quote Modify
|
Great, rbarriera. Thanks for finishing off your pair. Of course it is impossible for such a chaotic selection process to be entirely fair, but at least that will be one less "what might have been?" hovering around the outcome.
|
|
IP Logged |
|
|
|
UruramTururam
Forum Guru
Arimaa player #2537
Gender:
Posts: 319
|
|
Re: 2011 Arimaa Challenge
« Reply #56 on: Mar 31st, 2011, 2:50am » |
Quote Modify
|
We have an interesting situation now:
- The computer world champion, sharp, is likely not to win the Challenge qualification.
- Sharp seemed almost unbeatable at first and it had a substantial advantage at the beginning of screening period and then that advantage melted to less then nothing.
I think that it could explained by fundamentally different approaches sharp and marwin use to calculate their moves. Marwin represents a classical "search & evaluate" approach. Sharp uses a few clever heuristics to narrow its space of the moves of interest at very early stages of search.
What this is likely to cause: sharp plays more brilliantly. Just recall the "elephant in the hole" move during the crucial phase of its match against marwin during the championship. This approach is likely to take this bot opponents by surprise. But it leaves some space for weaknesses to explore: some tricks that mislead sharp's heuristics. When they are spotted sharp can be overrun by repetitive use of these tactics, while marwin still requires just careful strategical playing against it.
That may lead to a paradox that a human-like machine can beat machine-like one, but it's still not smart enough to fool well playing human opponents.
|
|
IP Logged |
Caffa et bucella per attactionem corporum venit ad stomachum meum.
BGG Arimaa badges - get your own one!
|
|
|
megajester
Forum Guru
Istanbul, Turkey
Gender:
Posts: 710
|
|
Re: 2011 Arimaa Challenge
« Reply #57 on: Mar 31st, 2011, 5:16am » |
Quote Modify
|
on Mar 31st, 2011, 2:50am, UruramTururam wrote:We have an interesting situation now:
- That may lead to a paradox that a human-like machine can beat machine-like one, but it's still not smart enough to fool well playing human opponents.
|
|
Well, that would seem to validate the Arimaa Challenge as a catalyst for genuine advances in AI. The future looks bright! We have 8 more years of titanic man-machine struggle to come...
|
|
IP Logged |
|
|
|
lightvector
Forum Guru
Arimaa player #2543
Gender:
Posts: 197
|
|
Re: 2011 Arimaa Challenge
« Reply #58 on: Mar 31st, 2011, 5:40am » |
Quote Modify
|
Actually, sharp is reasonably full-width. Any move is reduced in depth by at most 2 (and if it looks good, it gets re-searched at full depth again).
I maintain that sharp is really strategically weak. It just is reasonably good at tactics when the board gets really messy, and has one good trick. Maybe it's also a bad trick, I guess: give up a camel hostage and swarm a lot.
It turns out that being strategically weak is okay against other bots if you know how to do okay in messy fights, (although it's getting less and less okay), because other bots don't know how to consistently play "solidly" and punish strategic mistakes, and they make many as well. So my focus has been on making the eval tactically strong, because that's what beats other bots.
The problems are the usual ones for any traditional bot, namely that I didn't have any time to tune the "camel is afraid of elephant" evaluation (so sharp gives up camel hostages easily), and I didn't have time to tune the advanced rabbit evaluation (spent only a day writing the code, haven't revisited it since), etc. That is, it's a typical evaluation function with lots of holes.
Lots of work for next year though! 
|
| « Last Edit: Mar 31st, 2011, 5:45am by lightvector » |
IP Logged |
|
|
|
rbarreira
Forum Guru
Arimaa player #1621
Gender:
Posts: 605
|
|
Re: 2011 Arimaa Challenge
« Reply #59 on: Mar 31st, 2011, 2:50pm » |
Quote Modify
|
What is the actual screening period? In the 2011 challenge page it says "Mar 17, 2011 to Mar 29, 2011 ", but it started before that and there were at least two screening games after that, so the server might be still allowing screening games.
edit - oh, in the gameroom announcements it says "March 10 to March 31".
|
| « Last Edit: Mar 31st, 2011, 2:51pm by rbarreira » |
IP Logged |
|
|
|
|
Home
Help
Search
Members
Login
Register
Reply
Notify of replies
Send Topic
Print