Arimaa Forum - Print Page


    
      
        Arimaa Forum
        (http://arimaa.com/arimaa/forum/cgi/YaBB.cgi)
      

        Arimaa >> Events >> 2010 Computer Championship
        
(Message started by: omar on Mar 3^rd, 2010, 10:54am)

Title: 2010 Computer Championship
Post by omar on Mar 3^rd, 2010, 10:54am

In case you missed the live announcement regarding the computer championship and challenge after the The_Jeh vs 99of9 game you can hear it near the end of this video. It is about 2 hours into the video: 2:03:00
http://arimaa.com/arimaa/videos/2010wc/game136197.avi

I want to thank the bot developers for participating in this years tournament. I've seen a lot of effort put into the bots this year and there has been a noticeable improvement over last year.

I also want to thank everyone who has been helping out with the events this year. All the live commentary and excellent coverage in the wiki are really making the events quite interesting to watch and follow. There is no way I would be able to do this without the help of the community. It's incredible that we are keeping such a great historical record of the events at such an early stage of the game.

Before the computer tournament starts I select the defenders for the challenge match and give it in writing to the TD. We usually announce the challenge defenders after the computer championship. But this year we are announcing it at the start of the computer championship. Perhaps RW can start a thread for the challenge match and also announce the defenders there.

Title: Re: 2010 Computer Championship
Post by omar on Mar 3^rd, 2010, 11:11am

Yesterday, we encountered some problems with the bot_clueless vs bot_GnoBot game of round two.

Here is an account of what happened and included is some information which would be useful for the bot developers.

After the game timed out with GnoBot losing on time, my analysis of the log files seem to indicate that GnoBot had indeed picked a move within the move+reserve time that it had and tried to send the move to the server. I concluded that the fault was not GnoBot's and that probably a network lag or a HTTP reconnect attempt may have caused the timeout and suggested resuming the game. The TD approved resuming the game.

So we did try resuming the game and it timed out again. At that time we did not have enough information to know what caused the second timeout, but it did not seem to be GnoBot's fault since its logs showed that it picked a move and tried to send it and it used less time to pick the move than the move time plus reserve; leaving about 6 seconds to spare. We turned on logging to the netLog file also for GnoBot and resumed the game again and it again timed out. However from the netLog file we noticed that it took about 12 seconds between when I restored the game to when the bots joined the game. So GnoBot thought that it had 12 seconds more of reserve than what the server was giving it. After a few more attempts I was able to reduce the delay between restoring the game and bots joining the game to 4 seconds. However on that resume GnoBot picked a different line than what it had been picking before and used much less time to select the move leaving well over a minute of time in reserve. The game continued and finished with clueless winning the game by goal.

I considered the option of adding about 12 seconds of reserve to GnoBots time to compensate for the 12 seconds of delay between restoring the game and the bots joining the game. However that may not have worked if GnoBot decided to use as much time as possible and leave only 6 seconds in reserve. The game would again time out. The only way for GnoBot to know how much time it actually has for thinking on this move would be to use the formula:
"max thinking time" = "time per move" + "reserve at start of move" - "time used since start of move"
The three quantities on the right side of the equation are provided as parameters to the bot. However, I think GnoBot was not subtracting the "time used since start of move". Normally in a game which is not resumed this quantity is typically zero and so it never shows up when game are played continuously from start to finish. If GnoBot had used the above formula then adding some time to GnoBot's reserve when resuming the game would have prevented timeouts and helped to provide compensation for the delay between resuming the game and getting the bots to join. I am at fault for not clearly specifying to the bot developers the formula for calculating the "max thinking time" in the documentation I provide. Actually an even better way for the bots to calculate the "max thinking time" would be to compute "time used since start of move" on their own using other parameters provided to the bots:
"time used since start of move" = 2 * "current time stamp on game server" - "current time stamp on bot server" - "time stamp on game server when move started"
This would allow the bots to also compensate for any network lag in the message getting from the game server to the bot server.

The TD has suggested that in the future some guidelines should be provided to the bot developers on how much time to leave in reserve to compensate for network lag or server load that could delay the message between leaving the bot server and being recorded on the game server.

Title: Re: 2010 Computer Championship
Post by 99of9 on Mar 3^rd, 2010, 1:26pm

It seems a bit unfair on the restart bot to have any significant "time used since start of move". I'm certainly happy to account for it next year, but it concerns me that it can be 12s long.

The last method of calculation would cause problems if the two computers were not properly synchronized. I know that's not the case for the CC, but I can imagine it cropping up at some other point.

Title: Re: 2010 Computer Championship
Post by Fritzlein on Mar 4^th, 2010, 10:03am

Sharp continues to amaze me. Despite running on only one processor, it has outplayed OpFor, GnoBot, and Badger, all parallelized opponents using four cores. I'm guessing it will get a shot at clueless in round 5, a match I am suddenly looking forward to.

Title: Re: 2010 Computer Championship
Post by Hippo on Mar 4^th, 2010, 11:33am

Yes, I am looking forward even for Sharp-Marwin game.
Hope no more illegal moves.

Title: Re: 2010 Computer Championship
Post by aaaa on Mar 4^th, 2010, 12:27pm

on 03/04/10 at 10:03:14, Fritzlein wrote:

I'm guessing it will get a shot at clueless in round 5, a match I am suddenly looking forward to.

Sharp will only play against Clueless next if Marwin loses the next match against GnoBot (which, incidentally and funnily enough, would give Marwin the bye). Otherwise, Sharp will play Marwin, which to me seems like an equally interesting match up.

Title: Re: 2010 Computer Championship
Post by chessandgo on Mar 5^th, 2010, 2:58am

on 03/03/10 at 10:54:17, omar wrote:

http://arimaa.com/arimaa/videos/2010wc/game136197.avi

I listened a little bit to this video, the sound is terrible. Do we sound like this in the TeamSpeak room? If so we need to work on our mic settings. When I'm spectating the sound is much better.

Do you think there is a sound quality loss when encoding the video Omar? Losing image quality is no problem imo, but I feel like the sound is really what matters to make replaying these commentaries pleasant.

Title: Re: 2010 Computer Championship
Post by Adanac on Mar 5^th, 2010, 6:28am

on 03/05/10 at 02:58:37, chessandgo wrote:

Sometimes the audio sounds perfect during the broadcast but poor in the recording. During this Chessandgo-Fritzlein game I could understand Nombril perfectly but for some reason (only) his voice got distorted during the recording process.

http://arimaa.com/arimaa/videos/2010wc/game135491.avi

But then when Eric commentated the next game an hour later the quality of the recording was perfect:

http://arimaa.com/arimaa/videos/2010wc/game135493.avi

Title: Re: 2010 Computer Championship
Post by Eltripas on Mar 5^th, 2010, 6:29am

on 03/05/10 at 02:58:37, chessandgo wrote:

Yes it sounds terrible, no, it is not because "we" sound like that in Teamspeak, is the codec Omar used to encode the audio , sure it makes videos lighter but I think is too much sacrifice, when I recorded the game "The_Jeh vs woh" he recommended me to use that codec, I made a previous test, and thought it didn't was the best option, so the audio in my video is mp3 32kbps mono, you can hear it and give your opinion (http://arimaa.com/arimaa/videos/2010wc/game135493.avi) , I think it is like 20% heavier but it worths it.

Title: Re: 2010 Computer Championship
Post by omar on Mar 5^th, 2010, 12:36pm

Yes, I was trying to keep the size of the files as small as possible. But I'll increase the audio setting in future recordings.

Title: Re: 2010 Computer Championship
Post by omar on Mar 6^th, 2010, 8:09am

I used a slightly higher audio quality for the recordings in games 136706 and 136807. The audio sounds much better, but the problem now is that it gets more and more out of sync with the video as it progresses.

Title: Re: 2010 Computer Championship
Post by Fritzlein on Mar 12^th, 2010, 1:52pm

Congratulations marwin! I'm thrilled we made the format triple elimination just so that it would be possible to come from two games down like that. This certainly was the year for a dramatic showdown.

Title: Re: 2010 Computer Championship
Post by RonWeasley on Mar 12^th, 2010, 2:27pm

bot_marwin is this 2010's Computer Arimaa Champion. Congratulations on a close victory over worthy runner-up bot_clueless.

Both of these bots go on to the screening portion of the Arimaa Challenge. Good luck to both!

Title: Re: 2010 Computer Championship
Post by omar on Mar 13^th, 2010, 6:09am

Congratulations to both marwin and clueless (and their trainers of course :-) ) for finishing 1st and 2nd and moving on to the screening phase of the challenge match.

Also congratulations to OpFor and trainer Janzert for finishing 3rd.

This years computer championship was incredibly dramatic, I never would have guessed that marwin would come back and win it after having lost two games earlier.