Arimaa Forum - Print Page


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

        Arimaa >> Events >> 2013 Screening Fritzlein vs marwin timeout
        
(Message started by: omar on Mar 12^th, 2013, 10:23am)

Title: 2013 Screening Fritzlein vs marwin timeout
Post by omar on Mar 12^th, 2013, 10:23am

bot_marwin timed out in the screening game against Fritzlein.

From the logs, it seems marwin tried to make the move in time, but a network problem caused the move to be lost. The bot code should probably be improved to handle this better, but it's a hard case to handle since retrying may succeed later but give the bot a time disadvantage and not be noticed without examining the logs. We probably should have the bot interface software keep a "complaint" log as well to make such problems easier to detect.

For this game it needs to be decided if the game should stand as is, replayed from start or resumed from the current position.

We are in a bit of a awkward situation because the TD for the challenge/screening has by default been the TD for the WC; who as you know is Fritzlein this year. So for this game maybe one of the TCs for the WC can decide. Since I am a backup defender, Adanac is a primary defender, rbarreira has a bot in the screening, it leaves woh to make a ruling on the game.

So I am going to try and provide him the info here to make the decision.

The game in question:
http://arimaa.com/arimaa/gameroom/comments.cgi?gid=259831

marwin's net log file:
http://silver.arimaa.com/~marwin/logs/13810.netLog
At time Mar 11 16:38:27 2013 marwin tries to submit the move, but does not receive a response. About 10 seconds later it gets a game state that says it's still marwin's move and it submits the same move again at time Mar 11 16:40:21 2013. The server receives this, but the game has already timed out by now and the following request to get the game state tells marwin that it has timed out and so marwin leaves the game.

silver.arimaa.com load at the time:
http://silver.arimaa.com/logs/20130311/20/3801

arimaa.com load at the time:
http://arimaa.com/logs/20130311/20/3801

woh feel free to ask any questions you have before making a ruling.

Title: Re: 2013 Screening Fritzlein vs marwin timeout
Post by Fritzlein on Mar 12^th, 2013, 12:59pm

I understand the difficulty with restarting from scratch; marwin is disadvantaged by the fact that I know its move and have been able to think about the position in the interim. On the other hand, marwin's opponent (actually ziltoid more than me) would be disadvantaged to a much greater extent if the game restarted from scratch. Whenever a chunk of game has been played, it ought to count unless something makes it invalid. Recall that in the past we started from scratch when a process was stealing resources from the bot for the whole game, not when the bot was running fine and just had one network error.

One possible compromise is to restart from the final position with added time for marwin, perhaps added time equal to the maximum time per move. Not only does added time balance things better, it also creates a good chance that marwin will choose a different move than the one I can see in the logs.

I would rate the options:
1) restart from final position (good)
2) restart from scratch (poor)
3) let the result stand (bad)

Title: Re: 2013 Screening Fritzlein vs marwin timeout
Post by Boo on Mar 12^th, 2013, 2:40pm

Quote:

marwin's net log file:
http://silver.arimaa.com/~marwin/logs/13810.netLog
At time Mar 11 16:38:27 2013 marwin tries to submit the move, but does not receive a response. About 10 seconds later it gets a game state that says it's still marwin's move and it submits the same move again at time Mar 11 16:40:21 2013. The server receives this, but the game has already timed out by now and the following request to get the game state tells marwin that it has timed out and so marwin leaves the game.

So the question is, did arimaa.com server receive marwin's move the first time it was sent? If yes, but arimaa.com server failed to pass the move to Fritzlein - it is server's fault, and the game should be continued or restarted. If no, this is a failure at the client side, and the result should stand.
Otherwise the human player would be put at an unfair circumstances - it can happen that due to some kind of force majore (alien abduction or military invaders cutting off the internet cable) the move does not reach arimaa.com server, however the human-client doesn't even write logs, and it would be even harder for a human to prove he has clicked the "Send" button.
If we rule all connection/network glitches as client fault in HvH matches, why should we do an exception for a bot?

Title: Re: 2013 Screening Fritzlein vs marwin timeout
Post by Janzert on Mar 12^th, 2013, 5:00pm

on 03/12/13 at 14:40:03, Boo wrote:

For the world championship and challenge the bot server is run by Omar so any fault of either server (game or bot) or the network in between has been seen as a 'server fault', not the bots fault, thus properly deserving a redo. Basically the cutoff for what should count as a restart whether human or bot has been between what is under Omar's control and what is under the player's control. With the network connection between Omar's server and the player being the fuzzy, hard to determine, part.

Janzert

Title: Re: 2013 Screening Fritzlein vs marwin timeout
Post by omar on Mar 12^th, 2013, 10:08pm

Yes, probably the only way it would be the bots fault would be if the bot had a bug and crashed; or went into a loop and never even tried to send a move.

I like Fritzlein's suggestion of resuming the game and adding more time to the bot's reserve. It would be extra good if it caused marwin to pick a different move.

I had sent an email to woh this morning. I hope he sees it soon.

Title: Re: 2013 Screening Fritzlein vs marwin timeout
Post by woh on Mar 13^th, 2013, 6:23pm

After consulting the logs provided by the Tournament Coordinator Omar and in accordance with the Arimaa Challenge Rules (http://arimaa.com/arimaa/challenge/2013/) I decide the screening game Fritzlein vs marwin should be restarted from the final position with marwin to move on 24s without adding extra time to his reserve. The original game should be unrated.

Art. 4.2 under the section Technical Problems allows a restart in case the server fails to receive a move sent. At 16:32:58 marwin gets Fritzleins move 24g. Since mawins reserve is 7m42s he can use the maximum move time of 6m. At 16:38:27 he submits his move after spending 5m29s on the move. This is well enough within the 6m limit. At 16:38:37 the server responds by asking for the same move again. This is still within the maximum move time. I conclude the server is aware marvin has sent his move but the server fails to receive the move.

Art. 6 states the clocks should be restored as closely as possible to the state of the position from where the game is restarted. In this case the start of 24s with a reserve time of 11m11s for Fritzlein and 7m42s for marwin.

Title: Re: 2013 Screening Fritzlein vs marwin timeout
Post by Fritzlein on Mar 13^th, 2013, 7:16pm

Thanks for the ruling woh. I'll try to finish that game tomorrow.

Title: Re: 2013 Screening Fritzlein vs marwin timeout
Post by omar on Mar 14^th, 2013, 8:29am

Thanks woh. Fritz, just let me know when you are ready.

Title: Re: 2013 Screening Fritzlein vs marwin timeout
Post by 99of9 on Mar 18^th, 2013, 12:40am

on 03/12/13 at 12:59:52, Fritzlein wrote:

On the other hand, marwin's opponent (actually ziltoid more than me) would be disadvantaged to a much greater extent if the game restarted from scratch.

2) restart from scratch (poor)

Why is this? Is this always the case? Do you only know that this is so because you happened to know that you were winning? I haven't looked at the game, so without knowledge of the final position, I do not know who a restart would favour. It seems to me that always restarting server errors from scratch would be perfectly fair if it were a general rule, because overall it would give no advantage to either the human or the bot. (except the small advantage the human would get from getting experience against that bot... but in principle the bot gets this too, they just don't usually utilize it)

Whereas for your option 1, the human always gets the advantage of much pondering.

Title: Re: 2013 Screening Fritzlein vs marwin timeout
Post by Fritzlein on Mar 18^th, 2013, 7:05pm

on 03/18/13 at 00:40:42, 99of9 wrote:

Is this always the case? Do you only know that this is so because you happened to know that you were winning?

No and yes.

Quote:

It seems to me that always restarting server errors from scratch would be perfectly fair if it were a general rule, because overall it would give no advantage to either the human or the bot.

True, it would be fair as a general rule. Also it would be fair if, after every game, a coin were flipped, and the game were replayed from scratch if the coin flip came up heads. If every game had a 50% chance of being invalidated, neither the human nor the bot would be disadvantaged on average, so it would be fair.

However, in spite of being a fair rule on average, this rule would be unreasonable in each individual case that the coin came up heads. In each case, someone would be robbed of a victory that they had legitimately earned. The question is why a result that has been earned by a player should be thrown away by an event that was random, after the fact, and in no way connected to earning that result.

This is different from instances in the past where it was discovered that some rogue process had been running on the server along with the bot, stealing resources, and presumably weakening the bot's moves. Then the whole game played under unfair conditions should indeed be scrapped regardless of who is winning. Circumstances make all that has been played illegitimate.

My game with marwin, in contrast, was legitimate right up to the timeout. I earned my advantage fair and square. If marwin would have been ahead, that advantage would have been fair as well. Why throw away whatever advantage was there? Why shouldn't it count?

I submit that not counting the played portion of the game is arbitrary and wasteful, so there had better be a good reason for it.

Quote:

Whereas for your option 1, the human always gets the advantage of much pondering.

I proposed a mechanism to compensate for that. What do you think of giving the bot more time? Often this will result in a different move being played by the bot (note marwin didn't play the same move in the restart), so the human's extra thinking time has only general applicability. In any case extra thinking time would be an advantage to bot to balance the advantage to the human.