On-line engine blitz tourney September

[Joost Buijs]

On a lighter note, Myrddin's win over Nightmare was rather strange. Here's the position right after Myrddin left book.

[D]r3k1r1/p1qbnp1Q/p1n1p3/3pP3/5B2/P1p2N2/2P2PPP/R3KB1R w KQq - 0 14

Most engines give White at least a +1.5 advantage here. Nightmare must have huge bonuses for passers because in this position, instead of developing the kingside bishop like every other engine I tested, it uses it next THREE moves to push the h-pawn all the way to h6, giving Myrddin time to generate a nice queenside attack.
jm

This version of Nightmare was running on an old Intel Atom processor, I guess it was a little bit out-searched.
It is also possible that the tuning of the evaluation parameters is not what it should be.

A few months ago I introduced a stupid bug which messed up the automatic tuning of the evaluation parameters, recently I found the bug but since then I haven't done much to re-tune the engine.
It is difficult to use the parameters from a older version of the engine because in between many things have changed.

I'm working on a new version of the engine, totally rewritten from scratch, it is for a large part finished, but somehow I can't find enough inspiration to continue with it.
It is a lot of work to write an engine from scratch and each time I'm asking myself 'why am I doing this'.
Nowadays everybody seems to grab what they can from internet, and that is not exactly giving me inspiration.

[Modern Times]

There seemed to be in internet glitch somewhere: several participants suddenly lost connection simultaneously, and those that came back suffered the same fate several times in the 5-minutes or so that followed. Others were not affected at all, and the remainder of the evening there were no problems at all.

Yes that was weird. Cheng was in a dead-drawn game with Almere, and after re-connecting on the 4th time I think it was, the ICS then declared Cheng had lost on time, when it still had plenty on the clock. So Cheng lost half a point. Doesn't matter, these tournaments are just for fun mainly. These things can happen from time to time with the internet.

[op12no2]

I noticed that on Saturday my keepalive setting was not working - no date commands being issued. I just connected now (Monday) and it's fine. May or may not be relevant to the glitch. My connection is usually rock solid - it disconnected many times on Sat.

PS: John M: Myrddin went AWOL during its game with Lozza - while the scores were close I got the feeling Myrddin was getting an edge; so a likely +1 at least.

[hgm]

There seemed to be in inYes that was weird. Cheng was in a dead-drawn game with Almere, and after re-connecting on the 4th time I think it was, the ICS then declared Cheng had lost on time, when it still had plenty on the clock. So Cheng lost half a point. Doesn't matter, these tournaments are just for fun mainly. These things can happen from time to time with the internet.

Normally timeseal protects you from that, because it sends together with your move the time that actually expired between reception of the opponent's previous move and the one you send. If you really did flag, but have not sent a move yet, and the opponent gets impatient and tries to claim the game, the ICS sends a special signal to your timeseal to confirm that you exceeded the time, (i.e. how much time you used so far for this turn), and depending on the answer decides whether it will forfeit you.

The problem is that when the connection has gone dead, but the ICS does not know it, it will get neither a move nor a reply to the probe. At some point it becomes aware that you are disconnected (e.g. because you log in again on a new connection), and in that case it knows nothing better do do that subtract the time elapsed on the server clock since it sent you the opponent move from your time. The timeseal that knew how much you really had been thinking by then has been terminated together with WinBoard, without having been able to relay that info to the server. In this case your clock will be negative when the game is adjourned, and you will immediately forfeit when resuming it.

There is no easy solution for this.

[Daniel Mehrmann]

Yes that was weird. Cheng was in a dead-drawn game with Almere, and after re-connecting on the 4th time I think it was, the ICS then declared Cheng had lost on time, when it still had plenty on the clock. So Cheng lost half a point. Doesn't matter, these tournaments are just for fun mainly. These things can happen from time to time with the internet.

I had the same problem with one game (Rookie). After some reconnects Fruit lost on time with a position value of +8.xx.

However, a lot of people had disconnects at the same time. So, i guess a bigger internet node had a problem.
Well, I'm just playing for fun, so i don't have a problem with it and it's the internet: Such things can happen...

Regards,
Daniel

[hgm]

I wonder if there isn't anything that can be done against this. If the ICS would only know you disconnected, it would stop the clock at that point, and that would probably not make you forfeit.

But it seems wrong to subtract any time from the clocks when a game is adjourned due to a disconnect. The clock should basically be reset to the value it had when the player having the move started thinking. For engines this is obvious: you will have to reconnect with a new instance of the engine, which doesn't know anything about what the other did, and will see the adjourned position for the first time. It will even have lost its ponder work. Humans could in principle remember what the have already thought about. But they could resume the game after a few days, and have been thinking all that time without their clock running. So it makes little sense to only deduct the time that the ICS thought they were still connected. The only guard against such practices is to play with 'noescape', and then disconnecting is fatal in the first place.

[Daniel Mehrmann]

I have a question:

The last two tourneys my games was saved in the engine folder (subfolder of Winboard folder). Files: games.pgn (july) and games_online_tourney.pgn (august).
I don't use a special comand/parameter to save the games. So, i'm not sure how they are created.

This time no games was saved in a pgn file.

I'm not 100% sure, but i believe i didn't change anything on my winboard config or start file.

Do you know if a special parameter should be used to save games ?

My start batch:

Code: Select all

winboard -zp -ics -icshost winboard.nl -icshelper timeseal -fcp "polyglot _PG/fruit-reloaded.ini" -autoKibitz

fruit-reloaded.ini:

Code: Select all

; manual config

[PolyGlot]
EngineCommand = fruit-reloaded.exe
EngineDir = fruit-reloaded
Book = false
KibitzMove = true
KibitzPV = true
KibitzCommand = kibitz
[Engine]
Hash = 1024
BookMode = Tournament
NalimovPath = D:\Chess\TB\cis.uab.edu\hyatt\crafty\TB\3-4-5
NalimovCache = 256
Threads = 4

Regards
Daniel

[hgm]

The -saveGameFile (which is the name of the option that sets it, although you could use the shortcut -sgf) should be a persistent option, so once set you never have to set it again, unless you want to change it. Easiest way to set it is through the Save Game Options dialog. You also have to tick 'Save Games Automatically' there, and 'to file'.

Note that most of what you write in the polyglot.ini file (like hash size, number of threads, Nalimov cache) would be overruled by the GUI settings. Editing ini files is not recommended.

[JVMerlino]

On a lighter note, Myrddin's win over Nightmare was rather strange. Here's the position right after Myrddin left book.

[D]r3k1r1/p1qbnp1Q/p1n1p3/3pP3/5B2/P1p2N2/2P2PPP/R3KB1R w KQq - 0 14

Most engines give White at least a +1.5 advantage here. Nightmare must have huge bonuses for passers because in this position, instead of developing the kingside bishop like every other engine I tested, it uses it next THREE moves to push the h-pawn all the way to h6, giving Myrddin time to generate a nice queenside attack.
jm

This version of Nightmare was running on an old Intel Atom processor, I guess it was a little bit out-searched.
It is also possible that the tuning of the evaluation parameters is not what it should be.

A few months ago I introduced a stupid bug which messed up the automatic tuning of the evaluation parameters, recently I found the bug but since then I haven't done much to re-tune the engine.
It is difficult to use the parameters from a older version of the engine because in between many things have changed.

I'm working on a new version of the engine, totally rewritten from scratch, it is for a large part finished, but somehow I can't find enough inspiration to continue with it.
It is a lot of work to write an engine from scratch and each time I'm asking myself 'why am I doing this'.
Nowadays everybody seems to grab what they can from internet, and that is not exactly giving me inspiration.

I know how you feel. I've been trying to find motivation to fix Myrddin's very outdated eval, but just can't be bothered. There are also some problems in the SMP communication between the master process and the slaves. Still can't be bothered.

jm

[tttony]

I read this post too late

I would like to participate on this blitz tourney

Next time!!