Time loss in the IPON

[IWB]

Hello

Some know, some might be suprised but I am considering time losses as an engine feature. It IS part of the engine and if an engine loses on time I usually leave the result in the list!

So now the question is: How bad is it?

Here are some numbers with the current IPON results with 161200 games:

427 games lost on time overall, that is 0.26% of all games. Actually I could stop here as this is an irrelevant number ...

The Engine with the most games with time losses is Zappa, 66 of 9200 games. Right now I dont know how to sort this, but assuming they where all LOST by Zappa and it would have won ALL 66 games in a real match the performance would increase by 0.7% or something like 4 Elo - at best! As the elo average is higer than Zappa a better estimation would be that it might have won 50% of it which is a 2 Elo difference than.

Now I really stop, it IS irrelevant ...

Bye
Ingo

[Dr.Wael Deeb]

The number is irrelevant indeed Ingo....

Never the less,I don't allow a single time loss is my private rating list,but I operate the chess engines under completely different testing conditions....
Dr.D

P.S.I don't evaluate the time losing issue as a normal issue for the chess engines....a chess engine must never lose on time....this luxury is reserved exclusively for the humans

[IWB]

P.S.I don't evaluate the time losing issue as a normal issue for the chess engines....a chess engine must never lose on time....this luxury is reserved exclusively for the humans

I agree, but IF an engine looses on time and I decide to replay the game I manualy overrule the engine and interfere with the playing strength.

But it is academical as the number is so low ...

I found out that of the 66 time loss games, 16 where won on time losses and 50 where realy lost. The calculating is getting more complicated and it is not worth the whole writing here ...

Bye
Ingo

[MM]

P.S.I don't evaluate the time losing issue as a normal issue for the chess engines....a chess engine must never lose on time....this luxury is reserved exclusively for the humans

+1

[Sven]

a chess engine must never lose on time

In general I agree, but actually that depends on the time control. Especially the "sudden death" (i.e. no increment) type of games, which is not the usual case for rating lists but e.g. for testing based on many ultra-fast games, tends to produce some time losses when very long endgames are involved. That seems to be kind of unsolvable indeed, since even playing a "random" move instantly would consume at least some nonzero time.

The other type of games affected is when you use certain GUIs, of course ...

In my private testing related to my own development I remove all time losses from the PGN file, using a perl script, before evaluating it. Also I try to analyze all time losses of my own engine while ignoring those of the opponents. If an opponent engine produces too many time losses for my personal taste then I replace it by a different engine, to improve statistical quality.

Sven

[Engin]

some engine have a feature and they are prefer to lose on time if the position is hopeless.

what do you want to do with such engines ??

[Engin]

i am honestly i had many problems in the past with time losses too, everybody know that

especially reported last time many tournament managers to me that the version 4.80 to much lost on time with a given moves like near 40.

of course i fixed this so far, but its no guarantee that will no more happens in any other case. i dont implement a feature with prefer to lose on time if the position is anyway hopeless.

[IWB]

some engine have a feature and they are prefer to lose on time if the position is hopeless.

what do you want to do with such engines ??

Nothing! If such an engine exist you HAVE TO leave the time loss in or you overrule the intended "feature". That is why I usually leave time losses in!

Bye
Ingo

[Engin]

...and if this overtake the 0.7% ?

you say you will stop it.

[IWB]

especially reported last time many tournament managers to me that the version 4.80 to much lost on time with a given moves like near 40.

That sounds more like a problem with repeating time controls (While everyone usually states that repeating time controls is the natural way to play chess, it isnt! It is an admission to human capabilities. The natural way to play chess would be "Game in X Minutes" - done!)
Actually I read much more about problems with repeatiing time controls than with Fisher! I have the feeling that many just repeat what they usualy have heard or learned in the past...

Anyhow, your Tornade 4.8 did not lose a single game on time in the IPON. AS soon as it is a bit more quiet I will ad the latest Tornado as three monthes since the last inclusion are over at 24th of January as well.

Bye
Ingo