five good reasons for the losses of rybka vs. zappa

[Ryan Benitez]

I guess that we will hear no feeble excuses.

I guess that they are not happy about losing, but will be good sports about it. I have great respect for the Rybka team.

And how can you be so sure that they will finally lose anyway....?

In order NOT to lose they just need 2 wins and 2 draws in the next 4 games.
Not so improbable.....
In fact we are talking about Rybka so i would say the most probable.

Assuming that all previous events are independent (perhaps invalid if learning is currently configured...):

To score 75% would indicate an Elo difference of 190 points. So I guess that it is somewhat improbable that Rybka will draw or win. In reality, the engines are very close in strength, so I think Rybka will most likely score about 50% the rest of the way.

The engines may be close in strength but I think the Zappa team is clearly stronger than the Rybka team in the conditions being played. I expect the Zappa team to score at least 2.5 out of the last 4 games. When people call the results luck it will be a real shame.

[Dr.Ex]

I guess that we will hear no feeble excuses.

I guess that they are not happy about losing, but will be good sports about it. I have great respect for the Rybka team.

And how can you be so sure that they will finally lose anyway....?

In order NOT to lose they just need 2 wins and 2 draws in the next 4 games.
Not so improbable.....
In fact we are talking about Rybka so i would say the most probable.

Assuming that all previous events are independent (perhaps invalid if learning is currently configured...):

To score 75% would indicate an Elo difference of 190 points. So I guess that it is somewhat improbable that Rybka will draw or win. In reality, the engines are very close in strength, so I think Rybka will most likely score about 50% the rest of the way.

The engines may be close in strength but I think the Zappa team is clearly stronger than the Rybka team in the conditions being played. I expect the Zappa team to score at least 2.5 out of the last 4 games. When people call the results luck it will be a real shame.

I would call them luck. Erdo did some clever preperation, but he also was very lucky so far.

[Tony Thomas]

I would call them luck. Erdo did some clever preperation, but he also was very lucky so far.

When Rybka loses everything is because of luck not due to bad design.

[mclane]

i would not call the strongest chess program of the world BAD DESIGN.

of course in a tournament you can have bad luck.
or bad preparing. that happens.
its because the programmers are no machines.
what do you believe how often david levy has seen those bad-luck situations ?

[Tony Thomas]

The reason Rybka lost one of the games was due to bad/good design.

[Dann Corbit]

I guess that we will hear no feeble excuses.

I guess that they are not happy about losing, but will be good sports about it. I have great respect for the Rybka team.

And how can you be so sure that they will finally lose anyway....?

In order NOT to lose they just need 2 wins and 2 draws in the next 4 games.
Not so improbable.....
In fact we are talking about Rybka so i would say the most probable.

Assuming that all previous events are independent (perhaps invalid if learning is currently configured...):

To score 75% would indicate an Elo difference of 190 points. So I guess that it is somewhat improbable that Rybka will draw or win. In reality, the engines are very close in strength, so I think Rybka will most likely score about 50% the rest of the way.

The engines may be close in strength but I think the Zappa team is clearly stronger than the Rybka team in the conditions being played. I expect the Zappa team to score at least 2.5 out of the last 4 games. When people call the results luck it will be a real shame.

All the evidence that we have says that they are about the same strength.
The tiny number of games played so far has basically no impact on that.
It is (of course) possible that Zappa has picked up 200 Elo (I highly doubt it though). It may be that careful preparation (e.g. 1000 games played against Rybka to tune the opening book) has resulted in some easy wins.

At any rate, I guess that both programs are still about the same strength.
After the contest, when both engines are in the hands of the testers, we will find out the real answer (was it fluky or has Zappa passed Rybka in strength).

[Tony Thomas]

I got another one

1) Zappa is playing with a BH, DC, or a JA compile.

[George Tsavdaris]

At any rate, I guess that both programs are still about the same strength.
After the contest, when both engines are in the hands of the testers, we will find out the real answer (was it fluky or has Zappa passed Rybka in strength).

Testers have it already on their hands. Zappa Mexico is being sold commercially as an UCI engine. I don't know if the Zappa Mexico UCI is the exact same with the Zappa that is playing at Mexico but early results show:

http://computerchess.org.uk/ccrl/4040/r ... t_all.html

Code: Select all

Rybka 2.3.2a 64-bit 4CPU        3117	+27	−26  514  games
Zappa Mexico 64-bit 4CPU       3095    +82     −81    42  games only so far	
Zap!Chess Zanzibar 64 4CPU   3051    +22     −21  679 games

Also we have from the same list that:

Code: Select all

Zap!Chess Zanzibar 64-bit 1CPU	2904	
Zap!Chess Zanzibar 64-bit 2CPU	3027 +123	
Zap!Chess Zanzibar 64-bit 4CPU	3051 +24  +147

Rybka 2.3.2a 64-bit 1CPU	3072 	
Rybka 2.3.2a 64-bit 2CPU	3098 +26	
Rybka 2.3.2a 64-bit 4CPU	3117 +19 +45

If we translate these we find that:
Zappa: 147 = C· a1 ^ (log2(n))
Rybka: 45 = C· a2 ^ (log2(n))

Where C is the theoretical ELO increase per doubling speed, a1,a2 are the speed efficiency per doubling speed and n the number of CPUs.

So we have for 4 CPUs of the ratings that a1 = a2· 1.8
That means the doubling speed efficiency of Zappa Zanzibar is 1.8 times bigger than that of Rybka.
This seems illogical but the extremely high increase of Zanzibar from 1 to 2 CPUs justifies this. Probably the high increase of Zanzibar from 1 to 2 CPUs is just an anomaly and with more it will become smaller....

Also from Rybka's FAQ we have that:

Code: Select all

Cores   Efficiency
2          1.7 
4          2.8 
8          4.4 

So we have:
From 1 to 2 CPUs: =1.7
From 1 to 4 CPUs: =1.67
From 1 to 8 CPUs: =1.64
A bad one at 8 CPUs.....

Returning to the a1 = a2· 1.8 we have that:
The rating gain(compared to 1 CPU) of Zanzibar on 8 CPUs would be = C· 1.8^3 ·a2^3
While the rating gain(compared to 1 CPU) of Rybka on 8 CPUs would be = C· a2^3

So the gain on 8 CPUs for Zanzibar would be 5.8 times bigger than that of Rybka. Perhaps this is enough for Zappa to catch Rybka on 8 CPUs.....

[Uri Blass]

I guess that we will hear no feeble excuses.

I guess that they are not happy about losing, but will be good sports about it. I have great respect for the Rybka team.

And how can you be so sure that they will finally lose anyway....?

In order NOT to lose they just need 2 wins and 2 draws in the next 4 games.
Not so improbable.....
In fact we are talking about Rybka so i would say the most probable.

Assuming that all previous events are independent (perhaps invalid if learning is currently configured...):

To score 75% would indicate an Elo difference of 190 points. So I guess that it is somewhat improbable that Rybka will draw or win. In reality, the engines are very close in strength, so I think Rybka will most likely score about 50% the rest of the way.

The engines may be close in strength but I think the Zappa team is clearly stronger than the Rybka team in the conditions being played. I expect the Zappa team to score at least 2.5 out of the last 4 games. When people call the results luck it will be a real shame.

I think that we can know nothing because we do not have the engines that play.

rybka is not using the public version and I doubt if zappa use the default zappa mexico.

Uri

[Uri Blass]

You forgot the main reason.
Convekta sponsored the match and they did not have to do it.

Uri