an interview with shay bushinsky in hebrew

[Uri Blass]

link to the interview that is in hebrew by GM ram soffer

http://www.chess.org.il/users/ShowNewsU ... ntTypeId=2

One of the questions is if there is an interest in playing with houdini or Rybka and if Junior has chances.

Shay say that there is always an interest and the question is if we are going to find sponsors for it.
Shay also believe that Junior has chances.

Ram soffer mention the fact that programs like houdini are clearly stronger by the rating lists

Shay replied that the rating list are based on games on personal computers and with random openings and one of the disadvantage of Junior is that it is relatively weak at blitz.

Shay explained also that the developers fit the openings to the ability of Junior.

He claims that it is possible that some program is better than Junior as an analysis tool on personal computers but they still can beat it when both sides use strong computers.

[TimoK]

Hello Uri

Ram soffer mention the fact that programs like houdini are clearly stronger by the rating lists

Shay replied that the rating list are based on games on personal computers and with random openings and one of the disadvantage of Junior is that it is relatively weak at blitz.

That's what we experience in every tournament with long TC such as the WCCC or especially in the DCSC Thüringen Tournament. In this (privately organized) tourney all the top programs are participating. Junior always gets a higher rank than it would be expected looking at rating lists. Why? Our guess was Junior needs time and / or good hardware plus a matched opening book. Seems like Shay's statement confirms our guess.

Btw Junior finished shared 2nd place this year, passing Rybka, Houdini, Stockfish and so on.
http://forum.computerschach.de/cgi-bin/ ... ?pid=38523 (German CSS-Forum)

Best regards
Timo

[IWB]

Hello Timo,

...
Btw Junior finished shared 2nd place this year, passing Rybka, Houdini, Stockfish and so on.
http://forum.computerschach.de/cgi-bin/ ... ?pid=38523 (German CSS-Forum)

But just quoting the 2nd place (Shredder was second as well, IIRC, and we both know what Shredder played there ...), especialy in the context you made first, is only half the truth. You where there, I was there, we both know that a tourney has some circumstances which have nothing to do with the playing strength of a participating engine ... I think we both can agree that playing strength is just one part of many for a tourney success, right!?

Nonetheless, very good complete tourney performance by Shay and Amir, congratulations!

Bye
Ingo

[TimoK]

Hi Ingo,

I think we both can agree that playing strength is just one part of many for a tourney success, right!?

Yes, that's exactly my opinion. In a tournament there are several factors that are crucial for success:

- the engine's playing strength
- opening preparation
- hardware
- smart and attentive operator (especially for observing the internal and external clocks and adjust the internal clock if necessary)
- ...

But I also think that Junior is one of the very few engines which benefits a bit more from long time controls than others. I can't proove it (yet), but if one of the programmers say that Junior is a weak blitz-player, I feel confirmed with my assumption.

Best regards
Timo

[Don]

Hi Ingo,

I think we both can agree that playing strength is just one part of many for a tourney success, right!?

Yes, that's exactly my opinion. In a tournament there are several factors that are crucial for success:

- the engine's playing strength
- opening preparation
- hardware
- smart and attentive operator (especially for observing the internal and external clocks and adjust the internal clock if necessary)
- ...

But I also think that Junior is one of the very few engines which benefits a bit more from long time controls than others. I can't proove it (yet), but if one of the programmers say that Junior is a weak blitz-player, I feel confirmed with my assumption.

Best regards
Timo

Junior is not THAT far behind everyone else, and they always come loaded up with the very best hardware which is a great equalizer.

I think what really separates programs in the big tournaments that are within 200 ELO of each other is the open book preparation. If there is 100 ELO difference you expect to win only about 64 percent of the games which is a nice edge but doesn't make a win anywhere near certainty. I think really good opening preparation can nullify that pretty quickly. It can be as simple as knowing that your program plays closed position (or open positions) better than the other programs and that can give you 50 ELO or something.

Some of these teams understand the strengths and weaknesses of their programs intimately and carefully design opening strategies around them. It's not necessarily outbooking your opponent to get to a position where you have an advantage, although that may be part of it, but it is also about just getting into positions your program plays with strength.

A lot of computer chess games are decided right out of the opening to a greater or lesser degree.

I think Hiarcs and Junior are outstanding in that regard and there is a lot you can do to improve your odds in big tournaments that effectively give you 100 ELO or more.

[Don]

I forgot to mention one other secret to winning chess tournaments. You have to go to them. If your program is a top program, but not necessarily at the very top, but you go to all the tournaments you will win your share simply due to the law of probability and statics.

Also it's not generally understood that if your program is say, 100 ELO stronger than the second best program at the tournament you are by no means a certain winner. In fact your chances go down as the number of players go up.

In 1995 Fritz won the WCCC ahead of Deep Blue. What is interesting about this is that I had talked to Murray Campbell before the tournament and they estimated their odds of winning the tournament to be only around 50%, and this was despite the fact that Deep Blue was at the time clearly superior to any other program at that tournament. Their calculation was based on simple probability and statistics, even Deep Blue has a small chance of losing each game that it plays and any given program with only 5 rounds can have an exceptionally good score, well above it's expectations. As it turned out, in it's game against Fritz in a tactical opening it fell into a horizon effect situation and Fritz was able to capitalize. The Kittinger program had already taken half a point from Deep Blue so Deep Blue was out of it. I don't really know if their estimate of winning was correct, but there is no question that had 1000 of these tournaments been played, Deep Blue would have won at least half, and I suspect well over half. But nowhere close to the "near certainty" than many expected.

Hi Ingo,

I think we both can agree that playing strength is just one part of many for a tourney success, right!?

Yes, that's exactly my opinion. In a tournament there are several factors that are crucial for success:

- the engine's playing strength
- opening preparation
- hardware
- smart and attentive operator (especially for observing the internal and external clocks and adjust the internal clock if necessary)
- ...

But I also think that Junior is one of the very few engines which benefits a bit more from long time controls than others. I can't proove it (yet), but if one of the programmers say that Junior is a weak blitz-player, I feel confirmed with my assumption.

Best regards
Timo

Junior is not THAT far behind everyone else, and they always come loaded up with the very best hardware which is a great equalizer.

I think what really separates programs in the big tournaments that are within 200 ELO of each other is the open book preparation. If there is 100 ELO difference you expect to win only about 64 percent of the games which is a nice edge but doesn't make a win anywhere near certainty. I think really good opening preparation can nullify that pretty quickly. It can be as simple as knowing that your program plays closed position (or open positions) better than the other programs and that can give you 50 ELO or something.

Some of these teams understand the strengths and weaknesses of their programs intimately and carefully design opening strategies around them. It's not necessarily outbooking your opponent to get to a position where you have an advantage, although that may be part of it, but it is also about just getting into positions your program plays with strength.

A lot of computer chess games are decided right out of the opening to a greater or lesser degree.

I think Hiarcs and Junior are outstanding in that regard and there is a lot you can do to improve your odds in big tournaments that effectively give you 100 ELO or more.

[Kingghidorah]

Hi Ingo,

I think we both can agree that playing strength is just one part of many for a tourney success, right!?

Yes, that's exactly my opinion. In a tournament there are several factors that are crucial for success:

- the engine's playing strength
- opening preparation
- hardware
- smart and attentive operator (especially for observing the internal and external clocks and adjust the internal clock if necessary)
- ...

But I also think that Junior is one of the very few engines which benefits a bit more from long time controls than others. I can't proove it (yet), but if one of the programmers say that Junior is a weak blitz-player, I feel confirmed with my assumption.

I used Junior against other programs in blitz so the programmers's assessment is correct from what I have experienced.

Best regards
Timo

[Laskos]

Hi Ingo,

I think we both can agree that playing strength is just one part of many for a tourney success, right!?

Yes, that's exactly my opinion. In a tournament there are several factors that are crucial for success:

- the engine's playing strength
- opening preparation
- hardware
- smart and attentive operator (especially for observing the internal and external clocks and adjust the internal clock if necessary)
- ...

But I also think that Junior is one of the very few engines which benefits a bit more from long time controls than others. I can't proove it (yet), but if one of the programmers say that Junior is a weak blitz-player, I feel confirmed with my assumption.

Best regards
Timo

Junior is not THAT far behind everyone else, and they always come loaded up with the very best hardware which is a great equalizer.

I think what really separates programs in the big tournaments that are within 200 ELO of each other is the open book preparation. If there is 100 ELO difference you expect to win only about 64 percent of the games which is a nice edge but doesn't make a win anywhere near certainty. I think really good opening preparation can nullify that pretty quickly. It can be as simple as knowing that your program plays closed position (or open positions) better than the other programs and that can give you 50 ELO or something.

Some of these teams understand the strengths and weaknesses of their programs intimately and carefully design opening strategies around them. It's not necessarily outbooking your opponent to get to a position where you have an advantage, although that may be part of it, but it is also about just getting into positions your program plays with strength.

A lot of computer chess games are decided right out of the opening to a greater or lesser degree.

I think Hiarcs and Junior are outstanding in that regard and there is a lot you can do to improve your odds in big tournaments that effectively give you 100 ELO or more.

Very correct. I said the same here some 5 years ago about Junior and its successes above pure luck in WCCC. There was once a rating list (CSS?) based on thematic openings (some 12-15 of the more popular ones, if I remember). Although, on average, Junior performed below some top engines, in some few specific openings it was performing well above its usual strength, on the top of most others. It had a very unusual variance with the opening choice, for example a normal, strong engine, scoring on average 70% was in the range 65%-75% for all chosen openings. Junior, even if on average was scoring lower, say generally 60%, on certain (few) openings had 80% performance (and on some openings, very low, 40% performance). I think in this respect Junior was pretty unique, even if nominally weaker than some others. And it allows the building of a very specialized opening book for it, where even without obvious traps and such, Junior slowly prevails in certain openings. Junior team has an excellent understanding of their engine peculiarity and uses this opening book preparation for a decade already, the more curious fact is that some others are not using some specialized anti-Junior books. Maybe it's hard, or maybe it's not allowed to tweak the book before each game, I don't know.

Kai

[IWB]

Hello Timo,

Hi Ingo,

I think we both can agree that playing strength is just one part of many for a tourney success, right!?

Yes, that's exactly my opinion. In a tournament there are several factors that are crucial for success:

- the engine's playing strength
- opening preparation
- hardware
- smart and attentive operator (especially for observing the internal and external clocks and adjust the internal clock if necessary)
- ...

But I also think that Junior is one of the very few engines which benefits a bit more from long time controls than others. I can't proove it (yet), but if one of the programmers say that Junior is a weak blitz-player, I feel confirmed with my assumption.

Best regards
Timo

Basicaly we agree (even if I doubt the "long time control" thing as I cant see it in any rating list - starting from CEGT/CCRL 40/4 over IPON, SWCR, CEGT 40/20, CCRL 40/40 - despite what the programmer hopes to see (given a resonable lower boundary)) but I want to add another important factor: Luck! Sometimes I have the feeling that this is 50% of the success in a tourney

Bye
Ingo

[Laskos]

I forgot to mention one other secret to winning chess tournaments. You have to go to them. If your program is a top program, but not necessarily at the very top, but you go to all the tournaments you will win your share simply due to the law of probability and statics.

Also it's not generally understood that if your program is say, 100 ELO stronger than the second best program at the tournament you are by no means a certain winner. In fact your chances go down as the number of players go up.

In 1995 Fritz won the WCCC ahead of Deep Blue. What is interesting about this is that I had talked to Murray Campbell before the tournament and they estimated their odds of winning the tournament to be only around 50%, and this was despite the fact that Deep Blue was at the time clearly superior to any other program at that tournament. Their calculation was based on simple probability and statistics, even Deep Blue has a small chance of losing each game that it plays and any given program with only 5 rounds can have an exceptionally good score, well above it's expectations. As it turned out, in it's game against Fritz in a tactical opening it fell into a horizon effect situation and Fritz was able to capitalize. The Kittinger program had already taken half a point from Deep Blue so Deep Blue was out of it. I don't really know if their estimate of winning was correct, but there is no question that had 1000 of these tournaments been played, Deep Blue would have won at least half, and I suspect well over half. But nowhere close to the "near certainty" than many expected.

To illustrate quickly your point, say Deep Blue is 200 Elo points above the competition level (five 200 Elo points weaker engines), I plot the expected result in the tourney of Deep Blue (points and probability) and other five weaker engines.



As expected, the Deep Blue result tends to be good, centering at 4.0/5.0 points in the tourney, even possibly getting 5.0/5.0. Now the five weaker engines:



They are not so skewed, their only trouble is Deep Blue, and share points among themselves. Their center at about 2.0/5.0 points.

Seems clear, those five will get about 2/5 points, Deep Blue around 4/5. But... as you see, both distributions are wide. Five engines trying to get 4.0/5.0 or even 4.5/5.0 are valid tries for one engine to succeed. One can calculate the probability of Deep Blue winning the tourney: 48% In a field of five 200 Elo points weaker engines.

The chances sequence as follows:

If Deep Blue wins all the games, it has 100% probability to win the tourney.
If it wins 4 games and draws 1 game, it already has only 57% to win the tourney.
If it wins 3 games and draws 2 games or loses only one game, it only has 37% to win the tourney.

The third possibility is the most probable, therefore that 48% total percentage to win the tourney for Deep Blue, even being much stronger.

Kai