Not logged inRybka Chess Community Forum
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka vs Ehlvest TIME HANDICAP (WRONG!!)
- - By JhorAVi (***) Date 2007-07-08 07:40 Edited 2007-07-08 07:53
1. What's the sense of time handicap here?? You can just use half of the total Cores or use a slower computer and you'll arrive the same result in a fair match without humiliating us humans!

2. Rybkas strength reduction due to time handicap is INSIGNIFICANT against humans. It works only in Computer vs Computer since they scale better in time+hardware due to their consistency in search.

It turns out here that it's Ehlvest who's offering a handicap since he has lesser time to ponder.  The organizers should stay with equivalent time rule in order to give Ehlvest better chances.

As a Rule in Human vs Computer: Lesser Time control Equals Lesser chance for Humans.
                                                      Longer Time control Equals Better chance for Humans.

Set it to much higher time control instead! say 20/120 each! i'm sure Humans will stand a chance.

3. Given humans more time to think, even REVERSING the Handicap is possible!
Say 20/120 for Ehlvest vs 10/120 for Rybka.. Or 1 day for Rybka, or 1 week or 1 year! I'm sure it's Human who will benefit :)
Parent - - By Michael Hart (***) Date 2007-07-08 09:08
Rybka on inferior hardware, something like a Pentium II, would have made this match that much more interesting (for me at least), and of course made a time handicap seem unnecessary.   Anybody can afford or access one of these, but quads and the like are not readily available to everyone, for whatever reasons.  This is not to negate anything from this match, of course.  I have enjoyed the show very much so far and I am eagerly awaiting game six :)
Parent - - By -valdimor- (**) Date 2007-07-08 09:34
they want to show how strong Rybka can be so they use the best hardware

it's not about fairness
Parent - By Hetman (*****) Date 2007-07-08 09:49 Edited 2007-07-08 10:07
That Rybka is powerfull shows the ratinglist but Rybka camp is risking that  will be not next GM to fight. It is not a pleasure for top GM to be beaten by machine, even for 1500 $ in 4 days. The classical time control for human GM would be more appropriate. 2/40 + 1/20 etc.. and the match will be more interesting to watch.

Now, it is the play to one goal. One matter is the result other matter is marketing. It is good to  balance that but there will be comparison with Kramnik Fritz match. I am not marketing expert may it is the right way but I prefer to watch the footbal match where the powers of the teams are more less equal and result is in the margin of 3 goals of difference not 10.

Rybka is calculating on human time so time handicap is not so important and quick computer response might be confusing.

Rgds
Hetman
Parent - - By JhorAVi (***) Date 2007-07-08 09:35 Edited 2007-07-08 09:41
I agree and I still enjoy it and will continue to watch the match too..

But my point is that Time Handicap here is not just unecessarily humiliating to humans But also MISLEADING because many Dummies will falsely think that Rybka is playing HALF it's strength since it's playing half the time.  
This Handicap format is more of rybka promotion rather than Quest for truth..

I believe that humans will stand a chance against the fastest hardware in much longer but EQUAL time controls..
Parent - - By Uri Blass (*****) Date 2007-07-08 11:57
I think that the rybka team has every right to be misleading after the win of Fritz against kramnik was misleading and could cause wrong impression for many dummies that Fritz is the strongest program against humans.

Note that I was not impressed by the win of Fritz because I cannot be impressed by beating a player who cannot see mate in 1.
I am also not very impressed by the win of Rybka because it seems that Ehlvest was not prepared for the match(otherwise he could prefer to choose one opening and stick with it and this is usually what human players do in games that they play because they know that if they change opening every game they are going to surprise themselves more than surprising the opponent).

Uri
Parent - By Hetman (*****) Date 2007-07-08 15:21
I do not think that GM Ehlvest was not prepared for the match. He has not treated it in the result perspective, smaller loss or draw,probably. He had known that he is not able to win.
It is not difficult for player with such an experience to play with white 1. d4 d5 2. Nf3 Nf6 3 Bf4 e6 4. c3 etc or similar and build safe position.
He selected other way to play for fun of the audience and to show normal chess to the public. I can only say 'thank you' to him. He has shown in the games 2,4 and 5 that he is able to play on equal level. If he would have followed the strategy of early exchanges or blocked position we would have seen differrent result of games but other boring play either. Rybka would have not lost, for sure.

Regards
Hetman
Parent - By Uri Blass (*****) Date 2007-07-08 09:35
There is no rule in chess that you have to use all of your time.

Ehlvest offers no handicap because in case that rybka had the same time the programmers could still tell her to use less time if they think that it gives her better chances.

Maybe ehlvest gets no advantage from the fact that rybka is using less time but it is an absurd to claim that ehlvest is offering an handicap.

Uri
Parent - - By SR (****) Date 2007-07-08 09:49 Edited 2007-07-08 09:59
JhorAvi, I agree and your tread is in fact an elaboration of my post from the day before (the 7/7/07) where I wrote:

"Time handicaps are in my view complete nonsense. Since the search in chess to a large extend can be parallelized, one can also always give Rybka 8 times less, if one compensate by speeding up the search, by for example letting the future version of Rybka use 8 times as many processors. Well, I suppose in practice its a bit more complicated since searching in parallel on 8 machines can never be quite as efficient as giving one machine 8 times as much time. Anyway instead of taking of time handicaps, it is really hardware handicaps that matters. But I agree we want to see Rybka play for full strength and glory.

Also in real life the time handicap (without changing hardware) might in fact make it more difficult for the human, who have less time between his/her own moves."

I take Uri's point that no one prevent Rybka from using less time. Yes, using that argument less time for the computer can logically speaking not be an advantage, however then try to explain to the ordinary chess public (for a future match) that the computer is giving the human a new incredible handicap by promising that it will spend its time in a manner that maximize its playing strength.
Parent - - By JhorAVi (***) Date 2007-07-08 11:30
Yes.. I agree to let Rybka play at full strength and glory using the fastest hardware there is.. and using the maximum hash table.
These are my own corrections to the current rules:

1. Not Limit the Hash Table to 512MB because only very small ELO % is reduced (insignificant against Ehlvest)

2. Time handicap is useless as I stated. Give them EQUAL but longer time control because humans can benefit for it than computers do!
Hint: nobody has crossed the 40/120 barrier against computers so why not give it a try?

3. Too many of these insignificant handicaps give false impression that a severely crippled Rybka can still handily defeat a GM.

Simplify handicaps to only two that are very significant (currently implemented):

1. Limited Opening book to depth 3 because Opening book is not AI :)
2. No EndgameTable base because it's no AI either :)
Parent - By George Tsavdaris (****) Date 2007-07-08 13:25

>3. Too many of these insignificant handicaps give false impression that a severely crippled Rybka can still handily defeat a GM.


Why do you call them insignificant?
I agree they don't have the significance most they are talking about, but they have a great value and give some strong evidence.
When in the past such matches occurred the GM's were winning mostly, while now GM's almost all the times lose, then this has to make us understand some things.
At least to answer: Why this is happening? Did the GM's become weaker? Is the time control that is the factor here? If so why in the past at this time control GM's won?

My answer is simple to that: computers have reached the GM's and i guess they have surpassed them evenin classical time controls....
Not in the way they understand the game. But in the performance they demonstrate over the board....

>1. Limited Opening book to depth 3 because Opening book is not AI :-)
>2. No EndgameTable base because it's no AI either :-)


Why opening book and tablebases is not AI?
Do you suggest that we should remove hard disc drive ability to use from the computers?
Why?

Are some million loops and calculations of the form 0.4+0.2-0.1=0.5, being made on CPU, AI?

Computers use CPU and hard disc drive to play. Humans use their brain.
Why change this?
Parent - - By George Tsavdaris (****) Date 2007-07-08 13:12
I don't like any kind of handicaps too. I just hate them!
Let all opponents play with their usual and available options and weapons.

Human with his brain and computer with its compiled code, its hardware and everything that means, that is: Hard disc drives and CPU, that means opening books and tablebases and hash and bitbases and whatever other method of storing things and using them....

>2. Rybkas strength reduction due to time handicap is INSIGNIFICANT against humans. It works only in Computer vs Computer since they scale >better in time+hardware due to their consistency in search.


I don't agree at all.
A ratio of 1:2, that means half the time of your opponent, is significant against computers and against humans.
Against computers it has been calculated(for example the author of Chess Tiger supports that, as also the author of Rybka) that a double in speed gives ~+70 ELO more. 
Even if the +70, is smaller in human-computer matches, no logic i can think of says that it would be less than 10-20 ELO, for becoming insignificant.

>3. Given humans more time to think, even REVERSING the Handicap is possible!
>Say 20/120 for Ehlvest vs 10/120 for Rybka.. Or 1 day for Rybka, or 1 week or 1 year! I'm sure it's Human who will benefit :-)


Giving Rybka more time to think is obviously not better for human.
I don't know if it is better for Rybka but i can't find a reason to be better for the human.
Yes he would have more time to think when it will be Rybka's turn, but he would not have to think for one move but for various possible replies and this is the critical point where your suggestion is not logical....
Also giving Rybka 1 week or a year to think, is obviously much better for Rybka since the human will forget many of the variations he has thought during this time on Rybka's turn.
Parent - By Uri Blass (*****) Date 2007-07-08 13:23
It was not caluclated against computers.

1)Time handicap of 2:1 with pondering is smaller handicap than having hardware that is twice slower with no handicap
because time handicap of 2:1 means that the computer has more time to ponder.

2)We do not know if it is a significant handicap against humans.
The computer plays weaker with less time but the human also plays weaker if he has less time to ponder.

Uri
Parent - - By h1a8 (***) Date 2007-07-09 01:27

> Giving Rybka more time to think is obviously not better for human.
> I don't know if it is better for Rybka but i can't find a reason to be better for the human.
> Yes he would have more time to think when it will be Rybka's turn, but he would not have to think for one move but for various possible replies and this is the critical point where your suggestion is not logical....
> Also giving Rybka 1 week or a year to think, is obviously much better for Rybka since the human will forget many of the variations he has thought during this time on Rybka's turn.


As the number of plies searched goes up then the amount of time to the next ply increases exponentially. Meaning, that the size of the search tree grows beyond that which the engine can search through in finite time.
Thus, there is a limit to the number of plies an engine can usefully search through in reasonable finite time. Thus, increasing the time for the computer to think beyond its limiting ply depth would hardly increase its strength (if any at all). But in classical time controls, the strength of humans grows tremendously. For they are able to calculate almost as well as the computer  (accuracy and not depth) but with better strategy. I guarantee anyone that if ehlvest plays 40/120 tournament controls then the score would be a lot closer (if not better for him), no matter how long rybka thinks.
Parent - - By lkaufman (*****) Date 2007-07-09 04:48
Roughly speaking, each doubling of time allows Rybka to go one ply deeper (on average), which is quite significant. You are surely correct that doubling the human's time means more than doubling the computer's time, but I am pretty sure that you are not correct to say that a 50% increase in the human's time (roughly the difference between our time control and 40/2) would help him as much as a tripling of Rybka's time (i.e. also putting Rybka at 40/2). If both sides played at 30/2, then I might believe that the human's chances would be as good as they were under the time limits used in this match, as a doubling of the human's time might reasonably be worth as much as a quadrupling of the computer's. Of course I cannot prove this, it is just my best estimate based on having followed and/or been heavily involved with computer chess for the last twenty years.
Parent - - By h1a8 (***) Date 2007-07-09 06:07
I'm sorry if I am very much off here. Please correct me if I'm wrong about something.
I know almost absolutely nothing about computer programming (maybe a few basics that's it).
I was just using math. For example, if, on average, there are 20 possible legal moves at each ply then to reach the 10th ply
would mean 20^7 total positions searched which would take about 21 min to evaluate if calculating at 1000knps.
Now moving on to 8 ply would take about 7 hours to search through.

I understand that rybka weeds somewhat and uses hash transposition table to decrease number of positions searched.
But nontheless, the search tree still increases exponentially. And rybka must therefore weed out exponentially more moves just too increase her depth by one ply. This would make her definitely see very inaccurately beyond a certain point (since she is discarding exponentially possible plies).
Even if doubling the time allows her to go one ply deeper (this is still exponential) then she would still have a limiting ply depth in reasonable time.
For example, if it takes rybka 30min to calculate effectively to ply x then it would take her 1hr to ply x+1 and 2 hr to x+2 (here, she would lose on time).
So her limiting ply would be about anywhere from x-2 to x ply.
So whatever rybka's limiting effective ply is then she wouldn't be able to effectively search another ply deeper in the time limit given her.
Thus giving her more time (up to classical controls) wouldn't really increase much strength in her. But it would so the human.
But you are right (in another post), it probably isn't the best idea to play at classical time controls.
Parent - - By lkaufman (*****) Date 2007-07-09 13:59
I don't quite follow you. Going to 40/2 would have given Rybka roughly three times as much time (compared to 45'+15"). This means between 1 and 2 plies extra, if doubling = 1 ply. If the increased time is less than double, it just means that some moves get an extra ply and some don't (or a partial ply). Extra time always helps.
Parent - By h1a8 (***) Date 2007-07-09 16:09
Yes extra time helps rybka. But it helps the human way more.
The strength of both grows with extra time added (but with computers more slowly due to the exponential and limiting effect).
Meaning, the human would gain much more of an advantage than computers is time is increased for both. Thus effectively narrowing the gap
between man and machine.
Parent - - By Vasik Rajlich (Silver) Date 2007-07-08 17:51
The effect of search time on move quality in computer chess is very well documented. It's true that pondering does compensate for this somewhat. I don't have ponder statistics available, anyway this was discussed in a number of threads here.

One interesting idea is to give the human a really huge amount of time. For example, 1 hour for Rybka and 6 hours for the human. There is no question that the human will perform quite a bit better in this case. The question is if the audience can be kept interested.

Vas
Parent - By lkaufman (*****) Date 2007-07-08 18:52
Actually this was done in March, at the very same time as we were playing our pawn handicap match with Ehlvest. Program Zappa (on an Octal) played GM L'Ami at time odds of 4 1/2 hours to 25 minutes, with ponder off and no tablebases (but with full opening book, full hash tables, and alternating colors) a two game match in the Netherlands. Both games were drawn.
Parent - - By Sesse (****) Date 2007-07-08 23:53
Somehow I feel sorry for the human having to keep up with a match that potentially lasts seven hours -- I do realize that the GM will probably have time for breaks, though.

As mentioned in the chat during one of the games, the effect of ponder hit/miss in a very time-asymmetrical match will be huge, though. There are, IMHO, two good ways of solving that, depending on what handicap you want to give the GM:

* Either ponder will somehow need to take into account multiple different moves (ie. don't send a ponder move via UCI, just think for yourself without being directed by the interface), or
* Ponder should be turned off completely. The machine can display a cool screen saver or something ;-)

/* Steinar */
Parent - - By lkaufman (*****) Date 2007-07-09 01:00
     When the time handicap is moderate, like 2-1 in this match, normal pondering makes sense. When a huge time ratio is used, as in the L'Ami match, it is only logical to turn off ponder, as otherwise the true time handicap becomes highly variable and unknown. Also, when the time handicap is huge, the fact that the human can ponder while the computer thinks becomes insignificant.
Parent - - By Nelson Hernandez (Gold) Date 2007-07-09 11:04
I get the impression that some people will not be satisfied until a human wins against a computer, no matter how hamstrung the latter.  I think it is time to face the fact that desktop computers have not only left humans way behind, but they continue to pull away.  In the not-distant future the kinds of handicaps you would need to offer to attain parity will be so ludicrous such a match would have to be judged a pointless exercise.  I'll bet a chess program on a cell phone with no handicaps could trounce a GM now.
Parent - By lkaufman (*****) Date 2007-07-09 14:02
Depends on which GM.
Parent - - By Lee Ma Hong (**) Date 2007-07-08 23:53
One interesting idea is to give the human a really huge amount of time.  ...  There is no question that the human will perform quite a bit better in this case. The question is if the audience can be kept interested.

IMHO if the human is among the Top 100 in FIDE's list, you can get a larger audience and better press coverage ... and possibly a match sponsor from the corporate world, such as the manufacturers of the hardware components (to be) used by the Rybka team.
Parent - - By lkaufman (*****) Date 2007-07-09 00:57
     The human in this match (Ehlvest) is indeed among the top 100 in the current (and previous) FIDE list, and was once in the top five. We weren't exactly beseiged by corporate sponsors though.
Parent - By billyraybar (***) Date 2007-07-09 06:49
Has anyone talked with Intel PR?  One would think they would be interested in sponsoring something like this.
Parent - By Lee Ma Hong (**) Date 2007-07-09 12:43
The human in this match (Ehlvest) is indeed among the top 100 in the current (and previous) FIDE list

My bad (now where is that red-faced smiley?)
Parent - - By dragonx (**) Date 2007-07-09 11:48
What if Rybka were to simul 6 GM's all in 6 (or more if possible) game matches, with time odds of 1 hour to 6 hours, top GM receiving some sort of prize if he/she can achieve a plus score?  This should keep everyone interested during live games, and I bet a server like ICC would be willing to relay those games on ChessFM even if it consisted of top world or US GM's like Ehlvest. 

Matt
Parent - By Sesse (****) Date 2007-07-09 12:00
Getting one GM is hard enough, try getting six? :-)

Also, you'd probably need quite a bit of extra code if one copy of Rybka would be to simul six people at the same time.

/* Steinar */
Parent - By JhorAVi (***) Date 2007-07-09 12:22
Good Idea.. but Rybka isn't programmed to play simul. We can possibly run 8 instances of Rybka in an 8-core-CPU and play them against 8 GM's...... but that's no SIMUL......... i'ts TEAM CHAMPIONSHIP! :)

A better Idea is to allow the group of GM's to work as TEAM where they exchange IDEA's against Rybka..
It's no SIMUL either because Multiprocessor Rybka is also exchanging ideas with each Cores :)

Proposed Match Title:  MULTIPROCESSOR Rybka vs MULTIBRAIN GM   :) :)
Parent - - By Ankan (*) Date 2007-07-09 16:05
I too agree that longer time control would help the human player more. I suggest that for the next time try a match with ponder turned off,  standard 40/120 time (or more) for human and 40/30 for Rybka.
Parent - By Vasik Rajlich (Silver) Date 2007-07-10 12:52
This is not too different from the match which just finished. 90' + 30" gives the human 1 hour and 50 minutes for the first 40 moves, and he also gets two five-minute breaks.

At this point, the only meaningful experiments in this direction would be a really really long time control, like game in 6 hours.

Vas
Parent - By ubaldus (*) Date 2007-07-09 17:29
In 1980's, the standard time control was 150/40. Kasparov - Karpov matches were played with this time control. The point is that both spectators and grandmasters can easily handle a 5- or 6- hour game, and the quality of play only goes up.

IN the last match, Ehlvest blew two chances (one win and one draw) in time trouble. It is conceivable that with milder time control he would manage 2.5:3.5 or even 3:3 match.

One can have Rybka running on a fast machine (an octal) with 30'+10'' control, ponder off; and give GM 180'+60''. The game will last 4 or 5 hours, and time trouble issues will be greatly reduced. Rybka will still be her 2900+ self, even with all handicaps.
Up Topic Rybka Support & Discussion / Rybka Discussion / Rybka vs Ehlvest TIME HANDICAP (WRONG!!)

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill