Tony Thomas wrote:Is it possible that the differences in strength is due to the fact that both versions of Tiger didnt play against the same opponents? Tiger 15 played against much weaker opposition in my opinion and that could have possibily influenced the results a little bit.
It is possible but at some point you have to trust the guys doing the ratings. The SSDF and the CCRL are the two rating lists I trust the most because of their exceptional devotion and scientific approach.
There may be small errors in their testing but they do what is humanly possible to reduce these errors and report the possibility of the error.
And there are so many other chess programs to test that I understand they cannot spend too much time on one.
Tony Thomas wrote:George, if I remember correctly he estimated an 80 point improvement and in bullet time controls the prediction was indeed correct. It was Naum that expected a 100 point improvement and later changed it to 80.
You may be right about the figures Tony, or close. But i do remember him saying that his testing method was not the best, hence the less than expected elo increase on his part. Whatever the figures- that is not the point with me. I just hate to lose a programmer of his ability. I used to say Chess Tiger would come out of the box saying "Where is that damn King". Very interesting and enjoyable style of play.
"Where is that damn King"!
I just love this one. At one time I considered sending humorous comments to the GUI. It was not a novel idea, I remember a chess computer in the 80s (was it Boris?) displaying such comments while it was thinking. It added something to the magic of this computer.
I would certainly have added yours to the list of comments that Chess Tiger would have spit during the game!
// Christophe
Also, It would be much better for an engine to show these words in its analysis line just when it announces the checkmate in...or just when its eval jumps to >3
Tiger, Tiger, burning bright
In the forests of the night....
and just when the engine exits..the sound of the Tiger growling. (I know only one engine so far that actually says "good bye" just when the engine exits after the game, That's Homer!)
people like poem and music more than just pseud words...
swami wrote:
Also, It would be much better for an engine to show these words in its analysis line just when it announces the checkmate in...or just when its eval jumps to >3
Tiger, Tiger, burning bright
In the forests of the night....
and just when the engine exits..the sound of the Tiger growling. (I know only one engine so far that actually says "good bye" just when the engine exits after the game, That's Homer!)
people like poem and music more than just pseud words...
There's also Chispa smiling when things go well for him, and crying when losing...
Tony Thomas wrote:George, if I remember correctly he estimated an 80 point improvement and in bullet time controls the prediction was indeed correct. It was Naum that expected a 100 point improvement and later changed it to 80.
If memory serves weel enough, it happened more or less in both cases... That's life, and I guess it just means that those who's going to release commercial programs will do better in under-estimating, or people will say they felt cheated after a purchase.
tiger wrote:It is possible but at some point you have to trust the guys doing the ratings. The SSDF and the CCRL are the two rating lists I trust the most because of their exceptional devotion and scientific approach.
There may be small errors in their testing but they do what is humanly possible to reduce these errors and report the possibility of the error.
// Christophe
Hi Christophe,
I would be interested to hear, what conditions for testing are necessary to avoid most errors. Do you have some tipps?
tiger wrote:Zach is partly correct. I just don't call this cloning. I meant I could improve Chess Tiger the easy way by taking the source code of Strelka, changing it line by line until it could not be recognized anymore and finally adding a number of search tricks that I have in Chess Tiger and that are not in Strelka.
This would give a program that would approximately by 200 elo points stronger than Chess Tiger 2007.1
Whow!
Clone Tiger 2008 - the strongest single engine behind Rybka
The SSDF and the CCRL are the two rating lists I trust the most because of their exceptional devotion and scientific approach.
Where is CEGT leaving anything behind? I'm just curious
Best,
CEGT got a lot of games, but can you fiddle with their list like you do with CCRL. Also, do they have different lists, pure list, 32bit single list etc? Even their best list has more than one version of the same program. I think CEGT need to clone Kirill, then they will have the best rating list.
tiger wrote:It is possible but at some point you have to trust the guys doing the ratings. The SSDF and the CCRL are the two rating lists I trust the most because of their exceptional devotion and scientific approach.
There may be small errors in their testing but they do what is humanly possible to reduce these errors and report the possibility of the error.
// Christophe
Hi Christophe,
I would be interested to hear, what conditions for testing are necessary to avoid most errors. Do you have some tipps?
Not really. I know it's not easy, and my own testing, which has different goals but should be in theory easier, is difficult enough that I believe I have screwed it up for a long time!
However when I examine a rating list I'm going to have a close look to the rating of different versions of my own program, because I'm supposed to know with a good accuracy what they are. If the numbers I see, taking into account the error bars, are too far away from what I know, then I tend to believe that there might have been some errors in the testing methodology.
Now that does not help you much, I know.
However the ratings of the different versions of Tiger I see on the CEGT seem to be in line with what I expected.
tiger wrote:Zach is partly correct. I just don't call this cloning. I meant I could improve Chess Tiger the easy way by taking the source code of Strelka, changing it line by line until it could not be recognized anymore and finally adding a number of search tricks that I have in Chess Tiger and that are not in Strelka.
This would give a program that would approximately by 200 elo points stronger than Chess Tiger 2007.1
Whow!
Clone Tiger 2008 - the strongest single engine behind Rybka
...sorry, couldn´t resist
best
Don't push me too hard... What if I really do it!?
The SSDF and the CCRL are the two rating lists I trust the most because of their exceptional devotion and scientific approach.
Where is CEGT leaving anything behind? I'm just curious
Best,
CEGT got a lot of games, but can you fiddle with their list like you do with CCRL. Also, do they have different lists, pure list, 32bit single list etc? Even their best list has more than one version of the same program. I think CEGT need to clone Kirill, then they will have the best rating list.
I'm mostly concerned with reliablilty. And CEGT has it too. That's my point.
They were close to close down 2 year ago, but managed to make things work:)
Yes, more could be done when it comes to usefulness, but still, what's there is very good!
