Will Singleton wrote:position 3 might be busted, try rd5
What score and depth did you get?
Rybka has a score of +4 pawns at a depth of 22, and Zappa agrees almost exactly.
The best score I saw for Rd5 was from Colossus and Deep Frenzie, at +2 pawns, but the eval was falling. I will try analyzing the position after Rd5 with some strong engines to see what the result is.
It should probably at least be listed as an alternative (I will see what the score is after a good, long think).
The line is Rd5 Kf8 Rgd2 Rh8 c5, so I think it's just a case of white improving its position before finally playing c5. I used Rybka 1.0 and Shredder 7, but you should get a good score if you play Rd5 Kf8 Rgd2, then let any good engine analyze.
My program gets 830 at 1 minute, and 870 at 10 minutes. I think many positions are quite easy, even for Amateur. Still, it's a nice set of positions, and it can certainly be helpful in pinpointing specific areas for improvement.
Will
I see at least two programs that can improve mightily by implementing this concept:
Bright
Booot
Both scored abysmally compared to the others.
Dann Corbit wrote:
I see at least two programs that can improve mightily by implementing this concept:
Bright
Booot
Both scored abysmally compared to the others.
Good observations, Dann. I have already told Allard about it. I think Bright actually scores 80's when given more time for each test suites, Allard reported something around 80's-90's to me in 1 minute/each question time limit.
I agree it didn't do well when given less time, and that's the main problem but I doubt if that matters except in blitz tests.
Really interesting to see how a lot of engines respond to the test suites and their scores. I see Loop List and Junior having more scores than other engines, It does tell one thing: Loop and Junior have very high undermining skills.
I'd have expected a lot more from other engines as well, most of them seem to perform at 70's to 85's. Many Engines do have a lot to improve.
Perhaps I think I should add 100 more undermining problems for consistency so as to provide competitiveness and general overview of engine's individual performance, but I will get back to doing it once I have finished the rest of the chapters, and there are actually a lot more.
I just keep changing my mind on what chapters I should start doing. I have nearly finished "Activating inactive pieces" but suddenly I started doing "Open files and diagonals" just because I found this chapter very interesting. Atleast there's an enthusiasm in me, but I got to focus on one thing and finish that one completely before moving to another.
I must say Dann Corbitt has been really really helpful, a lot more enthusiastic, hardworking and more expierenced than myself, and I'm glad to have him in joint team as co-authors.
Dann Corbit wrote:
I see at least two programs that can improve mightily by implementing this concept:
Bright
Booot
Both scored abysmally compared to the others.
Good observations, Dann. I have already told Allard about it. I think Bright actually scores 80's when given more time for each test suites, Allard reported something around 80's-90's to me in 1 minute/each question time limit.
I agree it didn't do well when given less time, and that's the main problem but I doubt if that matters except in blitz tests.
For the given level of strength, it does not fit where we would expect it to fit in the list.
Given this observation, I expect that there will be a large boost in strength for Bright by implementaition of undermine test.
93 of 100 matching moves
13-01-2009 20:13:49, Total time: 0:23:50
Rated time: 12:39 = 759 Seconds
or 77/100 with 2 seconds/position.
A main difference with version 0.3a is the improved king safety, including pawn shield/storm.
Still, I wonder what the undermining concept is all about...
according to Swami, it's not king safety, so, what is it?
Dann Corbit wrote:
swami wrote:
Dann Corbit wrote:
I see at least two programs that can improve mightily by implementing this concept:
Bright
Booot
Both scored abysmally compared to the others.
Good observations, Dann. I have already told Allard about it. I think Bright actually scores 80's when given more time for each test suites, Allard reported something around 80's-90's to me in 1 minute/each question time limit.
I agree it didn't do well when given less time, and that's the main problem but I doubt if that matters except in blitz tests.
For the given level of strength, it does not fit where we would expect it to fit in the list.
Given this observation, I expect that there will be a large boost in strength for Bright by implementaition of undermine test.
93 of 100 matching moves
13-01-2009 20:13:49, Total time: 0:23:50
Rated time: 12:39 = 759 Seconds
or 77/100 with 2 seconds/position.
A main difference with version 0.3a is the improved king safety, including pawn shield/storm.
Still, I wonder what the undermining concept is all about...
according to Swami, it's not king safety, so, what is it?
{snip}
The idea of undermining or removal of the guard is to attack the chessman (usually a pawn) that is defending another chessman (often, but not always, a pawn).
Formal definition:
Removal of the Guard - A chess tactic where the defender is captured or forced to move so that it is no longer defending.
Will Singleton wrote:position 3 might be busted, try rd5
What score and depth did you get?
Rybka has a score of +4 pawns at a depth of 22, and Zappa agrees almost exactly.
The best score I saw for Rd5 was from Colossus and Deep Frenzie, at +2 pawns, but the eval was falling. I will try analyzing the position after Rd5 with some strong engines to see what the result is.
It should probably at least be listed as an alternative (I will see what the score is after a good, long think).
The line is Rd5 Kf8 Rgd2 Rh8 c5, so I think it's just a case of white improving its position before finally playing c5. I used Rybka 1.0 and Shredder 7, but you should get a good score if you play Rd5 Kf8 Rgd2, then let any good engine analyze.
My program gets 830 at 1 minute, and 870 at 10 minutes. I think many positions are quite easy, even for Amateur. Still, it's a nice set of positions, and it can certainly be helpful in pinpointing specific areas for improvement.
Will
Given these two positions after the key move, it appears that your move choice is nearly as good as the key move:
[D]1n6/4bk1r/1p2rp2/pP1RpN1p/K1P1N2P/8/P5R1/8 b - - id "Undermine.003 after Rd5";
[D]1n6/4bk1r/1p2rp2/pPP1pN1p/K3N2P/8/P5R1/3R4 b - - id "Undermine.003 after c5";
Yes, nice articles. My confusion is that I thought the suite was supposed to be strategic rather than tactical in nature.
Apperantly it is about reaching a better (strategic) position through an undermining tactic?
btw, if I run bright with material eval only, it still scores ~30/100 giving it only 2 seconds/position
Dann Corbit wrote:
Allard Siemelink wrote:(un)fortunately, bright 0.4a (private) already does much better:
93 of 100 matching moves
13-01-2009 20:13:49, Total time: 0:23:50
Rated time: 12:39 = 759 Seconds
or 77/100 with 2 seconds/position.
A main difference with version 0.3a is the improved king safety, including pawn shield/storm.
Still, I wonder what the undermining concept is all about...
according to Swami, it's not king safety, so, what is it?
{snip}
The idea of undermining or removal of the guard is to attack the chessman (usually a pawn) that is defending another chessman (often, but not always, a pawn).
Formal definition:
Removal of the Guard - A chess tactic where the defender is captured or forced to move so that it is no longer defending.
Depending on the position, it can be a strategic move or a tactical one.
See, for instance: http://rybkaforum.net/cgi-bin/rybkaforu ... l?tid=8813
It isn't totally clear that the removal of the guard is the best move (though Fischer won and Rybka liked it after 200,000 Seconds of searching).
Some of the positions are no-brainers and some of them are very difficult for the strongest engines. And of course, lots of them are intermediate in difficulty.
A weak program will probably have trouble with this test suite if we run it at less than one minute per position.
Rybka and Naum knock it flat even at 45 seconds/pos, so I doubt that they can learn anything from it.