Author: Amir Ban
Date: 03:06:06 07/13/04
Go up one level in this thread
On July 13, 2004 at 00:34:02, Uri Blass wrote: >On July 12, 2004 at 19:58:20, Amir Ban wrote: > >> >>Let me first thank those who sent their support during the tournament or posted >>their greetings after it. It is well appreciated. >> >>Time to say some random things about Junior, the event, and computer chess. >> >>Having the Championship at home was different, in some expected and some >>unexpected ways. I'm pretty used to these events by now, and I found that when >>away from home, isolated from family, work and current tasks, they are an >>about-right balance between the hectic and the relaxed. This time, with the >>tournament hall a half-hour drive from home, it was almost more than I could >>handle. I need a rest. >> >>The organization was a pleasant surprise. Playing conditions were superb. There >>were far more plugs and connections than necessary. Technical support was very >>competent. Commentary by Boris Alterman was first class. >> >>Nevertheless, there were some disappointments: PR work was not evident. The >>media were simply not there. Some media coverage would have attracted the >>crowds. As it was, the nearby auditorium they reserved for spill-over crowds (my >>daughter counted the seats: 320!) remained unoccupied. >> >>The traditional closing banquet was waived this time, in favour of speeches and >>glasses of wine. A pity. The opening promise by David Levy for social events did >>not materialize. >> >>The starting point for Junior in this tournament was Graz, from which I returned >> with a sense of failure, realizing that I have failed to progress beyond J8 >>levels. Still, that was almost good enough to win, so there was hope. >> >>I had quite a few major things mapped out to be tried. Most of them were >>abandoned after a week or two of work showed they were not productive. > >Did you try only changing the evaluation or also changes in the search? > Both. I'm more successful improving the evaluation than improving the search. >> >>In the midst of this open brain surgery we decided to pull out something to play >>in CCT6. It proved to be quite entertaining, but hardly ready for big time. >> >>Meanwhile Shay adopted a testing strategy that proved particularly unflattering >>to the new Junior, indicating it had slipped by a 100 or so rating points. I was >>skeptical about that, but decided to continue on the current path, trusting good >>results to follow. That proved exceptionally difficult. As a matter of fact, >>until weeks before the championship it was not clear that we were not heading >>for a rude embarrassment. What changed this was that at some point Boris started >>telling us that he liked what he saw. > >I am not sure if I understand correctly. > >Do you say in other words that Boris helped you by suggesting to test the >correct version(the version that he liked)? > I'm saying that testing results can be confusing and contradictory. Often you have to cut the Gordian knot by using judgement. >> >>Going into the first round, against Jonny, it seemed we have after all achieved >>nothing. Deep Junior was in imminent danger of losing for several moves, grossly >>misevaluating and apparently blind to tactics. However, it did manage to pull >>together the position and successfully negotiated a draw. >> >>After that game I threw away a feature I always knew was dubious to get the >>DeepJuniorEY version, which played the bulk of the tournament, and performed >>very well. > >If you always knew it was dubious why did not you do it before the first round? >Is it something that you planned to do and forgot about it? > Sometimes things test well, but your judgement tells you that they can't be good. > In the last day we threw in the EM version, which was apparently >>superior, based on inadequate testing. That was a gamble, but the idea was to >>try it out in the last two rounds before the expected playoff. EM did not fail, >>but there was no playoff for it to play ... > >Is the difference between EY and EM is evaluation or search or both? >Did the draw against Isichess influenced you to make some changes? > No. In this case the difference is the evaluation. >> >>Book work is tremendously important in these tournaments. > >Do you think that you won the tournament thanks to better book preperation that >helped Junior to get positions that it evaluates correctly inspite of slightly >inferior engine(relative to shredder8 because you cannot compare the engine with >Shredder of the tournament that you do not have). > >It is not a secret that Junior8 is weaker than Shredder8 and if you did not do a >significant improvement(I hope you did) then I do not expect Junior of today >even to be at the same level as Shredder8 if you use some book that is not >optimized for one of the engines like Fritz8.ctg or you use something like the >nunn match. > >Do you test with different book to try to evaluate progress of the engine or >alternatively use the nunn match(possibly with different random starting >positions from games) or do you plan to do it now after the tournament? > These are all parts of testing procedure, that contains other things as well. > > > It sometimes seems as >>if Boris is in charge, while I and Shay are delegated to the role of >>technicians. The symbiosis between Boris and Junior has grown and matured since >>the early days, when they hardly understood each other. That was the state of >>things at Paderborn '99, when Boris brilliantly planned a Sicilian line into >>which Fritz blindly followed. He "forgot" to tell Junior about it, who failed to >>understand what was expected of him, started playing elsewhere, and lost. >> >>The games between Junior, Fritz and Shredder were all sharp tactical draws. In >>Shredder-Fritz there were even some vaguely ridiculous fireworks just to force >>the obvious draw. Falcon-Shredder was similar, and probably the most spectacular >>game of the tournament, with some even more spectacular unplayed variations. The >>top programs now make you feel that going out of the opening they can see >>virtually till the end. Frightening thought. >> >>Junior and Shredder were obviously in top form for this championship. Shredder >>always gives the impression of near perfection, so it's very hard for me to >>judge whether it has made progress. > >I understood from Stefan that he did not have time to work much on the engine >recently so I guess that he probably has something that is only slightly better >than Shredder8. > working != making progress Amir > Fritz, on the other hand, has clearly not >>gone in the right direction. After Graz, where it played in superb consistent >>style, it seemed to have lost its footing to the wobbly performance that ruined >>some of its previous tournaments. >> >>Among the newcomers, the performance is mixed. Quite clearly, Jonny is the one >>who made the most progress. The promise of Sjeng, on the other hand, will have >>to wait for next year. Falcon has made progress, but lacks the consistency and >>solidity to convert it to results. >> >>Among the rest, Diep deserves mention. It had a good result, and there is >>obvious talent invested there, but the inconsistency that always characterized >>it was apparent in this tournament too. There is a FAQ in computer chess of >>whether it helps if the programmer is a good chess player. My answer (strictly >>IMO) is probably unexpected: It hurts. When you are a good player, you tend to >>think about the problem in all the wrong ways. Anyway one thing that is clear to >>me is that you have to be a very good programmer. No shortcuts there. >> >>Amir > >I agree that it is important to be a very good programmer(this is my >disadvantage and the code of today is badly written) but I do not think that >being a good player is a disadvantage. > >I think that it is important to test and the problem of some participants is >lack of testing. > >I talked with Gerd(programmer of Icichess) >Gerd is not a strong chess player. > >His program is even slower searcher than movei in nodes per second and >the programmer told me that most of the time is used on the evaluation. > >I asked him if he checked to test if what he added in the evaluation was >productive and the surprising reply was negative. > >I try to test terms that I add to the evaluation and unfortunately even things >that are cheap to add did not prove themselves to be productive. > >I have a code to evaluate backward pawns or knightoutposts and it is relatively >cheap to add it(no big difference in nodes per seconds) but none of them was >proved to be productive based on my tests games of the program against previous >version so I did not add them. > >I do not know if the reason is bugs in the evaluation or a different reason like >wrong definition of backward pawn or knight outposts(I try to test before >playing games by giving the new version some positions to analyze to see if the >evaluation is changed to the expected side) > >Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.