Author: Christophe Theron
Date: 14:56:15 04/23/01
Go up one level in this thread
On April 23, 2001 at 17:26:52, Uri Blass wrote:
>On April 23, 2001 at 15:39:22, Christophe Theron wrote:
>
>>On April 23, 2001 at 15:00:23, Uri Blass wrote:
>>
>>>On April 23, 2001 at 13:46:02, Christophe Theron wrote:
>>>
>>>>On April 23, 2001 at 02:43:23, Uri Blass wrote:
>>>>
>>>>>On April 23, 2001 at 02:17:26, Tanya Deborah wrote:
>>>>>
>>>>>>On April 22, 2001 at 12:29:34, Uri Blass wrote:
>>>>>>
>>>>>>>Tiger14(depth 3)-Deep Fritz(depth 3) 16.5-3.5
>>>>>>>Tiger14(depth 4)-Deep Fritz(depth 4) 7.5-12.5
>>>>>>>Tiger14(depth 5)-Deep Fritz(depth 5) 13-7
>>>>>>>Tiger14(depth 6)-Deep Fritz(depth 6) 14-6
>>>>>>>Tiger14(depth 7)-Deep Fritz(depth 7) 13.5-6.5
>>>>>>>Tiger14(depth 8)-Deep Fritz(depth 8) 10.5-9.5
>>>>>>>Tiger14(depth 9)-Deep Fritz(depth 9) 14-6
>>>>>>>Tiger14(depth 10)-Deep Fritz(depth 10) 9.5-10.5
>>>>>>>
>>>>>>>Tiger14(depth 4)-Deep Fritz(depth 3) 17.5-2.5
>>>>>>>Tiger14(depth 5)-Deep Fritz(depth 4) 19-1
>>>>>>>Tiger14(depth 6)-Deep Fritz(depth 5) 18-2
>>>>>>>Tiger14(depth 7)-Deep Fritz(depth 6) 15-5
>>>>>>>
>>>>>>>Tiger14(depth 3)-Deep Fritz(depth 4) 6-14
>>>>>>>
>>>>>>>Tiger14(depth 4)-Deep Fritz(depth 5) 8-12(this result should be 12.5-7.5 but
>>>>>>>deep fritz failed to mate because of having only part of the tablebases and the
>>>>>>>engines agreed to a draw in a position that the evaluation was mate against
>>>>>>>tiger).
>>>>>>>
>>>>>>>Tiger14(depth 5)-Deep Fritz(depth 6) 7-13
>>>>>>>Tiger14(depth 6)-Deep Fritz(depth 7) 8-12
>>>>>>>Tiger14(depth 7)-Deep Fritz(depth 8) 7-13
>>>>>>>
>>>>>>>The advantage of one ply seems to be very significant in the nunn match and you
>>>>>>>can see that the side that searched deeper won in all the matches.
>>>>>>>
>>>>>>>Deep Fritz at depth X seems to be 100 about elo better than Tiger at depth X-1
>>>>>>>when Tiger14 at depth X seems to be more than 200 elo better than Tiger14.
>>>>>>>
>>>>>>>It means that at small depthes 1 ply is probably worth at least 150 elo rating
>>>>>>>for the nunn match.
>>>>>>>There is of course a significant statistical error and we need more games.
>>>>>>>
>>>>>>>Uri
>>>>>>
>>>>>>Very interesting Uri!
>>>>>
>>>>>Here are more results
>>>>>Tiger14(depth 8)-Deep Fritz(depth 7) 14-6
>>>>>Deep Fritz(depth 9)-Tiger14(depth 8) 12.5-7.5
>>>>>Tiger14(depth 9)-Deep Fritz(depth 8) 8.5-1.5(this match was not finished because
>>>>>I stopped it in the middle)
>>>>>
>>>>>>
>>>>>>But how did you make that Fritz move when it is on depth4 or 5 ??
>>>>>>I believe that many of this games (at depth 3,4,or5) take only a few seconds.
>>>>>>Is this right?
>>>>>
>>>>>Correct games at these depthes takes only few seconds so I could finish the nunn
>>>>>match at small depthes in few minutes.
>>>>>
>>>>>I have some more results at small depthes(this time of Gambittiger2 against Deep
>>>>>Fritz
>>>>>
>>>>>Gambit2(depth 3)-Deep Fritz(depth 3) 8.5-11.5
>>>>>Gambit2(depth 4)-Deep Fritz(depth 4) 10.5-9.5
>>>>>Gambit2(depth 5)-Deep Fritz(depth 5) 13-7
>>>>>Gambit2(depth 6)-Deep Fritz(depth 6) 12-8
>>>>>
>>>>>Gambit2(depth 3)-Deep Fritz(depth 4) 4.5-15.5
>>>>>Gambit2(depth 4)-Deep Fritz(depth 3) 16.5-3.5
>>>>>Gambit2(depth 4)-Deep Fritz(depth 5) 9.5-10.5
>>>>>Gambit2(depth 5)-Deep Fritz(depth 4) 17-3
>>>>>
>>>>>You can see that the Deeper searcher at shallow depthes is always winning the
>>>>>match if you use tiger and fritz.
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>>
>>>>>>If the program using a 8x procesor machine against Kramnink, see one ply more
>>>>>>than the normal fast machine, how strong will be ?? in elo??
>>>>>
>>>>>I do not know.
>>>>>
>>>>>I believe in diminishing return and the results seem to suggest diminishing
>>>>>return but I need more games to prove it stattistically.
>>>>>
>>>>>Uri
>>>>>>
>>>>>>Tanya.
>>>>
>>>>
>>>>
>>>>Go on Uri, that's interesting.
>>>>
>>>>
>>>>
>>>> Christophe
>>>
>>>I will continue.
>>>
>>>Unfortunately I am not going to give more than few hours per day of computer
>>>time for it because I use my computer for other purposes.
>>>
>>>No news from the last update.
>>>
>>>I think to include
>>>Tiger14,Gambit2,Deep Fritz, Crafty at depth 3-10
>>>
>>>I think to play only matches between different programs when the difference
>>>in depth is at most 3 and it is going to give for every player:
>>>420 games for depth 6,7
>>>360 games for depth 5,8
>>>300 games for depth 4,9
>>>240 games for depth 3,10
>>>
>>>This process may take some months because of the fact that I am not going to use
>>>more than few hours per day and I do not promise to use the computer every day
>>>for this purpose.
>>>
>>>I am afraid games when the difference in depth is more than 3 are going to end
>>>often with the result 20-0 and this is the reason that I am not going to play
>>>them(we already had 19-1 result when the difference was only 1).
>>>
>>>Other people probably can reproduce similiar results but not exactly the same
>>>results because they probably do not have exactly the same set of tablebases
>>>that I have.
>>>
>>>Uri
>>
>>
>>
>>I think that a difference of 1 and 2 plies is more than enough to give
>>significant results. You can save the additional work with 3 plies differences.
>>
>>I suggest that you do not use tablebases. TB are the same thing as an infinite
>>search, so it breaks your rule of fixed depths.
>
>I understand
>Unfortunately I used tablebases in the first games so I need to repeat the games
>that I did in order to be consistant.
>Fortunately I did not play most of the games that are planned so I do not need
>to repeat most of them.
>
>I guess there is not going to be a big difference but I am going to do it.
>I expect that not using tablebases is going to help the deeper searchers.
>
>Uri
Well... thinking again about this, I think I'm not going to insist too much on
disabling TBs.
TBs can be seen as infinite searches, but they can also be seen as much better
knowledge (in the evaluation function). Perfect knowledge or perfect search?
They are not going to have a big influence on the result anyway, so you'd
probably better keep on with your current settings.
Christophe
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.