Author: Robert Hyatt
Date: 10:47:58 08/27/01
Go up one level in this thread
On August 27, 2001 at 12:19:46, Joshua Lee wrote:
>On August 27, 2001 at 04:46:06, Uri Blass wrote:
>
>>On August 27, 2001 at 04:33:31, Uri Blass wrote:
>>
>><snipped>
>>>for comparison Here is Deeper blue's logfile
>>>
>>>Guessing Nd2
>>> 7(4) #[O-O](10)##############################################[h5](18)# 18 T=1
>>>Ph7h5 qd1e2 Qd8d6 pa2a3 Bb4a5 o-o Ba5d2n bc1d2B
>>> 8(6) #[h5](-2)##[O-O](0)############################################ 0 T=11
>>>O-O o-o Bb4c5 nd2f3 Ne5f3n bg2f3N
>>> 9(6) #[O-O](-3)#########<ch> 'nd2'
>>>[153 sec (main.c:1849)][cont]###################################### -3 T=39
>>>O-O o-o Bb4d6 nd2b3 Nf6d5 ph3h4 Ph7h6
>>>10(6) #[O-O](-15)######[h5](-5)######################################### -5
>>>T=151
>>>Ph7h5 qd1e2 Ne5g6 pc2c3 Bb4e7 nd2e4 Ph5h4 bc1e3 Qd8a5 pg3g4
>>>11(6)[TIMEOUT] -5 T=178
>>>Ph7h5
>>>---------------------------------------
>>>--> 11. .. h5 <-- 29/101:33
>>>
>>>you can see that deeper blue needed 151 seconds to see 11...h5 and the
>>>main line of Junior at depth 16 is
>>>the same as deeper blue's in the first 4 plies.
>>
>>Junior7 agrees even on the first 5 plies after more time at depth 18
>>and here is the analysis at depthes 17,18
>>
>>1...h5 2.Qe2 Qc7 3.d4 Ng6 4.c3 Be7 5.Ne4 h4 6.Nxf6+ gxf6 7.gxh4 Rxh4 8.Be4
>> = (0.13) Depth: 17 00:05:55 184488kN
>>1...h5 2.Qe2 Ng6 3.c3 Be7 4.h4 Qc7 5.Nf3 0-0-0 6.Be3 Qa5
>> = (0.09) Depth: 18 00:17:08 529639kN
>>
>>(Blass, Tel-aviv 27.08.2001)
>>
>>Uri
>
>Uri I looked at the logs and Deep Blue saw h5 in it's search after White's 8th
>
> 10 T=34
>Ne5f3b qd1f3N Qd8b6 nb1c3 Ng8f6 ra1b1 Bf8c5 bc1f4 O-O nc3a4 Qb6b4
> 3(4) -2 T=0
>Qd8a5 nb1d2 Pd5e4p pd3e4P O-O-O
> 4(5) 7 T=0
>Pd5e4p bg2e4P Ng8f6 o-o Nf6e4b pd3e4N
>5(5)[dxe4](7) 7 T=0
>Pd5e4p bg2e4P Ng8f6 o-o Nf6e4b pd3e4N
> 6(5) [dxe4](10) 10 T=0
>Pd5e4p bg2e4P Ng8f6 o-o Nf6e4b pd3e4N
> 7(5) #[dxe4](26)######################################### 26 T=2
>Pd5e4p bg2e4P Ng8f6 o-o Nf6e4b pd3e4N Bf8d6
>8(6) #[dxe4](21)######################################### 21 T=4
>Pd5e4p bg2e4P Ng8f6 be4g2 Bf8b4 nb1c3 O-O
>9(6) #[dxe4](31)######################################### 31 T=13
>Pd5e4p bg2e4P Ng8f6 be4g2 Bf8c5 qd1e2 Qd8d4 bc1f4
>10(6) #[dxe4](20)######################################### 20 T=59
>Pd5e4p bg2e4P Ng8f6 be4g2 Bf8d6 o-o Qd8b6 nb1c3 O-O-O qd1d2
>
>11(6) #[dxe4](22)#[TIMEOUT] 22 T=174 Pd5e4p bg2e4P Ng8f6 be4g2 Bf8b4 nb1d2
>Ph7h5!!!!! qd1e2 Qd8d6 pa2a3 Bb4a5
> ---------------------------------------
>--> 8. .. dxe4 <-- 32/107:53
>---------------------------------------
>
>I am curious about a few things in the logs maybe you can explain.
>for 1 i noticed that when a move is going to be played it would be in the Pv
>however : In game 6
> 7. N1f3 <-- 33/111:6
>---------------------------------------
>Guessing Bd6
> 7(4) #[Qe2](29) 29 T=1
>qd1e2 Ph7h6 ng5e4 Nf6e4n qe2e4N
> 8(6) #[Qe2](29) 29 T=4
>qd1e2 Ph7h6 ng5e4 Nf6e4n bd3e4N Nd7f6 be4d3 Bc8d7 bc1e3
> 9(6) #[Qe2](29) 29 T=6
>qd1e2 Ph7h6 ng5e4 Nf6e4n bd3e4N Nd7f6 be4d3 Bc8d7 bc1e3
>10(6) #[Qe2](31) 31 T=40
>qd1e2 Ph7h6 ng5e4 Nf6e4n bd3e4N Qd8c7 bc1d2 Nd7f6 be4d3 Bc8d7
>
>11(6)<ch> 'h6'
>---------------------------------------
>--> Ph7h6 <--
>---------------------------------------
> 31 T=93
>qd1e2
>---------------------------------------
>--> 8. Nxe6 <-- 32/111:1
>---------------------------------------
>Guessing fxe6
> 3(4)[Bg6](-30) -30v T=0
>bd3g6 Ke8e7 bc1h6P Pg7h6b pc2c3
> 3(5) -54 T=0
>bd3g6 Ke8e7 bc1h6P Pg7h6b pc2c3
> 4(5) -24^ T=0
>bd3g6 Ke8e7 bc1h6P Pg7h6b pc2c3
>
>Notice the time given 50 seconds between 7.N1f3 111:6
> and 8 Nxe6 111:1
>The line before black's move was entered said 40 seconds
>10(6) #[Qe2](31) 31 T=40 <--------------
>qd1e2 Ph7h6 ng5e4 Nf6e4n bd3e4N Qd8c7 bc1d2 Nd7f6 be4d3 Bc8d7
>
>However after this
>11(6)<ch> 'h6'
>---------------------------------------
>--> Ph7h6 <--
>---------------------------------------
> 31 T=93
>qd1e2
>Where is this extra time comming from and why doesn't it reflect this on move
>8?? I can't say that IBM cheated but things like this do not look really good.
>If this single move was in book how come Deep Blue didn't search it like it did
>other book moves? or am i wrong about this?
>
>Also i had looked at Game 2 why wouldn't Deep Blue find Qe3 in it's search after
>about 10 seconds???? It didn't even show up on the previous move.
I don't think you can draw _serious_ conclusions from the logs, if you look
at that tiny type of detail. IE in a parallel search, there is lots of room
for strange things. I have seen the following sort of output from Crafty for
very shallow searches:
move XXX score +1.31 (crafty is searching for white)
Move YYY score +1.04 (same search, notice a slightly worse move comes
out after a better move. If you had checked the
time, you would notice the second move has a slightly
lower time value than the first, too, which should
be impossible).
This happens due to the way I "lock" things. Crafty will _always_ play the
move with the best score. But when it prints them (it searches root moves
in parallel too) it is possible that the worst move is printed second because
of acquiring a semaphore lock. I haven't tried to solve this because it is
only an issue in very rare cases, and the program plays the correct move in
every case, but the log output will look just a bit strange.
For that reason, paying too much attention to what is going on in a parallel
search is often misleading. There are good race conditions and bad race
conditions. Bad ones cause the wrong move to be played, and can hurt the
programs performance. Good race conditions have no harmful effects at all and
can be ignored, as fixing them might cause an execution penalty that is not
worth the cost.
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.