Author: Christophe Theron
Date: 06:49:03 10/20/99
Go up one level in this thread
On October 20, 1999 at 05:53:42, Enrique Irazoqui wrote:
>On October 20, 1999 at 00:29:23, Christophe Theron wrote:
>
>>On October 19, 1999 at 18:13:29, Enrique Irazoqui wrote:
>>
>>>On October 19, 1999 at 13:12:23, Christophe Theron wrote:
>>>
>>(snip)
>>>
>>>>I don't want to find any excuse if CM7000 beats Tiger.
>>>>
>>>>However I don't understand exactly why the discussions are taking place NOW.
>>>
>>>:)
>>>
>>>Because NOW we received Tiger 12.0 and test it. It is not that I am jealously
>>>protecting the honor of Tiger, but I am simply trying to figure out why our
>>>findings are different.
>>>
>>>Enrique
>>
>>
>>Maybe you could just mix the results of your tests to get closer to the truth.
>>
>>I run automatic tests against a given program here. When I launch a test, I
>>launch it on 4 computers simultaneously. Each computer runs a complete match
>>(both opponents are on the same computer). Here is a result I had recently (Oct
>>13, 1999):
>>
>>Computer A: 59.4% (32 games)
>>Computer B: 48.4% (32 games)
>>Computer C: 65.0% (20 games)
>>Computer D: 50.0% (16 games)
>>
>>Overall result: 55.5%
>>
>>If you look only at the result of Computer B and Computer C, you can say: "OK,
>>there is a problem, let's find WHY".
>>
>>But remember that on B and on C you have exactly the same opponents with the
>>same settings. Actually there is absolutely NO problem!!!
>>
>>Now mix together the results of B and C: you get 54.8% (don't forget the
>>weights: I do (48.4*32+65*20)/(32+20)), which is more reasonnable and quite
>>close to the final result.
>>
>>So even this 65% result I got with 20 games is rather far from what I get with
>>100 games (in the end I get 10% less). And the 4.5-2.5 result Didzis has
>>currently is close to 65%, isn't it?
>
>Sure, Christophe, that's something I have done and posted a few times. But this
>is not what I was referring to. After the games I saw, I had trouble imagining
>Tiger losing one single game to the King, let alone 3. That's why I was
>interested in the testing conditions. Another tester of Tiger told me the
>following: "it´s nearly unpossible that cm beats tiger. in 10 games the cm
>didn´t have the chance even to tickle tiger. only one game tiger lost due to a
>book-blunder. 6 games tiger won and the rest 3 games were a lucky draw for cm
>with a lot of advantage for tiger." This corresponds perfectly with my own
>findings. So it was not a matter of statistics but of looking at the games one
>by one and wondering how Tiger could have lost. Theoretically it may happen, of
>course. It even happened in Didzis games. I just wondered about this sort of
>anti-miracle in this specific case.
>
>Enrique
Maybe Mindscape has taken CM5555 parameters and included them in CM7000? The
engine is the same but maybe it behaves differently?
I suppose it should be possible to check this?
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.