# Computer Chess Club Archives

## Messages

### Subject: Re: I'm being too harsh, but still

Author: Uri Blass

Date: 01:46:48 12/18/02

Go up one level in this thread

```On December 18, 2002 at 04:20:05, Bruce Moreland wrote:

>On December 18, 2002 at 03:53:48, Dave Gomboc wrote:
>
>>My copy of the issue is at home, but my recollection is that he claimed
>>superiority over R=2, but did not claim superiority over R=3.  (Or if he did,
>>then this was based on experiments by others [notably Heinz] who already
>>demonstrated the superiority of R=2 over R=3.  I agree that the data presented
>>does not justify that claim without additional information.)
>
>I think that what is going on here is that Omid assumed that he could take the
>superiority of R=2 as a given, and build on that.
>
>The idea is that if R=3 is considered to be worse than R=2, if you use a variant
>of R=3 and it's better than R=2, it must also be better than R=3.
>
>The data in Omid's paper supports the conclusion that R=3 is better than R=2.
>
>On page 159, the number of WCS positions solved in 10 plies with R=2 is 850.
>With R=3 it's 849.  R=3 is more than twice as fast.  If R=3 would have been
>allowed to run longer than R=2, it would have found more positions, and
>surpassed R=2.
>
>You can draw the same conclusion from the table on page 157.  66 found with R=2,
>65 found with R=3, in half the time.
>
>I tested Gerbil with R=2 and R=3, on ECM, and Gerbil gets more solutions with
>R=2, for any time value between 1 and 20 seconds (assuming a correct answer
>holds to 20 seconds).  So for Gerbil, doing ECM pretty fast, R=2 is better.
>
>I have not tested Omid's variant.
>
>bruce

R=2 is better for Gerbil at fast time control.
It may be interesting to know if it is also better at slower time control
because I remember that omid claimed few months ago that based on his experience
R=3 is better than R=2 at long time control(he tested it in games and it was
without verification search).

It may be interesting to know if R=2 continue to solve more positions also at
180 seconds per position(of course this test may take a long time but you can
get first impression but solving only part of the ECM suite with R=3 and R=2 at
180 seconds per position).

Uri

```