Winboard Forum

by **Tord Romstad** » 17 Mar 2004, 12:15

Geschrieben von: / Posted by: Tord Romstad at 17 March 2004 12:15:06:

I have now released Mac OS X, Linux and Windows binaries of Gothmog 0.4.7.
There are a number of small bug fixes, and some playing strength improvements.
I am not sure how this version compares to the short-lived and not very
thoroughly tested version 0.4.6, but I think that 0.4.7 is about 50 rating
points stronger than 0.4.5.
Many thanks to Fabien Letouzey, who has reported several bugs.
The details:

A number of minor bugs in the xboard and UCI protocol support have been
fixed in order to make Gothmog run more smoothly in Fabien's UCI to xboard
adapter (still unreleased).
Another bug reported by Fabien: Gothmog was hanging at startup if the
gothmog.ini file contained empty lines at the end. This is now fixed.
Version 0.4.6 introduced a new selective search bug, which would often
cause too many moves to be pruned. The problem is solved in 0.4.7.
Some evaluation weights have been tuned.
The meaning of the "selectivity" setting in the init file (the parameter
is also accessible through the UCI interface) has changed. The exact meaning
of the "selectivity" parameter is difficult to explain, but higher numbers
make the search more selective. I have absolutely no idea whether the default
value of 4 is the best. Those who want to experiment with other values are
welcome to do so.

Tord

by **Kurt Utzinger** » 17 Mar 2004, 12:35

Geschrieben von: / Posted by: Kurt Utzinger at 17 March 2004 12:35:18:
Als Antwort auf:/In reply to: Gothmog 0.4.7 geschrieben von: / posted by: Tord Romstad at 17 March 2004 12:15:06:

Hi Tord
Many thanks for this new version I am going to download right away.
Kurt

by **Sune Fischer** » 17 Mar 2004, 12:48

Geschrieben von: / Posted by: Sune Fischer at 17 March 2004 12:48:17:
Als Antwort auf:/In reply to: Gothmog 0.4.7 geschrieben von: / posted by: Tord Romstad at 17 March 2004 12:15:06:

but I think that 0.4.7 is about 50 rating points stronger than 0.4.5.

Argh!
Regards,
-Sune (who keeps wondering when Tord hits the wall)

by **Mogens Larsen** » 17 Mar 2004, 12:54

Geschrieben von: / Posted by: Mogens Larsen at 17 March 2004 12:54:56:
Als Antwort auf:/In reply to: Gothmog 0.4.7 geschrieben von: / posted by: Tord Romstad at 17 March 2004 12:15:06:

I have now released Mac OS X, Linux and Windows binaries of Gothmog 0.4.7.
There are a number of small bug fixes, and some playing strength improvements.
I am not sure how this version compares to the short-lived and not very
thoroughly tested version 0.4.6, but I think that 0.4.7 is about 50 rating
points stronger than 0.4.5.

Hey Tord,
Any news on the (lack of) insufficient material claim?
There's a minor mistake in your tbook.pgn.
1. e4 c6? 2. d4! d5 3. cxd5 exd5 4. c4 *
Should be
1. e4 c6? 2. d4! d5 3. exd5 cxd5 4. c4 *
Regards,
Mogens

by **Kurt Utzinger** » 17 Mar 2004, 13:02

Geschrieben von: / Posted by: Kurt Utzinger at 17 March 2004 13:02:13:
Als Antwort auf:/In reply to: Re: Gothmog 0.4.7 geschrieben von: / posted by: Mogens Larsen at 17 March 2004 12:54:56:

I have now released Mac OS X, Linux and Windows binaries of Gothmog 0.4.7.
There are a number of small bug fixes, and some playing strength improvements.
I am not sure how this version compares to the short-lived and not very
thoroughly tested version 0.4.6, but I think that 0.4.7 is about 50 rating
points stronger than 0.4.5.
Hey Tord,
Any news on the (lack of) insufficient material claim?
There's a minor mistake in your tbook.pgn.
1. e4 c6? 2. d4! d5 3. cxd5 exd5 4. c4 *
Should be
1. e4 c6? 2. d4! d5 3. exd5 cxd5 4. c4 *
Regards,
Mogens

Hi Mogens
Thank you for this hint. I have changed the book.
Kurt

by **Uri Blass** » 17 Mar 2004, 13:05

Geschrieben von: / Posted by: Uri Blass at 17 March 2004 13:05:15:
Als Antwort auf:/In reply to: Gothmog 0.4.7 geschrieben von: / posted by: Tord Romstad at 17 March 2004 12:15:06:

I have now released Mac OS X, Linux and Windows binaries of Gothmog 0.4.7.
There are a number of small bug fixes, and some playing strength improvements.
I am not sure how this version compares to the short-lived and not very
thoroughly tested version 0.4.6, but I think that 0.4.7 is about 50 rating
points stronger than 0.4.5.

You progress too fast.
It seems that Gothmog is going to be the winner of the second division of Leo.
When you work on advanced ideas I still need to work on some basic stuff because I did a mistake when I started by copying part of my data structure from tscp.
For example I decided that I need to change moves to be integers and not ugly union of struct of 4 chars and integer for having code that is more easy to read(maybe it may help also in speed but the main reason that I do it is not speed and I do not like to read expressions with .m.b.from and defines are also not clear because in my code there are cases when I want to get the from square based on m and there are cases when I want to get the from square based on b.
Uri

by **Tord Romstad** » 17 Mar 2004, 13:15

Geschrieben von: / Posted by: Tord Romstad at 17 March 2004 13:15:09:
Als Antwort auf:/In reply to: Re: Gothmog 0.4.7 geschrieben von: / posted by: Sune Fischer at 17 March 2004 12:48:17:

but I think that 0.4.7 is about 50 rating points stronger than 0.4.5.
Argh!
Regards,
-Sune (who keeps wondering when Tord hits the wall)

As Marcus Prewarski pointed out yesterday: The key to rapid improvement is to
start with a lot of bugs. :-)

Tord

by **Tord Romstad** » 17 Mar 2004, 13:29

Geschrieben von: / Posted by: Tord Romstad at 17 March 2004 13:29:40:
Als Antwort auf:/In reply to: Re: Gothmog 0.4.7 geschrieben von: / posted by: Uri Blass at 17 March 2004 13:05:15:

I have now released Mac OS X, Linux and Windows binaries of Gothmog 0.4.7.
There are a number of small bug fixes, and some playing strength improvements.
I am not sure how this version compares to the short-lived and not very
thoroughly tested version 0.4.6, but I think that 0.4.7 is about 50 rating
points stronger than 0.4.5.
You progress too fast.
It seems that Gothmog is going to be the winner of the second division of Leo.
When you work on advanced ideas I still need to work on some basic stuff because I did a mistake when I started by copying part of my data structure from tscp.
For example I decided that I need to change moves to be integers and not ugly union of struct of 4 chars and integer for having code that is more easy to read(maybe it may help also in speed but the main reason that I do it is not speed and I do not like to read expressions with .m.b.from and defines are also not clear because in my code there are cases when I want to get the from square based on m and there are cases when I want to get the from square based on b.

Don't be too sure that my estimate of a 50 points improvement is accurate. I
have sometimes been too optimistic in the past.
I think the second division will be a very close race between several engines.
There are several engines with very similar playing strength, and I don't
see any clear favorite. Given the limited number of rounds, the winner will
need a certain amount of luck. I think your engine has at least as good
chances of winning as mine.
Actually I haven't implemented any new advanced ideas for quite a long time.
I played with some new ideas around Christmas, but I was not satisfied with
the results, and the code is currently disabled.
In fact, the current version of Gothmog hasn't changed much compared to
version 0.4.0 (released in October last year). The only important changes
have been lots of bug fixes, some modifications to search and eval parameters,
and a few pieces of new endgame knowledge.
Like you, I constantly struggle with old, stupid design decisions from the
time when my program was young. Some day I will have to rewrite it all
from scratch.
I also made the mistake of representing moves by structs (8-byte structs,
in my case) rather than integers. This is one of the things I intend to
change when I do a major rewrite.
Tord

by **Sune Fischer** » 17 Mar 2004, 13:56

Geschrieben von: / Posted by: Sune Fischer at 17 March 2004 13:56:36:
Als Antwort auf:/In reply to: Re: Gothmog 0.4.7 geschrieben von: / posted by: Tord Romstad at 17 March 2004 13:15:09:

but I think that 0.4.7 is about 50 rating points stronger than 0.4.5.
Argh!
Regards,
-Sune (who keeps wondering when Tord hits the wall)
As Marcus Prewarski pointed out yesterday: The key to rapid improvement is to
start with a lot of bugs.
Tord

Yeah, I remember back when fixing a bug got me 30 Elo,
those were the days....

I recall one incident quite clearly.
I was watching a Bruce Lee movie while chatting with the author of Oswald, I think it was on FICS.
We started talking about our qsearches and I explained that I culled at depth 3.
He suggested to me that I tried not doing that, I tried it his way and bang
50 Elo just like that

That's the kind of rush we are all looking for I guess, the one line change that will give +50.
I'm convinced they are out there, and given the rapid improvement of some
engines I think they are also known by some

-S.

by **Fabien Letouzey** » 17 Mar 2004, 14:00

Geschrieben von: / Posted by: Fabien Letouzey at 17 March 2004 14:00:05:
Als Antwort auf:/In reply to: Re: Gothmog 0.4.7 geschrieben von: / posted by: Sune Fischer at 17 March 2004 13:56:36:

That's the kind of rush we are all looking for I guess, the one line change that will give +50.

I believe that the line of code does not matter.
It's the years of study necessary to find it that count.
Fabien.

by **Uri Blass** » 17 Mar 2004, 14:03

Geschrieben von: / Posted by: Uri Blass at 17 March 2004 14:03:43:
Als Antwort auf:/In reply to: Re: Gothmog 0.4.7 geschrieben von: / posted by: Tord Romstad at 17 March 2004 13:15:09:

but I think that 0.4.7 is about 50 rating points stronger than 0.4.5.
Argh!
Regards,
-Sune (who keeps wondering when Tord hits the wall)
As Marcus Prewarski pointed out yesterday: The key to rapid improvement is to
start with a lot of bugs.
Tord

It is not enough.
You need also to fix your bugs.
There are programmers with a lot of bugs who did not get big improvement because they probably did not fix their bugs(see the cases of Rdchess or Biglion).
Uri

by **Tord Romstad** » 17 Mar 2004, 14:05

Geschrieben von: / Posted by: Tord Romstad at 17 March 2004 14:05:14:
Als Antwort auf:/In reply to: Re: Gothmog 0.4.7 geschrieben von: / posted by: Sune Fischer at 17 March 2004 13:56:36:

That's the kind of rush we are all looking for I guess, the one line change that will give +50.

Yes. Like the one I found in January this year, when I discovered that my
evaluation function included a bonus for NOT being able to safely advance
a passed pawn or backward pawn. :-)

Tord

by **Tord Romstad** » 17 Mar 2004, 16:13

Geschrieben von: / Posted by: Tord Romstad at 17 March 2004 16:13:24:
Als Antwort auf:/In reply to: Re: Gothmog 0.4.7 geschrieben von: / posted by: Mogens Larsen at 17 March 2004 12:54:56:

I have now released Mac OS X, Linux and Windows binaries of Gothmog 0.4.7.
There are a number of small bug fixes, and some playing strength improvements.
I am not sure how this version compares to the short-lived and not very
thoroughly tested version 0.4.6, but I think that 0.4.7 is about 50 rating
points stronger than 0.4.5.
Hey Tord,
Any news on the (lack of) insufficient material claim?
There's a minor mistake in your tbook.pgn.
1. e4 c6? 2. d4! d5 3. cxd5 exd5 4. c4 *
Should be
1. e4 c6? 2. d4! d5 3. exd5 cxd5 4. c4 *

Thanks, Mogens!
I will prepare new tbook.bin and tbook.pgn files for download later today.
I haven't found the time to look at the missing draw claim yet. :-(

Tord

Winboard Forum

Who is online