Andscacs - New version 0.84

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

User avatar
cdani
Posts: 2204
Joined: Sat Jan 18, 2014 10:24 am
Location: Andorra

Andscacs - New version 0.84

Post by cdani »

Hi!
I published a new version, 0.84.

Changes:
* Solved little bugs.
* Improved evaluation of pawns.
* Improved reductions.
* Some small speed improvements.
* Slight improvements on king safety.
* General reevaluation of most parameters.

http://andscacs.com/andscacs084.zip

Should be clearly better thanks to first time tuning with Texel method, probably more than 30 elo, if it holds at long time control, of course.
User avatar
Ozymandias
Posts: 1537
Joined: Sun Oct 25, 2009 2:30 am

Re: Andscacs - New version 0.84

Post by Ozymandias »

cdani wrote:Should be clearly better thanks to first time tuning with Texel method, probably more than 30 elo, if it holds at long time control, of course.
If it does, that'd be close to +50, over 0.82, in about… two months? :shock:
mar
Posts: 2685
Joined: Fri Nov 26, 2010 2:00 pm
Location: Czech Republic
Full name: Martin Sedlak

Re: Andscacs - New version 0.84

Post by mar »

Ozymandias wrote:If it does, that'd be close to +50, over 0.82, in about… two months? :shock:
Why not? IIRC I got 70 elo in 2 weeks using Texel method, of course you need untuned eval and you can only jump this much once.
User avatar
Graham Banks
Posts: 45675
Joined: Sun Feb 26, 2006 10:52 am
Location: Auckland, NZ

Re: Andscacs - New version 0.84

Post by Graham Banks »

cdani wrote:Should be clearly better thanks to first time tuning with Texel method, probably more than 30 elo, if it holds at long time control, of course.
Good that you're still making progress.
gbanksnz at gmail.com
IWB
Posts: 1539
Joined: Thu Mar 09, 2006 2:02 pm

Re: Andscacs - New version 0.84

Post by IWB »

Hello Daniel, hello all,

This version, as previous ones is not pondering after a ponderhit - at least not in the Shredder Classic GUI (Which I consider the "mother" of UCI :-) )

What happens is
1. Andscacs Ponders something
2. Opponent answers with the right ponder move (Ponderhit)
3. Andscacs answers immediately
4. Andscacs doesn't ponder

Can someone confirm that for the Shredder Classic GUI. I was told that it might work in Arena but neither do I have tha tinstalled (or any experience with it) nor do I know how UCI is implemented there (And I would consider the Classic as the "UCI Master" here).

Don't get this wrong, the engine is playing perfectly legal chess, but it is indirectly loosing time ... and therefore Elo!

Regards
Ingo
User avatar
Graham Banks
Posts: 45675
Joined: Sun Feb 26, 2006 10:52 am
Location: Auckland, NZ

Re: Andscacs - New version 0.84

Post by Graham Banks »

IWB wrote:Hello Daniel, hello all,

This version, as previous ones is not pondering after a ponderhit - at least not in the Shredder Classic GUI (Which I consider the "mother" of UCI :-) )

What happens is
1. Andscacs Ponders something
2. Opponent answers with the right ponder move (Ponderhit)
3. Andscacs answers immediately
4. Andscacs doesn't ponder

Can someone confirm that for the Shredder Classic GUI. I was told that it might work in Arena but neither do I have tha tinstalled (or any experience with it) nor do I know how UCI is implemented there (And I would consider the Classic as the "UCI Master" here).

Don't get this wrong, the engine is playing perfectly legal chess, but it is indirectly loosing time ... and therefore Elo!

Regards
Ingo
Do you mean it no longer ponders on subsequent moves after the instant response?

Many engines reply immediately even with ponder off due to the use of "easy move".
gbanksnz at gmail.com
IWB
Posts: 1539
Joined: Thu Mar 09, 2006 2:02 pm

Re: Andscacs - New version 0.84

Post by IWB »

Graham Banks wrote: Do you mean it no longer ponders on subsequent moves after the instant response?
Yes.
Graham Banks wrote: Many engines reply immediately even with ponder off due to the use of "easy move".
Yes, and after that "easy move" there is no ponder anymore for Andscacs! It waits idle for the next move to come ... (Looks lik a POFF game, and continues to ponder after the next move until ...)
NO other engine in my TOP 16 behaves like this - because they would lose time (or better depth) with a possible new ponderhit.


(Engines do reply immediately with Ponder off? Really? How, all out of hash? I am suprised, but have to admit I have no idea about POFF games - and don't mind much)

Regards
Ingo
User avatar
Graham Banks
Posts: 45675
Joined: Sun Feb 26, 2006 10:52 am
Location: Auckland, NZ

Re: Andscacs - New version 0.84

Post by Graham Banks »

IWB wrote:......(Engines do reply immediately with Ponder off? Really? How, all out of hash? I am suprised, but have to admit I have no idea about POFF games - and don't mind much)

Regards
Ingo
Yes, they do at times.

I'd imagine that if the expected response is played and their previous thinking had decided that there was one move much higher in its evaluation than others, it gets played instantly.

A programmer might be able to clarify.
gbanksnz at gmail.com
Ferdy
Posts: 4853
Joined: Sun Aug 10, 2008 3:15 pm
Location: Philippines

Re: Andscacs - New version 0.84

Post by Ferdy »

IWB wrote:Hello Daniel, hello all,

This version, as previous ones is not pondering after a ponderhit - at least not in the Shredder Classic GUI (Which I consider the "mother" of UCI :-) )

What happens is
1. Andscacs Ponders something
2. Opponent answers with the right ponder move (Ponderhit)
3. Andscacs answers immediately

4. Andscacs doesn't ponder
After receiving the ponderhit from gui, the engine would switch to normal search (not ponder search). It can send its move immediately if it wants to or extend its seach a little longer to get a good guess of opp next move.
In this case the engine should send something like,

Code: Select all

bestmove e2e4 ponder e7e5
If Andscacs is only sending bestmove e2e4 (after ponderhit command from gui), then there could be a problem. Did it do it like that always? It is possible that it failed to send its ponder move on that particular position only or on some other positions.

After sending the bestmove e2e4 ponder e7e5, the engine will wait for

Code: Select all

position ...
go ponder ...
from gui, before it will start to search again.
IWB
Posts: 1539
Joined: Thu Mar 09, 2006 2:02 pm

Re: Andscacs - New version 0.84

Post by IWB »

Graham Banks wrote:
Yes, they do at times.

I'd imagine that if the expected response is played and their previous thinking had decided that there was one move much higher in its evaluation than others, it gets played instantly.

A programmer might be able to clarify.
So, out of hash.

Anyhow, it is different here as this is a Ponder ON game.

Just as a rough estimation:
Assuming only 50% Ponder hit rate with imediate answer, every second of that is having no ponder at all. If the doubeling of time is worth only 50 Elo I would guess that Andscacs is loosing about 10-15 Eo here ... and that is a conservative guess.

And even with that it is most likely allready that good that it can beat any human :-D


Regards
Ingo