Latest Updates:
Normal Topic Engines & Openings (Read 645 times)
Bibs
God Member
*****
Offline



Posts: 2074
Joined: 10/24/06
Re: Engines & Openings
Reply #8 - 08/27/18 at 22:03:33
Post Tools
Ah, thank you Brabo.
I don’t read this bit of the board much, so didn’t notice. I’ll read that part now.

Thank you for the direction and for the ‘not unless free or +100’ note of your own approach.

Thanks!
  
Back to top
 
IP Logged
 
brabo
God Member
*****
Offline


Welcome chessfriend

Posts: 856
Joined: 02/02/07
Re: Engines & Openings
Reply #7 - 08/27/18 at 17:30:59
Post Tools
Bibs wrote on 08/27/18 at 12:59:14:
Hi all,

I'm wondering...

I have Komodo 10, which seems pretty good. I am pondering  upgrading to 12 and/or getting another software.

PURPOSE
For the first time, I am starting to study properly, finally. Now making opening databases, checking with engines. Building, building, studying.

Career settled and no probs, kids a bit older now and a bit less hassle, so time to improve my chess and get (back?) to a decent level. I am a low-rated FM at 2160-something next list. Noted, as level of OP helps re: advice I have found.

QUESTION
1) Is it worth upgrading K 10 to 12? Is there a noticeable difference? I do not play corr. at all.
This is just for analysis.

2) Would getting a different engine offer some variety of input? Do engines differ much in style and evaluations differences? I have used engines just a bit, but have no real knowledge of them, and I realise some here know lots.

Thank you if anyone may be able to offer advice! Much obliged Smiley


We already discussed this here a couple of months ago see http://www.chesspub.com/cgi-bin/chess/YaBB.pl?num=1327414061/151#151

The difference between Komodo 10 and 12 is 52 elo see http://www.computerchess.org.uk/ccrl/4040/cgi/compare_engines.cgi?family=Komodo&...
Personally I only upgrade an engine if it is free or I get a gain of 100 elo.
  
Back to top
WWW  
IP Logged
 
Bibs
God Member
*****
Offline



Posts: 2074
Joined: 10/24/06
Re: Engines & Openings
Reply #6 - 08/27/18 at 12:59:14
Post Tools
Hi all,

I'm wondering...

I have Komodo 10, which seems pretty good. I am pondering  upgrading to 12 and/or getting another software.

PURPOSE
For the first time, I am starting to study properly, finally. Now making opening databases, checking with engines. Building, building, studying.

Career settled and no probs, kids a bit older now and a bit less hassle, so time to improve my chess and get (back?) to a decent level. I am a low-rated FM at 2160-something next list. Noted, as level of OP helps re: advice I have found.

QUESTION
1) Is it worth upgrading K 10 to 12? Is there a noticeable difference? I do not play corr. at all.
This is just for analysis.

2) Would getting a different engine offer some variety of input? Do engines differ much in style and evaluations differences? I have used engines just a bit, but have no real knowledge of them, and I realise some here know lots.

Thank you if anyone may be able to offer advice! Much obliged Smiley

  
Back to top
 
IP Logged
 
Uhohspaghettio
God Member
*****
Offline


I Love ChessPublishing!

Posts: 510
Joined: 02/23/11
Re: Engines & Openings
Reply #5 - 08/26/18 at 21:49:51
Post Tools
ReneDescartes wrote on 08/26/18 at 16:41:50:
Uhohspaghettio wrote on 08/24/18 at 20:21:33:
0.8 is still a draw with perfect play though. I really think it's just the idiosyncratic way Stockfish evaluates the positions - the developers would change it if there were a clear flaw in its opening evaluation


I don't understand this. With perfect play, as in a tablebase, 0.8 is meaningless. There are no probabilities, only facts--win, loss, draw.


I agree but if any engine is showing 0.8 in the opening, it's almost certain that with perfect play by both sides the result will be a draw. Despite having higher numbers Stockfish isn't necessarily claiming white wins the position any more than Houdini is claiming it. 

ReneDescartes wrote on 08/26/18 at 16:41:50:
Uhohspaghettio wrote on 08/24/18 at 20:21:33:
At the end of the day the value of a pawn in chess is a completely abstract concept that is barely meaningful.


Well, you can't have a pawn (or other material imbalance) without having a position, and there are always features in a position other than the material. Engine evaluation functions are complicated affairs that only roughly use our notions of space or activity, etc.: you can't easily separate out factors. Even measuring space is complicated and can be done in a variety of ways. And the reported evaluation is of a position found by the search functions, which vary just as heavily.

If you want to see how engine evaluations converge, look at the graphs in TCEC games. The evaluations only sometimes (temporarily) converge on something like +1, even when one includes a multiplier for Stockfish to scale down its evaluations. Eventually, of course, evaluations usually converge--on very high positive numbers, very low negative numbers, or zero.

So our current engine evaluations are really a practical matter, for correspondence players, opening analysts, and programmers--not a theoretical matter in the philosophical sense.


I agree with this assessment, there 's certainly a tangible and practical meaning to the numbers. I will have a look at those TCEC graphs, sounds very interesting. 
 
  

"I don't recall saying good luck."
Back to top
 
IP Logged
 
ReneDescartes
God Member
*****
Offline


Qu'est-ce donc que je
suis? Une chose qui pense.

Posts: 933
Joined: 05/18/10
Gender: Male
Re: Engines & Openings
Reply #4 - 08/26/18 at 16:41:50
Post Tools
Uhohspaghettio wrote on 08/24/18 at 20:21:33:
0.8 is still a draw with perfect play though. I really think it's just the idiosyncratic way Stockfish evaluates the positions - the developers would change it if there were a clear flaw in its opening evaluation


I don't understand this. With perfect play, as in a tablebase, 0.8 is meaningless. There are no probabilities, only facts--win, loss, draw.

Uhohspaghettio wrote on 08/24/18 at 20:21:33:
At the end of the day the value of a pawn in chess is a completely abstract concept that is barely meaningful.


Well, you can't have a pawn (or other material imbalance) without having a position, and there are always features in a position other than the material. Engine evaluation functions are complicated affairs that only roughly use our notions of space or activity, etc.: you can't easily separate out factors. Even measuring space is complicated and can be done in a variety of ways. And the reported evaluation is of a position found by the search functions, which vary just as heavily.

If you want to see how engine evaluations converge, look at the graphs in TCEC games. The evaluations only sometimes (temporarily) converge on something like +1, even when one includes a multiplier for Stockfish to scale down its evaluations. Eventually, of course, evaluations usually converge--on very high positive numbers, very low negative numbers, or zero.

So our current engine evaluations are really a practical matter, for correspondence players, opening analysts, and programmers--not a theoretical matter in the philosophical sense.
  
Back to top
 
IP Logged
 
Uhohspaghettio
God Member
*****
Offline


I Love ChessPublishing!

Posts: 510
Joined: 02/23/11
Re: Engines & Openings
Reply #3 - 08/24/18 at 20:21:33
Post Tools
TalJechin wrote on 07/20/18 at 13:50:50:
Generally, I think Stockfish usually overestimates White's position in the early opening, especially with an early space advantage it goes up around +0.8 or more. On the other hand, sometimes the others may get 'too dynamic'.

One way to get an estimate is to pit the three big ones against each other, and perhaps add one's favourite Stockfish clone too, and see how the games go both move wise and result wise. Often they meet half-way, with one engine decreasing and another increasing White's/Black's advantage.


0.8 is still a draw with perfect play though. I really think it's just the idiosyncratic way Stockfish evaluates the positions - the developers would change it if there were a clear flaw in its opening evaluation. At the end of the day the value of a pawn in chess is a completely abstract concept that is barely meaningful. Stockfish is always like this. You just have to get used to how different chess engines evaluate positions, leading for example to taking Stockfish's 0.8 with a grain of salt - which may of course be part of what you mean but I'm just clarifying about comparing them. 

Since 1.0 is supposed to represent a pawn it might be thought Stockfish values position over material compared to other engines. But then again, as we all know material being lost translates very differently on the board depending on the situation, especially pawns, and that is reflected by engines only occasionally changing anywhere near 1.0 on loss or gain of a pawn. If you were going to try to directly compare the evaluations I think you'd have to weight them based on the fact that for example Stockfish almost always gives a higher opening evaluation, otherwise it seems like comparing finances in different currencies.      
 
  

"I don't recall saying good luck."
Back to top
 
IP Logged
 
HgMan
God Member
*****
Offline


Demand me nothing: What
you know, you know

Posts: 2320
Location: Up on Cripple Creek
Joined: 11/09/04
Gender: Male
Re: Engines & Openings
Reply #2 - 07/20/18 at 22:21:19
Post Tools
I'm coming to the conclusion that the opening is the part of correspondence chess where humans have the most influence. If most games are played by centaurs of one kind or another, being able to "trick" an opponent who overestimates an opening position is one way to make some progress, especially as Black (where I seem to be getting the bulk of my wins at the moment—in the Caro, which in general I think engines don't handle particularly well).
  

"Luck favours the prepared mind."  --Louis Pasteur
Back to top
 
IP Logged
 
TalJechin
God Member
*****
Offline


There is no secret ingredient.

Posts: 2892
Location: Malmö
Joined: 08/12/04
Gender: Male
Re: Engines & Openings
Reply #1 - 07/20/18 at 13:50:50
Post Tools
Generally, I think Stockfish usually overestimates White's position in the early opening, especially with an early space advantage it goes up around +0.8 or more. On the other hand, sometimes the others may get 'too dynamic'.

One way to get an estimate is to pit the three big ones against each other, and perhaps add one's favourite Stockfish clone too, and see how the games go both move wise and result wise. Often they meet half-way, with one engine decreasing and another increasing White's/Black's advantage.
  
Back to top
 
IP Logged
 
HgMan
God Member
*****
Offline


Demand me nothing: What
you know, you know

Posts: 2320
Location: Up on Cripple Creek
Joined: 11/09/04
Gender: Male
Engines & Openings
07/09/18 at 21:30:32
Post Tools
I'm curious about chess openings that chess engines tend to underestimate due to some kind of positional concession that—in the long run—tends not to be quite so decisive. For example, Stockfish hates the Bronstein-Larsen variation of the Caro-Kann. After 5...gxf6, it gives White almost +1.00. After considerable analysis, I don't doubt that the line is a little dubious, but by move 10-15, Stockfish's evaluation has dropped to ~+0.50 in most variations (which sounds about right). Are there other openings (for White or Black) that trip up engines in a similar way?
  

"Luck favours the prepared mind."  --Louis Pasteur
Back to top
 
IP Logged
 
Bookmarks: del.icio.us Digg Facebook Google Google+ Linked in reddit StumbleUpon Twitter Yahoo