Opening analyses: SF / Komodo to 88% similar, not nice!!

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

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Frank Quisinsky
Posts: 4852
Joined: Wed Nov 18, 2009 6:16 pm
Location: Trier, Germany
Contact:

Opening analyses: SF / Komodo to 88% similar, not nice!!

Post by Frank Quisinsky » Fri Jul 14, 2017 7:26 pm

Hi there,

normely in FEOBOS excel worksheet to the still running analyses you can see that the tested engines are max to 67% similar.

In case of Komodo and Stockfish I have a results from 88%.

This are results with 30 seconds ... I checked with both engines 22.034 collected positions (a new update database I am working).

Downlaod:
http://www.amateurschach.de/download/bo ... date-1.zip

I am very disappointed about it!!

Stockfish started after opening book moves with a higher eval. The differents are very small. I hope good to see with the 4 grafics I added in the message.

So, normely it's enough to used one of the two engines for opening analysis because most of the results will be the same with light differents in eval.

Have a look here ...
Stats based on 22.034 different positions!
nps is the same ... ply the same!

Versions:
Komodo 10.3 & Stockfish 8

Image

Image

Image

Image

Best
Frank
I like computer chess!

Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 10:41 am

Re: Opening analyses: SF / Komodo to 88% similar, not nice!!

Post by Lyudmil Tsvetkov » Sat Jul 15, 2017 6:03 am

right, reason is both top engines are fully clueless about the opening.

for example, taking the Dutch.

after 1.d4 f5, there is a single move that would give white advantage and winning chances, 2.d5!

[d]rnbqkbnr/ppppp1pp/8/3P1p2/8/8/PPP1PPPP/RNBQKBNR b KQkq - 0 2

unfortunately, SF and Komodo are fully clueless about that.

as well as all current chess theory, which even does not consider d5.

on the other hand, if black wants to enter the favourable stonewall construction, this is only possible with a transposition via the QG, for example 1.d4 d5 2.c4 e6 3.Nc3 f5!

[d]rnbqkbnr/ppp3pp/4p3/3p1p2/2PP4/2N5/PP2PPPP/R1BQKBNR w KQkq f6 0 4

again, neither top engines, nor current chess theory considers 3...f5 for black as the best move or even at all.

chess openings are a very difficult matter.

User avatar
Nordlandia
Posts: 2453
Joined: Fri Sep 25, 2015 7:38 pm
Location: Sortland, Norway

Re: Opening analyses: SF / Komodo to 88% similar, not nice!!

Post by Nordlandia » Sat Jul 15, 2017 6:16 am

Million dollar question -> how long until engines can play openings by themself above GM level. Or more precisely, when they can construct new openings.

Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 10:41 am

Re: Opening analyses: SF / Komodo to 88% similar, not nice!!

Post by Lyudmil Tsvetkov » Sat Jul 15, 2017 7:04 am

Nordlandia wrote:Million dollar question -> how long until engines can play openings by themself above GM level. Or more precisely, when they can construct new openings.
in 100-150 years, certainly top engines will be much better from now, but I am not sure they will be able to then play perfectly more than 50% of openings.(they play perfectly only 1% now)

certainly, your grand-grand-grand-children will see it.

now the million bucks. :)

Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 10:41 am

Re: Opening analyses: SF / Komodo to 88% similar, not nice!!

Post by Lyudmil Tsvetkov » Sat Jul 15, 2017 7:10 am

seriously, how easy do you think is to have perfect PV of some 400+ plies(with average game length with perfect play supposed to be at least 200 moves), provided that current tops have their difficulties even with some simple endgames requiring only 40 plies perfect play.

and one of those simple endgames constitutes the leaf nodes of a 1.d4 or 1.g3 main line.

obviously, without solving the leaves, you can not solve the root position.

for the time being, top engines and top humans are just playing, I do not know of more than 1 or 2 GMs who would mention perfect play at all, that is, I presume they exist, but so far am clueless who they are.

corres
Posts: 1663
Joined: Wed Nov 18, 2015 10:41 am
Location: hungary

Re: Opening analyses: SF / Komodo to 88% similar, not nice!!

Post by corres » Sat Jul 15, 2017 12:20 pm

Where is the similarity tester by Don Dailey?
:lol:

Nelson Hernandez
Posts: 100
Joined: Sun Nov 14, 2010 8:36 pm
Location: U.S.

Re: Opening analyses: SF / Komodo to 88% similar, not nice!!

Post by Nelson Hernandez » Sat Jul 15, 2017 12:32 pm

That's quite an interesting statement about 2.d5!?

Out of hundreds of thousands of Dutch games I've collected over the years I see only 98 instances of that move. If anyone else in this hobby shared your insight this number would be much, much higher.

I am not being critical. Only pointing out what an idiosyncratic proposition you put forward. It's worth trying!

bnemias
Posts: 373
Joined: Thu Aug 14, 2008 1:21 am
Location: Albuquerque, NM

Re: Opening analyses: SF / Komodo to 88% similar, not nice!!

Post by bnemias » Sat Jul 15, 2017 3:03 pm

Lyudmil Tsvetkov wrote:right, reason is both top engines are fully clueless about the opening.

for example, taking the Dutch.

after 1.d4 f5, there is a single move that would give white advantage and winning chances, 2.d5!
You don't explain why this should be the case.

2. d5, does nothing to help develop, thus is a wasted move at this stage unless it achieves something important. But does it? And how long lasting is that "something"?

Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 10:41 am

Re: Opening analyses: SF / Komodo to 88% similar, not nice!!

Post by Lyudmil Tsvetkov » Sat Jul 15, 2017 6:31 pm

Nelson Hernandez wrote:That's quite an interesting statement about 2.d5!?

Out of hundreds of thousands of Dutch games I've collected over the years I see only 98 instances of that move. If anyone else in this hobby shared your insight this number would be much, much higher.

I am not being critical. Only pointing out what an idiosyncratic proposition you put forward. It's worth trying!
well, I played a bit around with 1.d4 f5 2.d5, and white has better or much better in most, if not absolutely all lines.

apart from that, I have played and analysed extensively with SF the Dutch stonewall, so I know with complete certainty the stonewall is 0.0, full blocking/fortress, by perfect play from both sides, after some 25-30 moves or so.

having this 2 separate knowledge blocks, it is not at all difficult to legitimately suppose 2.d5, not allowing the stonewall, is the best choice.

anyway, it is relatively easy for a relatively strong human + SF to outplay/outanalyse any lone human player/theoretician out there, as well as any mostly stray top egnine output, no matter the length of the analysis.

this is just the beginning of complete transformation of opening theory: you will not know opening theory 10 years from now.

Lyudmil Tsvetkov
Posts: 6052
Joined: Tue Jun 12, 2012 10:41 am

Re: Opening analyses: SF / Komodo to 88% similar, not nice!!

Post by Lyudmil Tsvetkov » Sat Jul 15, 2017 6:35 pm

bnemias wrote:
Lyudmil Tsvetkov wrote:right, reason is both top engines are fully clueless about the opening.

for example, taking the Dutch.

after 1.d4 f5, there is a single move that would give white advantage and winning chances, 2.d5!
You don't explain why this should be the case.

2. d5, does nothing to help develop, thus is a wasted move at this stage unless it achieves something important. But does it? And how long lasting is that "something"?
if you looked closer, I posted also a second diagram, representing the Dutch stonewall.

as I have complete certainty the stonewall is completely drawn in a short while, only reasonable move for white seeking an advantage after 1...f5 is 2.d5.

for the very same reason, black should avoid playing the Dutch in a straightforward way, and instead get the stonewall via a QGD transposition.

Post Reply