Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Suggestion thread for TCEC
- - By 101 (*) Date 2015-10-11 21:29 Edited 2015-10-12 03:41
Note: Please forgive me for double-posting here as well as in the TCEC forum, but few TCEC regulars are members of that forum, among others Nelson Hernandez, whose opinion on the below-mentioned suggestion I would be interested in.

In this thread, I invite everyone to share their suggestions for TCEC, for season 8 or for future seasons if impossible to implement mid-season. To post such suggestions in a permanent medium like a forum should prove superior to posting it in the Chatwing chat.

Now to one suggestion I have for Martin and TCEC for season 9. We have already seen some very interesting endings this season. Some of them have unfortunately, in my opinion, been adjudicated when five pieces remained. One of these is the KBBKN ending from Gull-Hannibal (game 12 of stage 3), where Gull would only win within the 50 moves rule by seven plies with perfect play. This is one of the situation where two engines without tablebases could have produced an interesting endgame lesson, which I believe many would have enjoyed watching.

Here are some other examples of endings which I would like to see "played through", i.e. converted, in TCEC:
(KQvKR) - interesting for humans, and some weeker engines might fail to win, especially against a tablebase-equipped defender
(KQvKRP)
KQPvKQ - often almost impossible to play perfectly, even for strong engines, without tablebases
KQNvKQ - a win or a difficult-to-defend draw in some exceptional situations (general result: draw)
KQBvKQ - a win or a difficult-to-defend draw in some exceptional situations (general result: draw)
KRPvKR - in some special cases difficult (but possible) to win or defend
KRBvKR - a win or a difficult-to-defend draw in some exceptional situations (general result: draw)
KRNvKR - a win or a difficult-to-defend draw in some exceptional situations (general result: draw)
KBPvKB - long and difficult wins are few and far between for engines, but this should still be an exception both for those cases and for the often still instructive value of this endgame in the simpler cases
KNPvKN - long and difficult wins are few and far between for engines, but this should still be an exception both for those cases and for the often still instructive value of this endgame in the simpler cases
KBPvKN - in some cases instructive to see converted
KNPvKB - in some cases instructive to see converted
KNNKP - often, not always (but then sometimes difficult to draw), a win disregarding the 50 moves rule. Sometimes a win, sometimes a draw, under FIDE rules; obviously, extremely precise play is often required to convert within the 50 moves rule.
KPPvKP - can transpose into one of the above-mentioned endings

Here is one addition to the list from Chatwing user Spliff Jiffer:
"KPPKN can be relevant..eg:
8/8/8/8/7k/1K2P3/P7/4n3 w - - 0 32
1.Kc4 wins while latest SF dev. seems clueless...Komodo gets it easily for example...regards"

There are even some extremely exceptional cases in addition to these, which I don't mention (like the longest win of KBNKN). My primary suggestion is this: Replace 5 piece adjudication with 4 piece adjudication, possibly with an exception for the KQKR ending. A secondary suggestion is to use the list above, a solution which is good but not perfect (there will always be some extremely exceptional cases in addition which one would have liked to include but would make the list much longer and further reduce the point of even having a 5 piece adjudication instead of 4).
Parent - - By Nelson Hernandez (Gold) [us] Date 2015-10-12 00:30
Why half-measures? 

I would favor 6-man tablebase adjudication for all games that simplify to six pieces and all moves played out to the game's conclusion if a win, or an immediate halt if a draw when the board simplifies to six pieces.  To do this you might need dual Nalimov/Syzygy support.

Now, the practicalities of this I can't address.  Martin would know.
Parent - - By 101 (*) Date 2015-10-12 02:08
One problem with this is that in some drawn positions, if the defending engine doesn't have tablebases, it might lose with inaccurate play. This consequence of not having tablebases should be reflected in the outcome. More importantly, this is still a tournament for showing the differences between engines. Suppose one engine had a 32 men tablebase support. Would we allow 32 men tablebase adjudication, if possible? Of course not! The fun would instead be to see how long the other engines would hold their own against that engine.
Parent - - By Nelson Hernandez (Gold) [us] Date 2015-10-12 20:37
If 32-man adjudication were possible there would be no need for a tournament.  Or for this hobby.  (You can turn out the lights.)

Okay on your objection.  If an engine lacks tb support it doesn't get the benefit of a draw upon simplification.
Parent - - By 101 (*) Date 2015-10-12 21:10
There are several points here.

Say 20 men adjudication were possible. One could obviously argue for this, if one consistently want to end things where you have a clear picture. If that is the case, one should, if possible, allow for 6 men adjudication or 7 men. Note that this in the end would go against what I think is the intention: To reduce the length of games unnecessarily in the vast majority of cases where the result is given, cf. the comments from Trym among others cited in this thread. Even at 7 men, a large majority of games could safely be adjudicated today, but when you approach 8, 9, 10, you will miss a universe of positions where powerful engines with tablebase access, be it 5 or 10 men, could prove their superiority against engines without (the same) tablebase support. I thought that separating engines strengthwise was the basis for TCEC.

The only consistent opinion is to say that we simply sacrifce something for a reduction of time, which I think is not that big, by the way. (In most cases, the shuffling would take place before 5 men is reached.) This is done with evaluation based adjudication, since a tiny percentage of games would have had another result than a draw of those adjudicated after move 39 where there has been a |evaluation|<=0.05 agreement for 5 moves with no pawn move or capture. The difference is, as I pointed out to Trym, that with a tablebase adjudication, engines don't necessarily agree, and in the Gull-Hannibal case, Gull thought it had an advantage of several pawns, though not enough to have an evaluation based win adjudication, which is fair, because this rule should only kick in where it is almost guaranteed that the engine with the advantage will manage to win. This was certainly not the case against Hannibal, which by the way showed nearly a zero score. Neither Gull nor Hannibal would, without tablebases, have provided perfect play, and thus, I find it highly unclear that Gull would have won the final KBbKN ending with a distance to zero of 93 under optimal play.
Parent - - By Nelson Hernandez (Gold) [us] Date 2015-10-13 01:21
In the final analysis endgames are not my thing so I offer my blessing to your effort to improve TCEC standards without getting personally engaged. 

I favor the use of whatever is available to make the engines play the endgame stronger, 1) reducing the quantity of farcical outcomes and 2) giving us the assurance that a game that reduces to six pieces will have an accurate outcome, i.e. TB adjudication.  The rest is just details.
Parent - - By 101 (*) Date 2015-10-13 20:22
Thank you for the encouraging words.

To what you write: I too favour the use of "whatever is available to make the engines play the endgame stronger", as long as it is supported by the engine itself. Tablebase adjudication where at least one engine does not support tablebases does not fall into this category, in my opinion. One could argue for it when both engines would have produced optimal moves, and anyone interested could just play through the rest of the endgame here, here or here (in the cases where the 50 moves rule makes a difference, this is the one to use). I for one don't like this separation and I find that it does not do justice to the many moves already played to reach the endgame. It would also give me some feeling of pleasure to see the engines play the rest perfectly on the TCEC board which I would not get in the same way from the links above.

"1) reducing the quantity of farcical outcomes"
Not sure what you are thinking about here.

"2) giving us the assurance that a game that reduces to six pieces will have an accurate outcome, i.e. TB adjudication."
I don't get the point of having an "accurate" outcome if an engine by itself (including whatever extensions it supports) is not capable of providing it.
Parent - - By Nelson Hernandez (Gold) [us] Date 2015-10-14 12:51
I think this summarizes our views.

If you have a game where both engines use tablebases, at the instant they simplify to six pieces,

  In all cases, the game should be adjudicated.  You would argue that the mates should be rapidly played out from there.  I don't care if they play out or not.

If you have a game where one engine uses tablebases and the other does not, at the instant they simplify to six pieces,

  Case 1: EGTB indicates EGTB-using engine will win - the game ought to be adjudicated a win because the outcome is certain.
  Case 2: EGTB indicates EGTB-using engine will lose - you would argue play on, I would argue adjudicate.
  Case 3: EGTB indicates engines will draw - you would argue play on, I would argue adjudicate.

If you have a game where both engines do not use tablebases, at the instant they simplify to six pieces,

  You would argue the engines should play on, I would argue the game ought to be adjudicated.

As things stand we adjudicate games at five pieces already even if the engines lack tablebase support.  All I am proposing is that we adjudicate at six pieces.  (If we had ten piece tablebases I would argue that we adjudicate there.)

What is the logic of using tablebases in cases where the engines don't have tablebase support?  Instant adjudication certainly shortens the games, a desirable outcome, especially if you are facing a tedious drawn position.  But more fundamentally tablebases are within the realm of mankind's knowledge now even if they are not within the realm of that particular engine's knowledge.  If tablebases indicate the engines have reached a drawn or decisive position then that ought to be the outcome.  Spare us the embarrassment of an engine blowing a six-piece position.  Now, whether you play out the mate or not I am indifferent toward.  I suppose the main question there is if the tournament software will support that.
Parent - - By 101 (*) Date 2015-10-14 18:05 Edited 2015-10-14 21:31
I notice that what you personally would like to see is as flawless chess as possible (even when this means one has to cut off a part of the game (case 2 and case 3 where the engine without tablebases has winning chances)). I respect that view, but then I would imagine that you don't see the point of having many of the weaker engines play, which I would assume have almost no areas of the game where they are the strongest engine. If so, it would make more sense to have Stockfish or Komodo play those games, unless you find TCEC as a battleground for weaker, developing engines, a higher priority. I myself would rather have the stage system replaced by an accordingly longer Stockfish-Komodo superfinal, thus coming much closer to the ideal of perfect chess, which I we are both interested in seeing technology approaching. Such a permanent match would come much closer to what can be achieved in top level correspondance chess. If you want to see what technology can achieve, it is at least clear that given n € [5,7] men adjudication, every engine other than Komodo or Stockfish should have its n + (m € [1,30 - n]) men endings played out by Komodo and Stockfish, since they will almost always play better. You might argue for TCEC as a testing ground for weaker engines' endgame play, but that argument seems suspect in relation to case 2 and 3 when you would have wanted 10 men table adjudication. Sorry for arguing against an undeclared opinion of yours; I guess you too would have wanted to see a permanent Komodo-Stockfish final ...

Edit: Or not?

Rozgar
@Cato: Would Season 9 have more participating engines?!   
18 minutes ago

Cato the Younger (VIP)
I don't know Rozger; I would prefer to see 32  
18 minutes ago

101
Cato want[s] to see 30 far weaker engines, but would rather not see their "weak" play with 10 pieces or less (ref. http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?pid=556518#pid556518). Why is the endgame so much less interesting than the middlegame? I don't get it.
5 minutes ago
Parent - By Nelson Hernandez (Gold) [us] Date 2015-10-15 02:47
I think you need consistent rules for the whole season, so no, I would not favor changing the EGTB regime for the Superfinal.  And by no means am I advocating or secretly favoring two-engine hegemony.

I think that in general you make a good argument for letting the engines play to the limits of their organic skill.  But really, all we are discussing is more in the realm of philosophy.  The decision has to take into account practicalities which outside my area of expertise.  This all really comes down to a decision that Martin and Anton have to make based on the practical limitations of the tournament software and maybe to a lesser extent the limitations of the hardware.
Parent - - By turbojuice1122 (Gold) [us] Date 2015-10-14 20:57

> As things stand we adjudicate games at five pieces already even if the engines lack tablebase support.  All I am proposing is that we adjudicate at six pieces.  (If we had ten piece tablebases I would argue that we adjudicate there.)


Maybe I missed this in a previous post of yours, but why not argue that we adjudicate at 7 pieces, since Lomonosov covers these?
Parent - - By Nelson Hernandez (Gold) [us] Date 2015-10-15 02:41
I would argue for 7 piece adjudication if the engines supported Lomonosov.  The problem, if I am not mistaken, is that you would need someone stopping the game and looking up the result, which is administratively impractical.  If it were practical then sure, I'd go for that.

I might add that 6 piece adjudication may be impractical as well.  It is much more possible, of course, and I have argued to TCEC management that we should adopt 6-man adjudication going forward.  But you would need dual Syzygy and Nalimov support to carry it out as 101 proposes.  You need Nalimov to play out and display the rest of the game perfectly.  (If I am wrong about this I stand corrected; I am not an expert in EGTBs.)
Parent - By Dr.X (Gold) Date 2015-10-15 03:56
You do not need to stop the game to check for results.

The game can remain unhindered and continue in progress as long as you have access to the current move  notation placing it onto a clipboard and pasting the notation in something like a sandbox.

I do it all the time while games are in progress -placing the notation in a sandbox where TB7 online are accessible.
Parent - - By 101 (*) Date 2015-10-15 14:56
I agree about the impracticality of 7 piece adjudication, though it might be possible. In many endings, there would only be seconds to have an immediate adjudication before the next move is played. It could of course be possible to have the adjudication slightly delayed, nullifying moves played after the adjudicated position. Another problem is that we cannot, from my own experience, expect the Lomonosov connection to be reliable at all times. Because of this, a tiny percentage of 7 men positions, in many cases every one in a single game, will not be checked. It is up to Martin to decide whether or not this is a decisive problem. In my opinion, it is not. It is at least clear that Lomonosov tablebases should only be used for 7 men positions; for 6 men and less, Syzygy should be used (see below).

I don't think dual Syzygy and Nalimov support is needed, at least not for the adjudication system you want. Quite the contrary, Syzygy tablebases will always give a consise, definitive answer to which of the three results should be assigned to a given position (with or without he 50 moves rule), while I don't believe the Nalimov tablebases unless modified will take the 50 moves rule into account. (I don't think you need the full tablebase set for this either, only the WDL files.) I thought that you did not want the remaining moves to be played out to mate in case of an adjudication. Neither do I, that, is, automatically. Even if you want that, the DTZ files should, in addition to the WDL files, be enough to do this, provided that you let a Syzygy-compatible engine execute the moves. I must state a reservation that I don't know if a look-up feature can be more easily implemented with Nalimov tablebases, provided that a 50 men modification is possible. Note that you would also have to have such a modification for the Lomonosov tablebases, but I believe at least the latest Aquarium version provides this. Whether or not Martin can combine this with his GUI, I don't know.

Let me underline that I myself want tablebase adjudication with as few pieces as possible, but the engines to access tablebases with as many as possible.
Parent - - By Nelson Hernandez (Gold) [us] Date 2015-10-15 23:46
I just learned today from Martin that the proposed 40-core server that is planned for the Superfinal is going to be a rental.  I had always imagined that such a machine would be bought and serve as the standard for a minimum of three years.  Not so.

The implications of this for 6-man EGTB support are potentially stark.  EGTBs simply loaded onto hard drives simply won't do.  You need SSDs.  Will Anton be willing to shell out whatever incremental amount is needed to supply that hardware for the sole purpose of boosting EGTB from 5 to 6-man for some engines but not others?  I kind of doubt it.

Of course if you owned such a server you would custom-build the thing from the ground up.  If you're renting your options may be limited.
Parent - - By 101 (*) Date 2015-10-16 01:18 Edited 2015-10-16 10:52
I will not say anything categorical, just stress that my opinion is that Chessdom and Martin Thoresen should be free to act upon what I assume they agree with me on: that in this superfinal, we will most probably see both engines having tablebase capabilities. (The probability is high for this in future superfinals as well, as the competition to get that far becomes stronger and stronger. For a new challenger engine to beat Stockfish or Komodo and enter the superfinal, dedication and willingness to implement features like tablebases is imperative from the engine author. Of course, it could happen from superfinal to superfinal that for some reason, tablebases are not implemented, but I would assume this to be infrequent. Season 5 (with the Komodo TCEC equivalent) is the only exception so far, and it should happen more and more rarely.)

Removing the tablebases in the superfinal would certainly be looked upon as a clear downgrade, and the negative elo change will be comparable to the (hopefully) positive one the core double will provide. Another argument is that the rules on this should be the same for the whole season. (I don't agree with that, as there is meta-rule saying this, but if you (and Martin) think so, I don't mind utilize that view here.)

Addendum:
Giving it some further thought, I actually find that my desire to have engines "utilizing" 7 men tablebases, even only in 7 men endgames, outweights that the only way to do this is to have a (de facto) adjudication of all <= 7 men endgames. (Of course, I forgot in my next to latest post that since any <7 men endgame has been a 7 men endgame at a prior stage, Syzygy tablebases would not be needed for adjudication at all unless there is a technical problem.) I upheld my point that most unnecessarily long endgame phases take place before 5 men are reached, and typically in a position of material imbalance, which is more likely to happen with an odd number of men, like 7. An example is the Stockfish-Gull RPPvRP farce we just saw.

Why have I changed my mind so dramatically? I have realized that the difference between a tablebase implementation (partly) provided by an engine author and some form of mechanism to combine an engine and a look-up feature is in one sense artificial: the corresponding engine performances are both a function of the code of the engine in question. To be clear: I prefer my "secondary" suggestion, subsidiarily my "primary" suggestion, in the opening post, to status quo; I prefer 7 men look-ups to either suggestion. Extending 5 to 6 men adjudication I would probably still oppose, partly because in the vast majority of cases where these positions are dead drawn, this will already be reflected by the engine (Hannibal with negative contempt is an exception we have to live with) due to an even number of men.

There are two cases for a 7 men implementation: 1) the superfinal. We will either see a) SF vs. Komodo or b) either of those vs. Gull.

2) season 9. Giving weaker engines without a 5 men tablebase implementation the benefit of 7 men look-ups seems radical, but remember that they don't get the benefit of 5 or 6 men search in positions with 8 men or more.

I thus support 7 men adjudication for all 1a), 1b), and 2). Quite an experience, to agree more and more with, you, Nelson, with regards to this. If I get you right, you don't support it for the superfinal, but it would probably not be possible to implement in time, anyway.
Parent - By Dr.X (Gold) Date 2015-10-16 04:22

> I will not say anything categorical


:smile:

We strode across that lonely plain like men
      who seek the road they strayed from and who count
      the time lost till they find it once again.

                                                                                          111  Canto 1
                                                                                         The Puragtorio
Parent - By Nelson Hernandez (Gold) [us] Date 2015-10-17 01:14
I am not a policymaker, not even in the realm of openings (though I do have considerable influence in that field).  Thus I think you've expended a lot of effort laying down your policy points when really the decision-maker is Anton, and his frames of reference (as well as Martin's) are revenue and expense, administrative overhead, adherence to rules and IT feasibility.  The aesthetics you outline are a comparatively minor matter in the larger scheme.  I wish it were not so, but practicalities are paramount in an enterprise such as this one.
Parent - By Sesse (****) Date 2015-10-29 11:27
You don't need to do manual lookup, the Lomonosov tablebases have an (internal) API. It's used, in particular, by Aquarium.

/* Steinar */
Parent - - By Peter Grayson (****) [gb] Date 2015-10-12 11:24
Consideration must be given to the fact that engines using EGTB's can access them surprisingly early in the game with the exceptional ply depths now being reached through efficient programming techniques and the type of  hardware being used in the TCEC tournament. The absence of EGTB's may require some reprogramming to include additional knowledge for the engine to perform at the desired level in this type of public tournament.

Engines that cannot access EGTB's are likely having their move selection affected by that lack of predetermined knowledge accuracy much earlier in the move chain and well before the EGTB positions are reached that may influence the final result much more than by the time the game reaches a known end game result. It may also be the case that engine authors reduce endgame knowledge if it is provided from the use of EGTB's.

From a consumer perspective, I see no value in an engine that does not use EGTB's because I would not consider using it.

One other issue that must be considered from the tournament organiser's perspective; if no or reduced EGTB's are used, is the consequential extended game time wanted? Running private matches myself,  I can confirm that even at much faster time controls it is undesirable to see an extra 100 moves wasting time when the outcome was already known through EGTB's.

Therefore I see no benefit in requiring engines to play out already established endings.

PeterG
Parent - By Dr.X (Gold) Date 2015-10-12 12:52 Edited 2015-10-12 12:56
They could always put in place  the  Lomonosov Endgame Tablebases with 7 pieces or less left on the board. :grin:

That would be an end to this debate!
Parent - - By Uri Blass (*****) [il] Date 2015-10-15 09:03
I prefer instead of using 20 cores to have games with only one core at time control that is 20 times longer and hash that is 20 times lower when people can choose which game to watch so if there is a boring game you can switch to look at a different game.

I prefer no adjudication based on evaluation and also no adjudication based on tablebase positions.
The only case when I like to see adjudication is based on the fifty move rule or repetition because many engines do not know to claim draw correctly because the programmers assumed that the interface is going to do it for them.

For opening it is better to have unbalanced openings like 1.e4 h5 when the engines play both sides so we get less draw and the stronger engine has chances to win 1.5:0.5.
Parent - - By marcelk (***) [nl] Date 2015-10-15 21:31
I'm working on an UCI engine now, and I don't know how to claim draw correctly because unfortunately the UCI specs I found don't mention anything like that. Do you have a reference how to do that?
Parent - By h.g.muller (****) [nl] Date 2015-11-23 18:27
Creative use of the specs wouldlead you to the following method: print "bestmove 0000", with a "score 0 cp". The null move should make it dawn on the GUI that you are not going to play on. Similarly, a null move in combination with a negative score could express the intent to resign.
- By 101 (*) Date 2015-10-12 02:44 Edited 2015-10-12 21:22
Discussion about the matter in the Chatwing chat
101 
Cato: I want Martin to consider 4 piece adjudication instead of 5 next season. I might need some support in order to be heard, though. I am interested in your opinion on this: http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=30630
39 minutes ago

  Cato the Younger  (VIP)
I don't want to tangle with you, 101, but I am gung-ho in favor of 6-man tablebases
38 minutes ago

  nsousa254 
im in favor of 7-men lookup tables
38 minutes ago

  Cato the Younger  (VIP)
not practical msousa, but if it were, I'd agree with that too
38 minutes ago

  Albert van Houten 
yeah, I don't see the point of playing out LSB vs. DSB with one pawn on the board for an eternity.
38 minutes ago

chatWING
  kan  (VIP)
7 men might be worse than ponder atm
38 minutes ago

  101 
Cato, don't get this wrong. I too want six men tablebases, but not for adjudication.
37 minutes ago

  nsousa254 
it's 7-men lookup, not full lines
37 minutes ago

  Cato the Younger  (VIP)
101, that statement kind of stuns me...that's really interesting
36 minutes ago

  Cato the Younger  (VIP)
so you just want the moves from tb to be played out very rapidly once 6-man is attained?
36 minutes ago

  kan  (VIP)
maybe for educating the audience cato?
34 minutes ago

  101 
Yes, Cato. Or, if one of the engines don't have tablebases, the result might be different. In any case: Interesting to watch, even if we know what the theoretical result is.
34 minutes ago

  101 
Especially the endings KBBvN and KNNvKP I find aesthetically pleasing. They have both been reached in TCEC, but unfortunately, we have not been able to see them being played out.
34 minutes ago

  Cato the Younger  (VIP)
I could be more of an extremist then...6-man tbs and play out the rest of the game to mate...why not?
33 minutes ago

  Albert van Houten 
but 101, most of those will be dead draws like DSB vs. LSB + pawn, with the pawnless king covering the promotion route ...
33 minutes ago

  Cato the Younger  (VIP)
or if a draw halt on the spot of achieving 6-man
32 minutes ago

  Albert van Houten 
some engines are programmed to get to the draw asap, but others will hang on almost indefinitely to that pawn.
32 minutes ago

  dosamer 
Cato that is a great idea, because there are engines without table base. So even if the tablebase says its a win, it could be a draw/loss.
31 minutes ago

  101 
Albert: I think that is the price one has to pay. After all, using this logic, one should probably have evaluation based adjudication after move 30 instead of 40 since the vast majority of cases will be drawn anyway.

We miss some of the most beautiful and fascinating endgame finishes because of 5 men adjudication.
28 minutes ago

  Albert van Houten 
well, the only really drawn out endgames with 5-men would be QP vs Q, and Q vs R+minor piece. The latter is pretty rare over here
27 minutes ago

  Albert van Houten 
it is tricky 101. Unless the automated adjudication is set up to discriminate with regards to what endgame is being played. Just to get rid of the different coloured bishops boredom
26 minutes ago

  101 
Well, Albert, much of the time of TCEC is boredom. Does a slightly higher ratio really matter?
25 minutes ago

  joseph ( tick tock ) 
hours of boredom followed by moments of terror
24 minutes ago

  Albert van Houten 
well, with these time controls you'd be happy to avoid a 3-hour marathon of DSB vs LSB. At least some engines will be happy to give the pawn away, just to get it over with ...
24 minutes ago

101 
The engines will seldom have that amount of time in a KBPvKbP ending.
23 minutes ago

  Albert van Houten 
one hour each + increment adds up quite quickly. Who knows how long Gull and SF could have gone on in their game ...
22 minutes ago

101 
Albert: I am pretty sure that Gull would have shown <=0.05 in these five man endings.
20 minutes ago

  Albert van Houten 
Gull maybe, but last season there was nearly a 0-1 simply because two engines could not evaluate properly at all.

  Albert van Houten 
Personally, I'd prefer it if the automated adjudication kicks in depending on the material conditions. Thus play out Q vs RB, but don't play out DSB vs LSB plus one pawn that has been blocked by the opposing king.

  101 
Well, such things happens in lower stages, which I am not really that much interested in. What you state I read as an argument for more conservative win evaluation adjudication rules.
17 minutes ago

  Albert van Houten 
well, we have seen wins declared by TCEC win rule, when it is completely not obvious to the human eye why it is won too
17 minutes ago

Albert van Houten 
other times, it is so obviously won, but the eval just hovers around 3 or 4.
16 minutes ago

Castling is useless 
@Albert, pv's say everything
16 minutes ago

101 
Albert: KQvKRB is in the vast majority of cases a draw, even if engines which don't have tablebases think it is slightly better (>0.05) for the side with the queen, so probably not a good example (though no pawn moves).
16 minutes ago

Albert van Houten 
we can't adjudicate on the basis of PVs, especially not when there are (potential) fortresses involved.
16 minutes ago

Albert van Houten 
well, what are the supposedly interesting cases then? KR vs KBP?
15 minutes ago

Castling is useless 
KR vs KBP is easy draw in most cases
14 minutes ago

Castling is useless 
R can draw by eating pawn and even KB vs KR is a draw
14 minutes ago

Albert van Houten 
yes, so what are the interesting 5 men endgames that we're dying to see?
14 minutes ago

101 
Albert: I actually think this is a good point. Perhaps the win eval threshold should be significantly raised, for example to something along 20-100. In many cases, the top engines don't need many moves from 6.5 to 20.

A list: http://open-chess.org/viewtopic.php?f=43&t=2896
13 minutes ago

Albert van Houten 
no, but Gull for instance loves to troll with +3 or +4, when an engine like SF or K would be like +10 ...
13 minutes ago

101 
As you can see in the list, Albert, there are quite a few classes of interesting endings, even if they don't happen very frequently.
12 minutes ago

Albert van Houten 
well, what are the interesting 5 men endings?
11 minutes ago

101 
Albert: I just posted the link. http://open-chess.org/viewtopic.php?f=43&t=2896
10 minutes ago

Albert van Houten 
I have seen the list, but most of those endings are really not that interesting. KBPKB is not that interesting, sorry.
7 minutes ago

[On this I agree, in the vast majority of cases, but there are a few interesting study-like positions, especially with same-coloured bishops, like this:
8/8/6B1/8/2P5/8/2K5/k4b2 w - - 0 1
(mate in 51, distance to conversion in a few moves less. Click "Lookup" to play through this fascinating win-maneouvre!).]

Albert van Houten 
anyway, time for me to call it a night.
6 minutes ago

Spliff Jiffer 
@101 and Albert...also KPPKN can be relevant..eg:
8/8/8/8/7k/1K2P3/P7/4n3 w - - 0 32
1.Kc4 wins while latest SF dev. seems clueless...Komodo gets it easily for example...regards
3 minutes ago

Some other excerpts
j.sparks 
Ai Au [101]? Simple Question to you: You as a human hessrsplayer, do you think you could have learned something from this endgame Gull vs. hannibal - for your own play OTB?
6 hours ago

101 
j.sparks: I might have picked up a few points, but if I really wanted to learn the full technique, I would probably have to read the chapter about the KBBvKN ending in John Nunn's Secrets of Minor Piece Endings.
6 hours ago

j.sparks 
Well, 101: Probability to get this Gull vs Hannibal endgame is 1/1000. The time you invest into this endgame is wasted, just my 2 cents. Better to go for other endgames, a very good book to study is Glenn Flears Practical endgames, beyond the basics, where the term NQE, not quite an endgame occurs, when each side hase 2 pieces.
6 hours ago

101
J.sparks, thanks for the suggestion.
6 hours ago

j.sparks
You are welcome, of course you are right, it is not clear if Gull would have found the win in 48 [really 42, I gave the wrong number, as pointed out by Trym] moves. That is right.
6 hours ago

101 
Trym! I would very much want your opinion on the current 5 men win adjudication rules, considering the Gull-Hannibal BBvN finish we never got to see: http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=30630
41 minutes ago

  Trym  (VIP)
101: My opinions are of no relevance to anything, certainly not chess. I'm good with the rules as they are.
40 minutes ago

101 
Ok. My impression was that you was pretty informed in the ending. I would have liked to see it played out, with only a two-move margin to the 50 moves rule.
39 minutes ago

  Trym  (VIP)
As far as I remember it was a 94-ply distance to zero win. Possibly could have ended differently without adj., but if we go that route a lot of games will draw out for hours. I do think that most engines play 5-man endings accurately without tbs.
36 minutes ago

  101 
Trym: It seems I remembered the wrong number. Anyway, I think Gull would not have managed to win it against perfect play. Hannibal would not provide this, so I consider likely the result in that case unclear. We would have got to see a more "human" struggle, though. My point is that one thing is to adjudicate wins or draws where both engines agree about the result. Another is to adjudicate a win where one engine certainly does not agree (Hannibal showed nearly 0).
27 minutes ago

Trym  (VIP)
101: Fair point. You already know Anton's address. I'm just a "helper" here, I have no influence whatsoever.
26 minutes ago
- - By 101 (*) Date 2015-10-14 18:29
It seems that there are two main reasons for having tablebase adjudication: One upheld by most people who have given their opinion in this debate, including Albert van Houten, Trym (though I might have convinced him of my view), Nelson Hernandez, and others, that this will prevent a lot of unnecesary game prolongation. The other concerns cases where the outcome is not clear: Nelson Hernandez is here more interested in having the "correct" (theoretical) outcome than in the "correct" outcome in terms of engine capability.

If my view were to be taken into consideration by Martin Thoresen and Anton Mihajlov, it would be most natural for me, as a compromise, to recommend the secondary suggestion of using the list given in the opening post as exception cases. That list covers the vast majority of interesting (four and) five men cases. I don't if it is supported by the GUI, though. (Probably it is not, unfortunately.)
Parent - - By Graham Banks (*****) [nz] Date 2015-10-14 18:54
As somebody who uses a lot of cpu time and electricity to run tournaments at longer than blitz time controls, I'd go with automatic tablebases adjudication all the time.
Parent - - By h.g.muller (****) [nl] Date 2015-10-14 19:23
You could of course also just look which engine thought it came out of book best, and award the win to that. This is guaranteed to reduce your electricity consumption by at least a factor 100! When you are not really interested in the capabilities of the engines, it is truly amazing how many tricks one has at his disposal to improve efficiency.

In WinBoard I implemented an interesting method for saving end-game time that is much less disruptive than adjudication: you can limit the search depth on a per-engine basis in positions that are an EGT draw. This prevents that the engines would waste a lot of time searching very deep for something you already know they cannot possibly find, because it is just not there. It is mainly meant for use on engines that do support EGT, so that you know they can never make any mistake no matter how short you make them think. They would still play by their evaluation to keep an opponent that does not have EGT under pressure, giving it the chance to bluder the game away.
Parent - - By Ozymandias (****) [es] Date 2015-10-14 21:00

>When you are not really interested in the capabilities of the engines, it is truly amazing how many tricks one has at his disposal to improve efficiency.


When you are interested in the engines' capabilities, which really matter, the number of tricks is greatly reduced.
Parent - - By h.g.muller (****) [nl] Date 2015-10-14 21:52
Yes, and awarding wins and draws by adjudication to engines that would, respectively, draw or lose these games is one of the ticks that would get axed...
Parent - - By Ozymandias (****) [es] Date 2015-10-15 10:39
No, it wouldn't. Tablebase adjudication helps reduce the time needed, to determine the engine's strength, before reaching positions which don't matter anymore.
Parent - - By h.g.muller (****) [nl] Date 2015-10-15 14:41
So you claim awarding engines points for games they would not be able to win a good way to determine their strength? You must have a funny defenition of 'strength' then. Normally people would understand by 'strength' the ability of the engine to win games.
Parent - - By Ozymandias (****) [es] Date 2015-10-15 16:39
You're choosing to leave part of my statement out, I wonder why that is so? Indeed, "awarding engines points for games they would not be able to win [is] a good way to determine their strength", if "you are interested in the engines' capabilities, which really matter" and you stop the game just "before reaching positions which don't matter anymore".
Parent - - By 101 (*) Date 2015-10-15 18:29
The root of the misunderstanding here is the comma you placed before "which really matter", which changes your intended meaning completely.

Then, I still don't understand why the (late) endgame is so less interesting or even doesn't matter, which you and Nelson tend to think. (Nelson hasn't really answered me on that ...)
Parent - - By h.g.muller (****) [nl] Date 2015-10-15 19:21
Well, even without the comma it remains weird. Apparently some think that it doesn't matter whether engines can win games.

This TCEC seems to have awfully little to do with Chess according to FIDE rules.
Parent - By turbojuice1122 (Gold) [us] Date 2015-10-15 20:31

> This TCEC seems to have awfully little to do with Chess according to FIDE rules.


FIDE rules sometimes have awfully little to do with chess.  Adjudicating games due to an opponent being late instead of simply starting the clock and letting it run down?  Draws for games that can technically be won with a few more moves?
Parent - By Ozymandias (****) [es] Date 2015-10-15 19:48
I think this was a clear answer "If 32-man adjudication were possible there would be no need for […] this hobby." I can only explain it, if that's what you require, but I wouldn't be giving you a different reply.
Parent - - By Peter Grayson (****) [gb] Date 2015-10-15 08:20

> You could of course also just look which engine thought it came out of book best, and award the win to that.


So what is the difference between an openings book and end game table bases? The engine is not required to do anything in both cases where they are used. Therefore if the aim is to prevent human interference that provides any sort of game direction then no opening books or table bases should be allowed otherwise as soon as books are allowed the arguments against end game table bases is lost.

PeterG
Parent - - By h.g.muller (****) [nl] Date 2015-10-15 09:44
The argument is not against the use of EGT per se, but by using them for adjudication. If engines can and do use EGT to improve their end-game, fine. Then it is a property of the engine. Wat distorts matters is to let an engine benefit rom knowledge it doesn't have, through GUI adjudication.

One could indeed argue the same thing for opening books: engines that suck in opening play get that repaired by skipping this phase of the game, starting from a line provided by the GUI book. Engines might not even know how to castle, and get away with that because most opening lines do contain the castling. This is an argument for letting engines take care of the opening without GUI assistance, if needed by playing from their own opening book.
Parent - By Peter Grayson (****) [gb] Date 2015-10-15 18:13
Thanks for clarification. I was not aware that Gull and Hannibal do not use end game table bases, but now I am aware I better understand the argument.

PeterG
Parent - By h.g.muller (****) [nl] Date 2015-10-14 19:10 Edited 2015-10-14 19:24
Well, it depends on your goals, I guess. If you see this as a serious competition between engines then any form of adjudication is of course highly undesirable. It corrupts the result by awarding points to engines that would never have managed to collect those on their own capabilities. On the other hand, if you just see this as a show to produce entertaining high-level Chess games, then you would of course want adjudication based on perfect end-game tables at the earliest opportunity. You could even consider to invoke a special end-game analyzer like FinalGen to take over from the playing engine to flawlessly play out the game once it becomes within the possibility of the latter to determine the theoretical outcome.

Under the current rules TCEC is not even a true Chess tournament: they apply winning and drawing conditions that are not mentioned in the FIDE rules. You might as well play King-of-the-Hill with engines thinking they are playing ordinary Mad Queen...

Funny thing is that these adjudication rules are highly exploitable. It is only a matter of time before it will dawn on one of the participants that the adjudication will screw you when you report too negative or near-zero scores, and that it would to your advantage to always add 10 Pawns to your own eval before printing it!:grin: Or always report 0.00 when you think you are worse.
- By 101 (*) Date 2015-10-26 01:27 Edited 2015-10-26 01:33
New tablebase/adjudication/rules discussion:

[My suggestion is: Clarify the rules as soon as possible. Does the 50 moves rule apply to all positions, or only to >5 men positions?]

Warp 
Some 5-man endgame positions are winnable, but only with more than 50 moves. When using the 50-move rule, they are draw. Is this taken into account in the tablebases?
an hour ago
chatWING
  Mark Lefler (Komodo)  (VIP)
Warp: With Syzygy there is a 50 move rule option, which chould be "checked" to enforce the 50 move rule. YOu can turn it off if you want to see things like how long a mate would check without the 50 move rule. At least, that is how we implemented it in Komodo.
an hour ago
chatWING
  Warp 
Does TCEC take it into account in tablebase positions?
an hour ago
chatWING
  RocksEm (Stage 3 fan) 
The Black Bishop threatens the White Queen again! Who will win this battle of wills?
an hour ago
chatWING
  Ronald 
TCEC uses 5-piece tablebases for adjudication which are unaware of the 50-move rule
an hour ago

chatWING
  Warp 
So in theory it could adjudicate a win for a game that would actually be draw by 50-move-rule if let continue?
an hour ago
chatWING
  Ronald 
so if it ends KNNvKP, SF or Komodo might think it's a draw but still win (as white) or lose (as black)
an hour ago
chatWING
  101 
Really, Ronald?
an hour ago
chatWING
  Ronald 
yes, TCEC adjudicates using Gaviota TBs
SF and Komodo could be set up to ignore the 50-move rule in the TBs, but probably they are not
it is unlikely to become a problem, but it could happen
an hour ago
chatWING
  RocksEm (Stage 3 fan) 
The White Queen retreats again. Are we looking at a 3 fold repetition?
an hour ago
chatWING
  Arun 
K will avoid it and move something else
an hour ago
chatWING
  Mark Lefler (Komodo)  (VIP)
I would be surprised if TCEC did not take the 50 move rule into account. I believe the system uses a modified cutechess to play the games. We can ask Jeremy, who I am pretty sure did the mods.
an hour ago
chatWING
  Dragon Mist 
yes, 3-fold coming up
an hour ago
chatWING
  Warp 
How good are modern engines at endgames without tablebases? I think they were notoriously bad at it in the distant past (sometimes even losing won games due to weak endgame playing).
an hour ago
chatWING
  101 
Mark/Ronald: You might read the debate about it here: http://rybkaforum.net/cgi-bin/rybkaforum/topic_show.pl?tid=30630
an hour ago
chatWING
  RocksEm (Stage 3 fan) 
The audience seems divided here. We'll just have to wait and see...
an hour ago
chatWING
  Mark Lefler (Komodo)  (VIP)
Thanks 101. I will take a look soon.
an hour ago
chatWING
  Agreedy ½♥ (chare fan) 
Fish ready to clinch its umpteenth super final in a row!
an hour ago
chatWING
  Lee Sailer (I gree) 
When modern engines play out the endgame, they do fine, even against human GMs. The complaint is about how human GMs can tell at a glance what the plan is, while engines still have to do all the searh.
an hour ago
chatWING
  Warp 
This looks dangerously close to 3-fold repetition.
an hour ago
chatWING
  RocksEm (Stage 3 fan) 
And the Black bishop once again threatens the White Queen!
an hour ago
chatWING
  Agreedy ½♥ (chare fan) 
Two plies to 3fold
an hour ago
chatWING
  RocksEm (Stage 3 fan) 
Will the White Queen retreat to c2 again setting up a 3 fold repetition?
an hour ago
chatWING
  Jhonny 
for the force of rule TCEC, no matter how the game will be terminated. The eval from thoses engines is enought to determine the adjudication.
an hour ago
chatWING
  Ronald 
@Mark: TCEC takes into account the 50-move rule, but not for tablebase adjucation of 5-piece positions.
a KNNvKP drawn by the 50-move rule will be adjudicated as a win
an hour ago
chatWING
  Andrew Jacob 
wont it be adjudicated as a win before it reaches 50 moves?
44 minutes ago
chatWING
  Warp 
1 ply to 3-fold repetition.
44 minutes ago
chatWING
  Ronald 
I did not know about the KBBKN Gull-Hannibal ending
44 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
Yes, The White Queen moves to c2. Will the Black Bishop move to b3 and end this with a 3 fold repetition?
Will Komodo settle for a Draw?
43 minutes ago
chatWING
  Ronald 
KNNKP probably won't be adjudicated as a win before it is reached, unless both engines incorrectly believe it is completely lost for the KP
43 minutes ago
chatWING
  Agreedy ½♥ (chare fan) 
Hanni thinks it's better but black will avoid 3-fold... Silly
43 minutes ago
chatWING
  Warp 
We have Komodo's programmer here. Will it set for a draw by 3-fold repetition?-)
43 minutes ago
chatWING
  101 
The Gull-Hannibal ending was won even under the 50 moves rule.
42 minutes ago
chatWING
  Ronald 
Yes, but it might have ended in a draw
41 minutes ago
chatWING
  101 
But this is a serious flaw in the adjudication system. If, say, SF enters this endinig, believing it will get a draw, it will lose half a point.
41 minutes ago
chatWING
  Arun 
k will play rb5 or something else other than bb3
41 minutes ago
chatWING
  Ronald 
It would be better if TB adjudication was limited to "trivial" cases
(better in my personal view)
41 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
No one would blame Komodo for settling for a Draw with Black. Will he do it? Will his fighting spirit prevail??
40 minutes ago
chatWING
  101 
That is essentially what I am suggesting in the Rybka forum thread.
40 minutes ago
chatWING
  Ronald 
indeed
40 minutes ago
chatWING
  Arun 
the other thing is few engines dont support tb so how are we to say its a draw ....... we shud say its a draw only if both engines support tb
for eg in this case hannibal vs K ,....... hannibal can blunder with 5 pieces on board
40 minutes ago
chatWING
  LPZ 
SF team can simply ask that Syzygy50MoveRule be set to False.
38 minutes ago
chatWING
  Warp 
Is Komodo really taking over 6 minutes for this move, or is the GUI stuck again?
38 minutes ago
chatWING
  101 
LPZ: The problem is that the rules make no mention of this tournament not following FIDE chess rules.
38 minutes ago
chatWING
  SF TCEC 8 champ! 
Mark Lefler: Kramnik vs Deep Fritz game 6 (Brain in Bahrai 2002) was one excellent example of a Chess Fortress position: (FEN):
8/5k1p/5q2/8/1R6/6P1/5P2/6K1 w - -

Are you working with positions like the one above to develop an algorithm that will recognize a chess fortress? Do you think that the concept of a "fortress position" is too difficult a concept to teach to a chess program?
37 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
The tension mounts....
37 minutes ago
chatWING
  LPZ 
@101 True, the rules do not make this clear.
37 minutes ago
chatWING
  Agreedy ½♥ (chare fan) 
over!
36 minutes ago
chatWING
  Warp 
After 8 minutes it accepted the draw finally.
36 minutes ago
chatWING
  Dragon Mist 
draw ... bring on da fish!
36 minutes ago
chatWING
  101 
Arun: I agree. If there should be a 5 men adjudication for non-trivial cases, it should only apply when both engines have a functional 5 men tablebase implementation.
36 minutes ago
chatWING
  Terence Rattray 
wow quick game
36 minutes ago
chatWING
  LPZ 
"Cutechess will also adjudicate 5-men or less tablebase endgame positions automatically." Is there more?
36 minutes ago
chatWING
  physica ingocnito 
yyy eeeee sssss!!!!
36 minutes ago
chatWING
  Dragon Mist 
will Gull blunder with 9.f3 as SF did in first game?
35 minutes ago
chatWING
  Terence Rattray 
what is wrong with 9.f3?
35 minutes ago
chatWING
  Arun 
even if sf looses this it will still qualify mostly right ...... 2.5 points to bridge is too much with 4 games left
34 minutes ago
chatWING
  Joseph Ellis 
I thought SF won that game?
34 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
And for the third game in a row the battle ends in a 3 fold repetition. Komdoo settles for a Draw which will not clinch it places in the Superfinal. Perhaps Stockfish can beat Gull in the next game. If so it would guarantee a place for both Komodo and Stockfish in the Final.
34 minutes ago
chatWING
  Dragon Mist 
it allows early ..d5 and equality
SF won later, but played opening rather poorly
34 minutes ago
chatWING
  Agreedy ½♥ (chare fan) 
why isn't the archive getting updated?
34 minutes ago
chatWING
  Ronald 
@LPZ: setting Syzygy50MoveRule to false would indeed work here, because 6-piece tables are not used. If 6-piece TBs are used by the engines but not for adjudication, it gets even more messy.
33 minutes ago
chatWING
  LPZ 
Right
33 minutes ago
chatWING
  Sebastian Rodriguez 
what move prevent ... d5 now?
33 minutes ago
chatWING
  Jerry Olsen 
[teaching engine to recognize fortress] I wonder if it's necessary - perhaps the search would find fortresses naturally if the engine can recognize and correctly evaluate "unable to make progress" positions.
33 minutes ago
chatWING
  101 
LPZ: Of course, it might not be much of a real issue, since we must assume that Martin can/will manually overturn the result in favour of the correct one in the rare event that this happens.
32 minutes ago
chatWING
  Ronald 
a 6-piece 50-move draw could be drawn (if the 50 moves occur in the 6-piece phase) or won or lost (if the 50 moves occur in the 5-piece phase)
32 minutes ago
chatWING
  Dragon Mist 
both 0-0 and f4 are better now than f3
32 minutes ago
chatWING
  Ronald 
@101: I'm not so sure about that
32 minutes ago
chatWING
  Dragon Mist 
well done Gull
32 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
Gull lost this opening with Black against Stockfish. Can it turn the tables and win with White? We shall see.
32 minutes ago
chatWING
  Agreedy ½♥ (chare fan) 
Gull castled kingside
31 minutes ago
chatWING
  An Toshka 
agree f 3 is not really doing anything
31 minutes ago
chatWING
  Dragon Mist 
f3 is ok when you develop lsb to c4, not in this classical Dragon
31 minutes ago
chatWING
  LPZ 
FYI, lazy_smp just switched from O-O to ... f3 depth 27 and then back to O-O depth 28
31 minutes ago
chatWING
  n0brain 
@Ronald but doesnt syzygy contain 50move info?
31 minutes ago
chatWING
  Terence Rattray 
f3 usually supports g4-g5
31 minutes ago
chatWING
  101 
Can or will (want)? :)
31 minutes ago
chatWING
  An Toshka 
right dragon
31 minutes ago
chatWING
  Dragon Mist 
f3 coupled with Be2 are rarely a good idea
30 minutes ago
chatWING
  Agreedy ½♥ (chare fan) 
f3 prevented h5 Ng4?
30 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
Gull's first move is a King side castle. Gull looks for defense from the start against the very powerful Stockfish.
30 minutes ago
chatWING
  n0brain 
@Dragon its one of the main lines in the KID
29 minutes ago
chatWING
  Jerry Olsen 
This is Gulls best opportunity to finish in a not-too-distant 3rd place.
29 minutes ago
chatWING
  Mark Lefler (Komodo)  (VIP)
SF TCEC 8 champ: I have answered this too many times. Sorry but I am tired and not feeling well tonight
29 minutes ago
chatWING
  Dragon Mist 
in this opening you need f pawn to move to f4; how you prevent eventual Ng4 is another question
@nobrain KID has nothing to do with this opening
29 minutes ago
chatWING
  Barry Bague 
i like this particular opening selection
28 minutes ago
chatWING
  101 
Ronald: If Martin doesn't want to do it, TCEC will lose all credibility, since the adjudication will obviously work against the engines' inbuilt knowledge of the 50 moves rule. If he cannot do it technically, he can at least adjust the final table when the stage/superfinal is over.
28 minutes ago

chatWING
  n0brain 
@Dragon i know, but u spoke of f3 and Be2 in general...
28 minutes ago
chatWING
  Ronald 
@n0brain: yes it does. (but we are talking about TCEC adjudicating 5-piece positions with Gaviota TBs that are not aware of the 50-move rule)
27 minutes ago
chatWING
  Mark Lefler (Komodo)  (VIP)
Warp: Sure Komodo will take a 3 rep draw, if its search (and Contempt) could not find any higher scoring move.
27 minutes ago
chatWING
  Dragon Mist 
I did not, but even so, what KID has f3 and Be2 coupled with g4? Saemisch? not really ...
27 minutes ago
chatWING
  n0brain 
@Ronald ah, makes sence
27 minutes ago
chatWING
  Mark Lefler (Komodo)  (VIP)
Might I suggest we talk to Jeremy about the scheme used for calling a game?
26 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
This is an interesting opening for Gull-Stockfish. Let's hope it is not wasted like our last game which had a splendid opening and some lively action before ended in a 3 fold repetition.
26 minutes ago
chatWING
  Ronald 
@101: maybe, but one could also argue that rules are rules
(even when they deviate from FIDE rules)
26 minutes ago
chatWING
  n0brain 
@Dragon qote: "f3 coupled with Be2 are rarely a good idea",
that is pretty general and where do u get something with g4 from???
25 minutes ago
chatWING
  Dragon Mist 
you are right nobrain, I should have been more specific, thought that was clear I was speaking of sicilian setups
24 minutes ago
chatWING
  Nikolaos Konstantakis 
you go g4 with Bc4 usually
24 minutes ago
chatWING
  Ronald 
@101: hmm, the TCEC rules don't seem to mention TB adjudication...
24 minutes ago
chatWING
  n0brain 
noone hurt @Dragon :D
24 minutes ago
chatWING
  Dragon Mist 
that is what I am saying Nikolaos
f4 now the only move
23 minutes ago
chatWING
  Ronald 
@101: in fact they do: Cutechess will also adjudicate 5-men or less tablebase endgame positions automatically.
"or less"... interesting
I tend to agree that incorrect TB adjudications should be corrected.
22 minutes ago
chatWING
  Agreedy ½♥ (chare fan) 
for example when wind wipes the board
21 minutes ago
chatWING
  n0brain 
gn8 @all, need some sleep :D
21 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
Stockfish advances it's light square Bishop setting up a solid center.
21 minutes ago
chatWING
  101 
Ronald: Martin/Chessdom can do as they wish, of course, but it would make no sense to use the merely practical measure which adjudication is to "judge" a position using a different set of rules than the engines use.
21 minutes ago
chatWING
  Dragon Mist 
gn nobrain
21 minutes ago
chatWING
  SF TCEC 8 champ! 
Mark Lefler: Do you have a link on your web site for FAQs like the one I previously asked about ""Fortress positions"? If not, would you please consider posting a FAQ to answer questions like these?
21 minutes ago
chatWING
  Agreedy ½♥ (chare fan) 
swipes
21 minutes ago
chatWING
  Ronald 
@101: I think H4 uses Gaviota this season, so the engines don't agree on the rules either
20 minutes ago
chatWING
  Mark Lefler (Komodo)  (VIP)
I was thinking of using this image for Komodo. What do you think? http://postimg.org/image/3xc6nd7lx/
20 minutes ago
chatWING
  Eliot 
Too nice looking
20 minutes ago
chatWING
  Terence Rattray 
I like it
20 minutes ago
chatWING
  Knight Moves (#1 Fan of Gull) 
sexy
20 minutes ago
chatWING
  Ronald 
and correspondence chess has now switched to using Nalimov TBs to adjudicate 6-piece positions
20 minutes ago
chatWING
  Mark Lefler (Komodo)  (VIP)
SC TCEC 8 champ: now we have no FAQs on fortress positions, nor will I ever make one.
19 minutes ago
chatWING
  LPZ 
@101 Both Komodo and Stockfish can be configured to avoid entering into lost positions thinking they are 50-move draws.
19 minutes ago
chatWING
  Agreedy ½♥ (chare fan) 
Mark that looks very positional
19 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
Seems very friendly mark. We're you looking for friendly?
19 minutes ago
chatWING
  Mark Lefler (Komodo)  (VIP)
There is something about that eye.
19 minutes ago
chatWING
  101 
H4 uses Nalimov, but I guess there is no DTZ/DTC information there which H can retrieve.
19 minutes ago
chatWING
  Agreedy ½♥ (chare fan) 
yes, the flashlight
18 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
it looks like it just ate something yummy.
18 minutes ago
chatWING
  101 
LPZ: I know that, and that is useful for correspondence chess.
18 minutes ago
chatWING
  Mark Lefler (Komodo)  (VIP)
Some night I just get tired of answering questions over and over. This has been a rough day. I will try and be nicer tomorrow.
18 minutes ago
chatWING
  Jhonny 
correct TB adjudacation in the end of stage are no good idea because the time usage and because the adjudication no allow the correction in the future (after all)
18 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
Like Bobby Brown said, that's your prerogative Mark.
17 minutes ago
chatWING
  101 
Or indeed if you are just curious if there is a possible different "parallell" evaluation of a >7 or >6 men position.
17 minutes ago
chatWING
  LPZ 
@101 It would also prevent what you are worrying about here, at least for the Superfinal.
17 minutes ago
chatWING
  Arun 
mark is the most nice person in the chat ...... just leave him if he is not well ..... he need not answer to any1
17 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
Stockfish starts out with a 0.00 evaluation. I hope the web site doesn't crash again.
15 minutes ago
chatWING
  LPZ 
depth 33 f4 +0.03
15 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
Gull advances it's F pawn 2 places menacingly....
15 minutes ago
chatWING
  101 
LPZ: I It would, but I think it would be better if Martin manually overturned the adjudication (i.e. adjudicated the game himself) in the rare event that this happened. If not, he would have to mention that the superfinal would not use the 50 moves rule for >=5 men endings.
*<=
14 minutes ago
chatWING
  Jhonny 
the pv for both engines are quit to advance of pawl f4 and g4
13 minutes ago
chatWING
  Joseph Ellis 
yes, there is no reason a little common sense can not be applied
13 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
Stockfish moves it Rook off the a file in anticipation of a King side attack.
12 minutes ago
chatWING
  Lee Sailer (I gree) 
Common sense is to follow the rules.
12 minutes ago
chatWING
  LPZ 
The rules are not clear. That should be corrected.
11 minutes ago
chatWING
  An Toshka 
el cool magnifico vs el cool fantastico
11 minutes ago
chatWING
  Garrett 
Gull got an unfair half point in game 12 from adjudication - it was adjudicated as a win, and while technically true, there is absolutely no way Gull would have actually been able to win the BBvN ending
10 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
Stockfish seem to be settling into a defensive position. Is a Draw enough?
10 minutes ago
chatWING
  Martin Thoresen  (Operator)
People wanted TBs, people got TBs. Good night folks.
10 minutes ago

chatWING
  101 
Martin: Have you understood the issue?
9 minutes ago
chatWING
  Pablo Rouca 
stop whinning with the lost on time Garrett
9 minutes ago
chatWING
  Terence Rattray 
I am sure he has but why complicate things?
9 minutes ago
chatWING
  Garrett 
that was not about loss of time - that was Gull's game against Hannibal!
9 minutes ago
chatWING
  101 
I am happy that there are tablebases, and surely, it is not a very big problem. :)
8 minutes ago
chatWING
  RocksEm (Stage 3 fan) 
GN Martin. Thanks for all your hard work today. BTW, I don't think table bases should be used in a game for anything exception adjudication.
8 minutes ago
chatWING
  Nikolaos Konstantakis 
BBvN is very easy for gull, or any engine to win without tb
even on single core
7 minutes ago
chatWING
  Garrett 
Nikolaos, it is not - set it up and let Gull try. Gull cannot even win KNBvK
7 minutes ago
chatWING
  MINDBREAKER 
Very happy with 5-man on RAM
7 minutes ago
chatWING
  Lee Sailer (I gree) 
Nothing pleases everyone
7 minutes ago
chatWING
  MINDBREAKER 
Good decision
7 minutes ago
chatWING
  Garrett 
I did this experiment, Gull has terrible difficulties trying to do this, and ultimately fails both
7 minutes ago
chatWING
  101 
Whether or not to have engines access tablebases, and this tiny issue with regards to the rules and tablebase adjudication are separate issues.
6 minutes ago
chatWING
  Garrett 
absolutely. I would say let engines use TBs, and only adjudicate drawn positions with endgames. if it it a winning endgame, let the engines prove it
5 minutes ago
- By Dr.X (Gold) Date 2015-10-26 16:17

>Garrett  absolutely. I would say let engines use TBs, and only adjudicate drawn positions with endgames. if it it a winning endgame, let the engines prove it


A Reversal in the Kerouac notion of first thought best thought. In this case last thought best thought!  :wink:  Upvote on this idea!
- - By 101 (*) Date 2015-11-12 14:27 Edited 2015-11-12 15:00
Modified adjudication rules for the next superfinal
101 
This ridiculous amount of draws actually makes me become in favour of, in superfinals, a draw adjudication rule not ignoring captures and pawn moves, and starting from move one. If that means some games are drawn which would have been decicisive, so be it, because in the vast majority of cases, they should be drawn, when both Komodo and SF agrees. This could probably reduce the length of the superfinals several times.
22 minutes ago

Permanent Brain 
Exactly my words 101! I have proposed that yesterday, for season 9.
22 minutes ago

AlbinCGambit 
This large amount of draws was expected before the superfinal started.
22 minutes ago
chatWING

  Stock Overfish (Isaac, fanatic o... 
+1 albin
21 minutes ago

Permanent Brain 
Akthough, I disagree as for starting so early. Engines should do, let's say 35 moves minimum.
22 minutes ago

101 
PB: If both engines show that evaluations for ten moves before move 35, something is wrong with the opening, and we can safely move on to the next game.
21 minutes ago

Santiago Méndez 
Spectators should understand that this is what happens at this level of chess.
20 minutes ago

101 
Santiago: Yes, but to what extent? 50 percent of superfinal 7 and game 1-20 of superfinal 8 were theoretically critical lines as determined by me and the Aquarium book, and the win rate in those games was the double of the remaining games.
18 minutes ago

sasuke2690 
anyways, its painful to see these games continued after they've reached 0.00 so long back
2 minutes ago

On my last point, I will make a separate thread in this forum.
Parent - - By APassionforCriminalJustice (***) [ca] Date 2015-11-13 08:45
Just choose sharper, more aggressive lines. There are still very complex openings that can fool the engines like the Semi-Slav Defense variation consisting of dxc4. Draw adjudications seems fine of course - but the openings thus far have been pretty much crap. We cannot have boring, pawn-symmetrical, and overly positional positions - and then expect fun, exciting, and downright awesome chess. It feels like we are watching mechanical Capablancas going at each other.
Parent - By Venator (Silver) [nl] Date 2015-11-13 19:35
There are still very complex openings that can fool the engines like the Semi-Slav Defense variation consisting of dxc4.

You mean the Semi-Slav Botwinnik variation? Isn't that analysed up to a draw by now?

Or do you mean the Anti-Moscow gambit?
- By 101 (*) Date 2015-11-17 09:42 Edited 2015-11-17 09:50
Marker of divergence from previous PV
Martin Thoresen has already implemented a nice feature which which marks (with a @ sign) where the two engines' last PVs diverge. I have often also wanted to see if one engine discovered any new moves compared to its last PV. Therefore, I suggest to mark this as well, for instance with the ¤ sign (which is almost never used, but nonetheless has a prominent place on many keyboards).

Two suggestions from parallell thread
A user has submitted two new suggestions on OpenChess, one of which is an alternative to the one suggested by me regarding adjudication.
See "Suggestion thread for TCEC" in the TCEC Season 8 subforum of OpenChess: http://open-chess.org/viewtopic.php?f=43&t=2896&p=22409.
Up Topic The Rybka Lounge / Computer Chess / Suggestion thread for TCEC

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill