Why is this forum so quiet?

[vb4]

When I first came across this forum I thought it was a great place to talk about egtb's. Only 4 posts in the last month is disappointing since there are so many interesting things to learn about eg's. I know that we have finished the 5-1 sets and this is a very strong interest of mine. Unfortunately the statistics that are generated for these are of no use to me since they are not in a similar format to Eugenes output. At one point there was some general talk about somenone maybe writing a routine to rectify this but havent seen anyone getting involved. Perhaps if we can all discuss some viable approaches to extracting the correct statistics maybe from the egtb's we could find a way of easily doing it but with no discussions at all its a much tougher road. I have been studying egtb's for many years and really need the statistics for the 5-1 sets to finish my study.

I also remember at one point someone had started generating sets for the 7 piece egtb. Can anyone tell me the status of that project, whose involved and is their a web site we can go to see how things are going??

Anyway I guess I just want to see tis forum pick up some speed. There are alot of you guys out here much more talented then me with some great ideas.

Lets keep it rolling,

Les

[h.g.muller]

Indeed. I only discovered this forum recently, but am also a little bit concerned by the low level of posting frequency. On Talkchess.com there is generally more response, even on a specialist subject like EGTBs.

[guyhaw]

It's quite reasonable that this b'board should be quiet for the moment if people don't think there is much to discuss at the moment.
Eiko Bleicher and I have completed some work on the 6-man maxDTM positions.
Eiko has also surveyed all his 6-man EGTs for the zugzwang positions of Types 1-3 and most of the 'new' types 4-6. I am doing some post-processing work on his results now - finding maxDTM mzugs of each type for example.
We may also put the 6-man P-less EGTs found to the DTC EGTs which exist with John Tamplin - to find the maxDTC mzugs.
Thinking about it, the positions which are most 'zugzwangy' if I can put it like that are the positions featuring the minimum DTM wins, so maybe we give the minDTM and min DTC mzugs as well

With regard to the 5-1 EGTs, there is no point persisting in existing for non-Nalimov statistics in Nalimov format. FEG stats are likely not to agree with Nalimov stats on P-less endgames because they treat the case of "two Kings on the long diagonal" differently.
g

[vb4]

It's quite reasonable that this b'board should be quiet for the moment if people don't think there is much to discuss at the moment.
Eiko Bleicher and I have completed some work on the 6-man maxDTM positions.
Eiko has also surveyed all his 6-man EGTs for the zugzwang positions of Types 1-3 and most of the 'new' types 4-6. I am doing some post-processing work on his results now - finding maxDTM mzugs of each type for example.
We may also put the 6-man P-less EGTs found to the DTC EGTs which exist with John Tamplin - to find the maxDTC mzugs.
Thinking about it, the positions which are most 'zugzwangy' if I can put it like that are the positions featuring the minimum DTM wins, so maybe we give the minDTM and min DTC mzugs as well

With regard to the 5-1 EGTs, there is no point persisting in existing for non-Nalimov statistics in Nalimov format. FEG stats are likely not to agree with Nalimov stats on P-less endgames because they treat the case of "two Kings on the long diagonal" differently.
g

Hi Guyhaw,

I understand the difference to some degree. I guess what I am interested in knowing is there any way of extracting the stats from the already generated egtb's?? Since we know have all of the 5-1 sets done can you or anyone out here think of a way of extracting the stats. I am not an expert on egtb's but I would be willing to give it a shot if it is not to complex <s>. I have been studying all prior stats from all the other sets and finding some interesting things. My thought is if we have the 5-1 sets and they contain all perfect knowledge then there may or should be a way of arriving at the stats.

Thanks,

Les

[Kirill Kryukov]

I think this forum is quiet because 6-men tablebase distribution problem is solved now (3-3 and 4-2 sets), and 5-1 and 7-men bases are not available in any practical form yet. "Practical" means that the tablebase is available as well as some engine able to use it.

Also most of us are probably just busy with other things (at least I am).

[guyhaw]

The (5-1) FEG (DTM) EGTs are not as accessible as the Nalimov ones for three reasons. First, they are only accessible by CHESSMASTER. Secondly, there is no ancillary access process analogous to TBGEN. Thirdly, the index-scheme was good for the EGT-generation phase but is bad for repeated access in a forward-search process 'otb': so linear access would be ok but random access would be relatively poor.

There are two ways forward to get to a Nalimov-standard set of 5-1 statistics. One is to extend TBGEN to generate 5-1 DTM EGTs directly. The second is to convert from FEG form to Nalimov form, something Marc B has done on an 'as needed' basis for the 'last 16' of the 3-3p endgames.

Marc B must have generated a few 5-1 EGTs, probably to the DTC metric rather than DTM, because he's generated the likes of KBBBBKQ.

I can't think that there's too much that is interesting about 5-1 EGTs: perhaps the major-side-to-move draws! They are needed for completeness, sure. They are the most obvious example of why one should generate to the DTZ (Depth to move-count--Zeroing move) metric: the EGTs would compress to much smaller sizes.

Marc hasn't said anything about his 7-man EGT work since the announcement about a new maxDTC (KRBNKQN, 8/1R6/8/6N1/5K2/1B6/3k3n/7q w) record at 517 moves. I think he's aiming towards the ~KQPPKQP and ~KRPPKRP EGTs, the '~' indicating that P-conversions are assumed to be P=Q, increasingly not true as the initilal P-positions become distant from the last rank.
g

[vb4]

The (5-1) FEG (DTM) EGTs are not as accessible as the Nalimov ones for three reasons. First, they are only accessible by CHESSMASTER. Secondly, there is no ancillary access process analogous to TBGEN. Thirdly, the index-scheme was good for the EGT-generation phase but is bad for repeated access in a forward-search process 'otb': so linear access would be ok but random access would be relatively poor.

There are two ways forward to get to a Nalimov-standard set of 5-1 statistics. One is to extend TBGEN to generate 5-1 DTM EGTs directly. The second is to convert from FEG form to Nalimov form, something Marc B has done on an 'as needed' basis for the 'last 16' of the 3-3p endgames.

Marc B must have generated a few 5-1 EGTs, probably to the DTC metric rather than DTM, because he's generated the likes of KBBBBKQ.

I can't think that there's too much that is interesting about 5-1 EGTs: perhaps the major-side-to-move draws! They are needed for completeness, sure. They are the most obvious example of why one should generate to the DTZ (Depth to move-count--Zeroing move) metric: the EGTs would compress to much smaller sizes.

Marc hasn't said anything about his 7-man EGT work since the announcement about a new maxDTC (KRBNKQN, 8/1R6/8/6N1/5K2/1B6/3k3n/7q w) record at 517 moves. I think he's aiming towards the ~KQPPKQP and ~KRPPKRP EGTs, the '~' indicating that P-conversions are assumed to be P=Q, increasingly not true as the initilal P-positions become distant from the last rank.
g

Hi Guyhaw,

Thank you for the info. Listen you seem to know much more about this stuff then I do certainly. Listen is it a major problem for me to run Eugenes generator for the 5-1 egtb's? I thought I read somewhere a while back that the code would have to be slightly modified! Is that true and if so do you think someone here in the forum could do it? I will try to run them if someone can supply me with the generator.

Just a thought,

Les

[guyhaw]

I haven't seen the Nalimov code lately, but imagine it does have to be modified for 5-1 EGTs. The index-ranges have to be defined somewhere (for some reason that's not clear to me as they can be worked out at runtime). Also, whenever one pushes a code into diffferent territory, if it not generic enough, one will discover problems with it: that's actually a definition of 'generic enough'
I'm sure Marc B can produce an appropriate version of TBGEN if necessary, and I think he helped Eugene N with the indexing before. Not sure if MB is still visiting this board.
g

[vb4]

I haven't seen the Nalimov code lately, but imagine it does have to be modified for 5-1 EGTs. The index-ranges have to be defined somewhere (for some reason that's not clear to me as they can be worked out at runtime). Also, whenever one pushes a code into diffferent territory, if it not generic enough, one will discover problems with it: that's actually a definition of 'generic enough'
I'm sure Marc B can produce an appropriate version of TBGEN if necessary, and I think he helped Eugene N with the indexing before. Not sure if MB is still visiting this board.
g

Hi Guyhaw,

Listen if you or anyone out here knows his email address or how I can reach him give it to me and I will see if we can get things going. I am sure if he can do the necessary changes since he is familiar withthe code I am sure I can get this done in terms of generating them.

Thanks,

Les

[guyhaw]

Actually, Les, I am listening - but as a matter of standard protocol and courtesy to their owners, I don't pass on email addresses.
Neither am I interested in 5-1 endgames enough at this time to make a request to Marc: he has far better things, even chess EGT things, to do with his time. Apologies if that disappoints.
I think a more interesting challenge for both you and me would be to have a look at the Nalimov source and understand it enough to extend it to 5-1 EGTs if that is necessary.
Despite being interested in chess EGTs as a vehicle for illustrating aspects of algorithms and data-mining, and despite describing Eugene's algorithm retrospectively with EN and Ernst Heinz, I've never been totally au fait with EN's code.
Martin Kreuzer has given a lead on 5-1 EGTs, and I'm happy to await his return and his further results. Your only issue would seem to be with the statistics for P-less endgames as the statistics for P-ful endgames would be the same as Nalimov's. Anyway, you haven't said why you have a problem with the stats you have in the first place.
g

[h.g.muller]

Well, I am back from a great vacation, so you will hear from me soon. I came back with lots of ideas. First I had to implement those concerning my engine Joker, though, which is competing in OpenWar2, and coud use some improvement...

Now that is done, I am going to code my latest EGTB algorithm for two test cases: In-RAM generation of KQKR without using symmetry (which I consider a model problem for a P-slice of other end games with 4 non-Pawns). My aim is to get the time needed to build it below one second.

The second model problem will be building KQKBN (initially also symmetry-less), making use of the harddisk, for testing the leapfrog algorithm with alternating scans in the white K and the white Q direction. By replacing the scans over individual white pieces by scans over pairs of pieces, this would then allow building 4-3 7-men EGTBs without significant modification.

[vb4]

Actually, Les, I am listening - but as a matter of standard protocol and courtesy to their owners, I don't pass on email addresses.
Neither am I interested in 5-1 endgames enough at this time to make a request to Marc: he has far better things, even chess EGT things, to do with his time. Apologies if that disappoints.
I think a more interesting challenge for both you and me would be to have a look at the Nalimov source and understand it enough to extend it to 5-1 EGTs if that is necessary.
Despite being interested in chess EGTs as a vehicle for illustrating aspects of algorithms and data-mining, and despite describing Eugene's algorithm retrospectively with EN and Ernst Heinz, I've never been totally au fait with EN's code.
Martin Kreuzer has given a lead on 5-1 EGTs, and I'm happy to await his return and his further results. Your only issue would seem to be with the statistics for P-less endgames as the statistics for P-ful endgames would be the same as Nalimov's. Anyway, you haven't said why you have a problem with the stats you have in the first place.
g

Guyhaw I am not quite sure in your last 2 sentences. Are yousaying that Martins 5-1 stats for endgames with pawns are the same as Nalimov's and that the difference only exists with the 5-1 stats for egtb's that dont have any pawns?? I am confused so please explain. regarding having problems with the stats I have in the first place I assume you are referring to the 5-1 sets. That being the case I have over the years been studying all the egtb's and since I have all the sets completely except for the 5-1 I am not able to move forward in my study. Since there has been talk out here regarding the inaccuracies of the 5-1 stats I didnt involve them in my study in hope that someone would find a way of making the stats agree with Eugenes. Since all prior work I have been studying has been in his format its a bit late to change gears at this time. And finally in closing I get the feeling that you ressnt some of my posts and I am not quite sure why. I didnt ask if you were interested in 5-1 endgames all I asked was if you or anyone out here can touch base with Marc to see if he could do something. I dont care if you give me his email or not as long as we can have someone try to touch base with him. I am sure Marc is an adult and doesnt need you to make decisions for him as to whether he has far better things to work on. Let him amke that call not one of us. I assure you that I too am very busy but always find time to post out here. Put your self in my shoes guyhaw, you have been doing a study over many years and all you need to move forward in the study is one last bit of information I dont think you would just chalk it up without trying to resolve things.

Anyway my $.02 and atleast thanks for your response,

Les

[guyhaw]

Apart from the format, the difference between Nalimov and FEG stats is confined only to the P-less endgames. When there are Pawns on the board, the results should agree.
The reason has been explained before. It is that that, when both Kings are on the a1-h8 diagonal, Nalimov will count both the position and its reflection in the a1-h8 'mirror' in his statistics. FEG I think does not [and Wirth's statistics before did not]. So, these _differences_ are not _errors_ but just different ways of dealing with the symmetries involved.
One can confirm this by comparing EN and FEG stats on say KQKR, KPKP, KRBKQ and KRPKQ.
I believe that both EN and FEG count in winner's moves to mate. I hope FEG takes 'e.p.' into account as well: EN does.

g

[h.g.muller]

Surely it wouldn't be so difficult to extract the required statstics if you have the corresponding TB? If you have one kind of statistics you can even greatly speedup the process by just probing for the difference. Only 21 out of the 462 King-pair positions have both Kings on the dagonal, so you would only have to probe 4.5% of the TB.

[guyhaw]

... a bit messy, but it works.
In fact, Eiko Bleicher has created zugzwang data from Nalimov EGTs and I'm scanning it with his software to identify the adjustments needed for the 'two Kings on a1-h8' syndrome. Same idea but in reverse.

Still got the KT paper to scan for you, hgm. I'll put it in the thread where I mentioned the paper.
g

[jkominek]

Hi,

It has been a while since poking my head in this forum, and I find myself amused by the off-topic exchange between (eager and frustrated) vb4 and (the tad haughty) guyhaw.

Actually, vb4, I'm on sufficiently familiar terms with tbgen to perform the extension. I'm interested in doing this, at some point. It's just a matter of available time and resources. Like many people, when faced with the question of "should I sink many tens of hours into a hobby, or get paid for what I do?" my answer usually sides with the paycheck. Plus, there doesn't seem to be a driving interest in 5-1 tablebases. (Present company excluded.)

Here's the skinny. Assuming a machine of sufficient size, generating 7-man 4-3 tables is trivially easy. Generating 5-1 tables requires devising an indexing scheme for this configuration, as such a thing does not currently exist. The new 5-1 scheme would need to be "in the spirit" of the existing EN indexing, then extensively tested. It would have to be made available to engine authors, and receive affirmation. If 5-1 tablebases were somehow dropped from the sky (yes, you only want the statistics, I know), no chess playing program could make use of them. Niether could Wilhelm.

So you see, what appears to be a simple request in fact ensnares the programmer in a significant chunk of work, for little more than a "thank you" and a small bit of chess immortality. Eventually, I surmise, that wasn't enough to keep Eugene going.

john

[vb4]

Hi,

It has been a while since poking my head in this forum, and I find myself amused by the off-topic exchange between (eager and frustrated) vb4 and (the tad haughty) guyhaw.

Actually, vb4, I'm on sufficiently familiar terms with tbgen to perform the extension. I'm interested in doing this, at some point. It's just a matter of available time and resources. Like many people, when faced with the question of "should I sink many tens of hours into a hobby, or get paid for what I do?" my answer usually sides with the paycheck. Plus, there doesn't seem to be a driving interest in 5-1 tablebases. (Present company excluded.)

Here's the skinny. Assuming a machine of sufficient size, generating 7-man 4-3 tables is trivially easy. Generating 5-1 tables requires devising an indexing scheme for this configuration, as such a thing does not currently exist. The new 5-1 scheme would need to be "in the spirit" of the existing EN indexing, then extensively tested. It would have to be made available to engine authors, and receive affirmation. If 5-1 tablebases were somehow dropped from the sky (yes, you only want the statistics, I know), no chess playing program could make use of them. Niether could Wilhelm.

So you see, what appears to be a simple request in fact ensnares the programmer in a significant chunk of work, for little more than a "thank you" and a small bit of chess immortality. Eventually, I surmise, that wasn't enough to keep Eugene going.

john

Hi John,

First thanks for understanding my position. I am not as frustrated as you may think but disappointed perhaps. I do understand all you have said and I know for sure there is really no rewards other then a little tiny itsy bit of fame <S>. I know this all to well. I have been working on a scheme that will allow me to store alot of chess positions in a very small amount of space. I have been playing with it for several years now and it grows constantly and has a mind of its own. My hope here was to see if some person much more talente and familiar with egtb code could implement it into the existing generator. I had no idea it was so involved and I do want to apologize to all that might have thought I was being inconsiderate. Quite the contrary I have a great deal of repsect for people.

John I thank you for your approach in addressing this thread and I must say you have a magnificint way with words.

Take care,

Les

[guyhaw]

I go along with almost (!) everything jk says.
The principles of the Nalimov indexing-system are described in a Nalimov/Haworth/Heinz ICGA_J paper (v23.3 pp148-162) and can be extended to 5-1 endgames.
However, taking the code into 'new territory' might identify new challenges for the code if it is not completely portable to all endgames. Thinking about it, the new challenge would be dealing with 'Four Like men': I have no idea whether EN's code can deal with that or not. This would affect the computation of KxxxxK where x = Q/R/B/N: it is not necessary to computer KPPPPK for stats purposes as the FEN stats can be transformed into EN form when there is a Pawn present.

I don't wish to prolong the discussion of e-protocol, but I'm respecting MB's e-silence and assuming that he's taking a holiday from chess EGT work: life moves on for all of us. If he starts an e-dialogue on same, I'll raise the 5-1 issue 'en passant'.
g

[vb4]

I go along with almost (!) everything jk says.
The principles of the Nalimov indexing-system are described in a Nalimov/Haworth/Heinz ICGA_J paper (v23.3 pp148-162) and can be extended to 5-1 endgames.
However, taking the code into 'new territory' might identify new challenges for the code if it is not completely portable to all endgames. Thinking about it, the new challenge would be dealing with 'Four Like men': I have no idea whether EN's code can deal with that or not. This would affect the computation of KxxxxK where x = Q/R/B/N: it is not necessary to computer KPPPPK for stats purposes as the FEN stats can be transformed into EN form when there is a Pawn present.

I don't wish to prolong the discussion of e-protocol, but I'm respecting MB's e-silence and assuming that he's taking a holiday from chess EGT work: life moves on for all of us. If he starts an e-dialogue on same, I'll raise the 5-1 issue 'en passant'.
g

Hi Guyhaw,

Ok I see the potential for complications which I didnt understand previously. Ok lets maybe drop the thread for the time being and hope someday someone will pick up where we left off. Take care and thx for your patience.

Les

[guyhaw]

Fine. As I forgot to mention this in a previous post ...
the papers on Nalimov's indexing approach are accessible via http://www.tinyurl.com/law6k .
g