CCRL Discussion Board

Its drawback, as one might easily imagine, is that it may sometimes give very large numbers who far exceed what might be achievable with more optimal play. But I don’t see for now how it could be possible to do better within the requirements quoted before. What you are describing is a table that gi...

Perhaps the subject of how you might have done (or did) DTMk EGTs is academic, but – given that sz has insisted that DTM50 EGTs should not walk users into 50-move-draw-claims – I now believe that it is possible to produce such DTM50 EGTs. Wrong. At least, Nalimov-type DTM50 tables will run into pro...

To go back to something you wrote earlier: - the DTZ50-minimaxing lines are not necessarily followed, which means that - some DTM50-minimaxing lines might be subject to a 50-move draw claim. This is incorrect. The fact that DTZ50-lines are not necessarily followed does not mean that DTM50-minimaxing...

I'm quite familiar with the fact that lines that minimax DTM without any k-move rule are not necessarily lines that minimax DTM in the context of a k-move rule. Do you agree that DTM50 should refer to DTM in the context of the 50-move rule? If yes, then how can a DTM50-line ever walk into a 50-move...

Well, if I have got the definition of mb's DTM50 metric wrong, I hope he will step in and correct me. Yes, I agree that if a position is decisive under a 50-move rule, there is a line that minimaxes DTM. Yes, and now it is important to realize that lines that minimax DTM in "chess without any ...

I don't understand why anyone would call a table "DTM50" if by design it does not properly take into account the 50-move rule. DTM give shortest mate against any play by the losing side, without any k-move rule. DTM50 gives shortest mate against any play by the losing side, with the 50-mov...

Given mb's track EGT-record which is impeccable, I would be very reluctant to say that anything associated with him is 'broken'. Admittedly, the DTM50 EGT 'EM50' is only a partial response to the 50-move rule, but that does not mean it is 'broken' if it does what it claims to do. First of all, I do...

My guess is that first and foremost, h.g.muller is defining a chess-like game without captures and promotions, so that he can first develop and test a (somewhat simplified) version of his algorithm, and so that he does not have to generate subtables first. Certainly a nice idea. The intuition about ...

I would like to understand more clearly what people have in mind when they say 'make the move-counter part of the index', just for the avoidance of misunderstanding. Do they mean, in effect 'extend the FEN to include the move-counter'? I suspect so. This would appear to multiply the number of posit...

Attached are ...-Marc Bonjour Marc, A friend of mine (in France) has produced the Nalimov-like DTM50 files (giving legal mates by the 50-move rule) for knnkp (and knnkq). Have you (or friend) produced something like that? We would be interested in comparing :) Note that in the general case a DTM50-...

The problem is that there is no solid answer, because fallible opponents are fallible in different ways Good luck teaching the computer to read minds What you can do is to have the information required to win positions that can be won when ignoring the 50-move rule. Even if you don't care too much ...

I still don't see what the point of using anything other then DTZ50 is DTZ50 is sufficient for perfect play, and it is also the smallest possible tablebase that has this property I fully agree that DTZ50 is all you need for perfect play. When playing an imperfect opponent some extra information mig...

Yes, this is one of the scenarios where DTR goes wrong. Another scenario is when both sides are playing according to the DTR-table and the table leads to the winning side playing a move with lower DTR than the winning side can actually achieve given the current move-counter. The problem is (again) t...

Before diving in too deeply, it is probably good to first explain the basic concept. A tablebase doesn't store a list of (all) positions with for each position the best move, but it assigns to each position an index value 0 ... N-1 and stores an array of N values (one value of each index value 0 ......

... But I doubt it will on the average give better results than doing a normal alpha-beta search and relying on the hash table. Oh yes, it will. Ah, I'm a bit surprised that that last sentence of mine is still there. I thought I had removed it after deciding that indeed it should help in some posit...

In its most simple form it would only work for totaly won (for wtm) Pawn slices. This seems a severe restriction, but in fact I believe it to already be extremely useful in practice, as it will cut off the path to successor phases with multiple queenings. It is to be expected that having a single (...

I think I get the idea now. With many pawns on the board, the slices become quite small (and many). It might be possible to a kind of minimax search on the slices which would allow some branches to be cut. This will probably work to some extent if the relevant slices are generally won for one side. ...

First I "bitpack" all positions. Second I compress the bitpacked data with a general data-compression-algorithm, which I got from internet. I work with Delphi and the unit is called ZlibEx. Have you tried without bitpacking? I would not be surprised if that gives better results (especiall...

Of course the non-sensical P-slices could be of interest to problemists and such, so eventually someone will have to generate them. But it would be an enormous advantage to be able to calculate the sensible 99% of P-slices of KPPPPKP without first having to do KQQQKQQ, KQQRQKQQ, KQQQKQR, ... And I ...

So the matter of ordering of the different Pawn constellations will simply never crop up. In a way it does crop up, because you're not fully free in the order you generate pawn slice files. I don't think ordering pawn slices is very difficult. Make sure lower "pawn values" are closer to p...

szyzygy - do you have a source/reference for that neat formula for 'Q' which enumerates the positions of like men? I came up with it myself, so I don't have a reference. The general formula works with binomials. Let Bin(n, k) denote the number of k-element subsets of an n-element set. Then Bin(n, k...

For kqqk, the indexing scheme can be further improved by removing the symmetry caused by having two pieces of the same kind. Each positions k,q1,q2,k corresponds to a position k,q2,q1,k. To get rid of this symmetry, only index pairs q1,q2 (of squares) with 0 <= q1 < q2 <= 63. There are 64*64 pairs (...

On the other hand, there is a compelling case that castling should reset the 50 move counter. I assume that the only reason that they didn't put it in the rules is that it never comes up in actual play. I agree that a stronger case can be made for letting castling reset the 50-move counter. But I d...

Tfy explanation. It never crossed my mind that someone would create an indexing-scheme and EGT to accommodate e.p.-positions (necessary if they are to be taken into consideration) and then not include them in the published EGT. Well, you haven't understood. It is not necessary to index them. Please...

Incidentally, I don't understand how we can have EGTs which are 'correct in the context of the e.p.-rule' without holding positions where there is an e.p.-capture opportunity. Construct an EGT with e.p.-positions and throw out the e.p.-positions (leaving the values for all other positions unchanged...