CCRL Discussion Board

Greetings,

It seems all of my problems have been due to a messed up g++ install. I was able to compile it on another machine here.

I've tried the examples, but I still get "Illegal tablebase name: 0s"
Anyone have an example command-line? Would like to start 5-1 egtb's

-Josh

I'm glad you succeeded to compile the generator! I did not try to compile or run it here, yet. I'm sure Martin will help you out. Good luck with 5-1!

You can't calculate 5+1 EGTBs with this code. I had only a short look on it till now but there are only defines for 4+2 and 3+3 EGTBs. The code has to be modified. During my holidays within the next three weeks I will try to understand the code and what to do to calculate the missing 5+1 tables.

I'm using the Microsoft VC++ compiler so I can't help you with your compiler problems.

The 'sticking point' is having the right code to generate 5-1 index-schemes, including code for '(4) men of a kind'.

I don't know how generic Nalimov's 'like men' code is.

Marc B is the expert on indexing-schemes as on all other things, and has I believe collaborated with Eugene Nalimov on the 3-3p index-scheme. he understands the indexing regimes of Nalimov, FEG and Konoval, and may have a generic index-scheme in his GTBGEN code.

So my recommendation is to wait for Marc B to advise us. Oh, my other one is to generate 5-1(p) EGTs to DTZ, not DTM.

g

guyhaw wrote:The 'sticking point' is having the right code to generate 5-1 index-schemes, including code for '(4) men of a kind'.

I don't know how generic Nalimov's 'like men' code is.

Marc B is the expert on indexing-schemes as on all other things, and has I believe collaborated with Eugene Nalimov on the 3-3p index-scheme. he understands the indexing regimes of Nalimov, FEG and Konoval, and may have a generic index-scheme in his GTBGEN code.

So my recommendation is to wait for Marc B to advise us. Oh, my other one is to generate 5-1(p) EGTs to DTZ, not DTM.

g

I had understood that EN would use this formula to trasform positions of n like men into a subindex:

subindex = X1 + (X2, 2) + (X3, 3) + ... + (Xn, n)

where X1, X2, X3, ... , Xn are the positions of the n like men from 0 to 63 with the condition X1 < X2 < X3 < ... < Xn, while (X2, 2), (X3, 3), ... , (Xn, n) are binomial coefficients, with the
convention that (a, b) = 0 if a < b.

But this formula would preclude the possibility of discarding from the file the positions where a check is given by a man close to the opponent king.

A little more complicate and less fast is the inversion process, which corresponds to solve a
diophantine equation in N integer variables.

guido

... replying to guido

To understand the subtlety of Nalimov's index-scheme, the best plan is to read the 'Nalimov, Haworth, Heinz' paper on the subject. It's in a past ICCA_J: see the ICGA website at http://www.icga.com

Nalimov does not place a stm-man on a square where it can give an unblockable check. So some squares, including those the Ks are on, are not available to the stm-men. Thus the 64 squares are numbered with some knowledge used of what has been placed before.

Even so, when all men are placed, the sntm-K may be in check, in which case the index-position is set to broken.

I always have to concentrate very hard to think through the general 'like men' formula even though I've understood it at on at least three occasions and written it up once. It's in the paper above (which is also on my "Guy Haworth" website at Reading Univ.

g