Gaviota 0.74.6 (special for EGTB builders)

[michiguel]

Ray seems to have a weird problem. I do not know whether is related to the bug I just fixed. It is possible. It is a pity because you are so close to finish!

Miguel

Now all finished !

I don't know what was different this time - I used Vista instead of W7, but that shouldn't have mattered

Yes!

You are the first to reach this milestone
Gabor will be there soon.

Now with the command 'tb' you can access info from a particular position
if you type tb after setboard FEN you get something like this

Code: Select all

setboard 3k4/8/3K4/8/R7/5B2/2r5/8 w - - 0 1
display
+-----------------+
| . . . k . . . . |
| . . . . . . . . |
| . . . K . . . . |
| . . . . . . . . |    Castling: 
| R . . . . . . . |    ep: -
| . . . . . B . . |
| . . r . . . . . |
| . . . . . . . . | [White]
+-----------------+

tb
 White wins in   21 moves
 ----------------------------------
 Rh4      wins in  21 ( 41 plies)
 Rg4      wins in  21 ( 41 plies)
 Bb7      wins in  21 ( 41 plies)
 Bc6      wins in  21 ( 41 plies)
 Ra7      wins in  22 ( 43 plies)
 Rf4      wins in  22 ( 43 plies)
 Be4      wins in  23 ( 45 plies)
 Bd5      wins in  23 ( 45 plies)
 Rb4      wins in  24 ( 47 plies)
 Ra8+     wins in  25 ( 49 plies)
 Ra1      wins in  26 ( 51 plies)
 Ra3      wins in  26 ( 51 plies)
 Ra5      wins in  26 ( 51 plies)
 Bh1      wins in  26 ( 51 plies)
 Bg2      wins in  26 ( 51 plies)
 Ke6     DRAW
 Ke5     DRAW
 Kd5     DRAW
 Ra2     DRAW
 Ra6     DRAW
 Rc4     DRAW
 Re4     DRAW
 Rd4     DRAW
 Ba8     DRAW
 Bd1     DRAW
 Bh5     DRAW
 Be2     DRAW
 Bg4     DRAW

You can access this info in winboard with help-->book (the EGTB is some sort of a book )

And of course, gaviota access it during play.

Miguel

[michiguel]

Ray seems to have a weird problem. I do not know whether is related to the bug I just fixed. It is possible. It is a pity because you are so close to finish!

Miguel

Now all finished !

Fine, Ray!

I am going to have 143 files very soon, I expect they'll be 145.

Only thing I noticed it does not seem to use all cores (cores 4 was used), CPU usage wanders between 25 and 77, no big deal, at least I can use the machine for other things.

Yes! it is 145.

Gaviota uses all the cores in parallel when it calculates all reversible moves. Those do not transition into another EGTB (i.e. captures or promotions) or are pawn moves. In the "transition phase", the algorithm is not parallel (yet). The last files you are generating are all pawns so most of the time is spent in transition phases. Even more now that each position with different pawn structures are calculated separately to save memory. What you see is a normal behavior. When you generate KRBKR you should see gaviota using all the cores most of the time.

Miguel

[Graham Banks]

Now we just need Ray or somebody to make these files available for download.

[Graham Banks]

Now we just need Ray or somebody to make these files available for download.

It would require a 2 GB or so place.

Okay. Too big. One needs to generate their own by the sounds of it.

[michiguel]

You are the first to reach this milestone
Gabor will be there soon.

Maybe not very soon.

KNpKp is ready too. KppKp ran for an hour or so then the screen began to scroll so fast I could not read it. HDD activity was very high. All I could read was 'absent egtb' or something like that and some chess moves.

That was about at slice 6000 out of 27000.

I restarted (this time using only 2 cores) but the same happened again.

The thing happened when I exited a program through whitch I was watching some live broadcast.

Not given up yet but I will restart it only before going to bed and see if it can generate the all-pawn tables if left alone.

Ok, at least we are making progress. You certainly have one file that is corrupted that may have been generated in the previous unsuccessful attempts the other day. The generator try to access it and fails. I just need to figure out which one is it. You do not have to try again because you most probably will have the same result. Just hold it until I figure what to do. I have an idea...

Miguel

[michiguel]

You are the first to reach this milestone
Gabor will be there soon.

Maybe not very soon.

KNpKp is ready too. KppKp ran for an hour or so then the screen began to scroll so fast I could not read it. HDD activity was very high. All I could read was 'absent egtb' or something like that and some chess moves.

That was about at slice 6000 out of 27000.

I restarted (this time using only 2 cores) but the same happened again.

The thing happened when I exited a program through whitch I was watching some live broadcast.

Not given up yet but I will restart it only before going to bed and see if it can generate the all-pawn tables if left alone.

Gabor,

I uploaded a zip file (gs.zip) with a debug version of Gaviota. Whenever you had the problem you describe, it will stop and show exactly what position was missing. That would correspond to the corrupted file.

I included a utility I just wrote to check the integrity of the files externally.
Put tbcheck.exe in the same folder of gaviota, go to the console and type

tbcheck gtb

(gtb is the name of the folder where all the gtb files are, the idea is to have tbcheck <folder>)
It will check all the files and tell you which ones are ok, absent, or corrupted. It should take ~15 min to check them all. You can also check individual files with

tbcheck -i path\file

for instance

tbcheck -i gtb\kppkp.gtb

If you delete the files that are corrupted, you should be able to resume and regenerate them with tbgen 5.

tbcheck is more powerful than tbtest, which is integrated to the engine. tbtest check the longest mate in a particular file. tbcheck compares hash signatures from your files with the correct ones I have here. Even if one byte changed, it should detect it.

Hope it works,

Miguel
PS: if it works well, I may include the utility tbcheck in the distribution or I may integrate it to the engine.
PS2: Do not run tbgen with memory 1 as I mentioned before. That reduces the memory of the tb cache too, slowing things down. Memory is not a problem now. So if you leave the hash memory as it is by default, it should be ok. You can adjust it down a bit in the ini file if you like.

[Spock]

Now we just need Ray or somebody to make these files available for download.

It would require a 2 GB or so place.

Okay. Too big. One needs to generate their own by the sounds of it.

Yes, they would compress to about 2GB with 7-zip

I could create say 4 zip files of them, but I'd need somewhere to upload them with the necesary disk space and bandwidth for subsequent downloads, and then people would need to have a decent un-capped internet connection to actually download them

Easier if people just generate them themselves if they want them

[Dann Corbit]

Version 0.74.6 =
1) I fixed a bug that prevented Gaviota to be used without logs.
2) Building Gaviota tablebases require much less RAM memory available (< 500 Mb).
3) The default download contains a smaller book, so people can download the engine without problems. If someone wants the bigger book, it is available separately.
4) Everything else is identical to 0.74

http://sites.google.com/site/gaviotachessengine/

Miguel

Here is a simple link (no PGO) 64 bit binary:
http://cap.connx.com/chess-engines/new- ... ta.exe.bz2

[Dr.Wael Deeb]

Version 0.74.6 =
1) I fixed a bug that prevented Gaviota to be used without logs.
2) Building Gaviota tablebases require much less RAM memory available (< 500 Mb).
3) The default download contains a smaller book, so people can download the engine without problems. If someone wants the bigger book, it is available separately.
4) Everything else is identical to 0.74

http://sites.google.com/site/gaviotachessengine/

Miguel

Here is a simple link (no PGO) 64 bit binary:
http://cap.connx.com/chess-engines/new- ... ta.exe.bz2

Again this archive doesn't open for me....thanks anyway....
Dr.D

[Dann Corbit]

Version 0.74.6 =
1) I fixed a bug that prevented Gaviota to be used without logs.
2) Building Gaviota tablebases require much less RAM memory available (< 500 Mb).
3) The default download contains a smaller book, so people can download the engine without problems. If someone wants the bigger book, it is available separately.
4) Everything else is identical to 0.74

http://sites.google.com/site/gaviotachessengine/

Miguel

Here is a simple link (no PGO) 64 bit binary:
http://cap.connx.com/chess-engines/new- ... ta.exe.bz2

Again this archive doesn't open for me....thanks anyway....
Dr.D

Do you have bzip2 (free)?
If not, do you have 7-zip (free)?
If not, do you have PowerArchiver (commercial)?

I guess that you are trying to open it with the wrong tool.