Strelka 5.5 x32 is released....

[Dr.Wael Deeb]

ouch ! This post by Robert will calm down many people

Gab

Robert is a seller man and every seller man must defend his product by any means necessary....
Dr.D

People like to feel they are buying of their own good judgment as a result of the information the salesman has given them

So true

[Dr.Wael Deeb]

Also, for Arasan14 test suite (5s+1ply / position) the results are so much in favour of Strelka 5.5 that I am suspecting some oddity.
Kai

Arasan suite / 30s:
Solved 140/175 in 23m (Critter MP 127 in 37m)

Strelka 5.5 has a database of best-moves stored in the executable file for many popular test positions. (29643 positions in total).

When it reaches depth 5 in such position, it excludes all other moves and searches only the stored best-move.

Wow....Thanks for the info

[michiguel]

Also, for Arasan14 test suite (5s+1ply / position) the results are so much in favour of Strelka 5.5 that I am suspecting some oddity.
Kai

Arasan suite / 30s:
Solved 140/175 in 23m (Critter MP 127 in 37m)

Strelka 5.5 has a database of best-moves stored in the executable file for many popular test positions. (29643 positions in total).

When it reaches depth 5 in such position, it excludes all other moves and searches only the stored best-move.

In the 90's this could have been considered serious cheating (when people judged engines based on the ability to solve problems). And that was the reason why many testers kept their positions secret.

Are those positions flipped north/south west/east? What if a useless pawn is added?

Miguel

[WuShock]

For what it's worth.........

i7 940 @ 3.75
F11 / 512 MB / ponder off / no egtb / klo_50 openings

Time Control : 4 mins + 2 secs

Strelka 5.5 x32 .................57.5 ............ +32,-17,=51

Komodo x64 .....................42.5

[gerold]

ouch ! This post by Robert will calm down many people

Gab

Robert is a seller man and every seller man must defend his product by any means necessary....
Dr.D

Well, after the results of +45, I figured that would make it about as strong as Houdini 1.5 x64, and started a g/5 match. After 30 games, Strelka was under 25%, so... If Strelka 5.5 is the second coming, I did not see it.

After 54 games 3/5 TC
Houdini 1.5 vs. Strelka 5.5 ..... 21-6-26. 64%.
Only 54 games but the new Strelka has a lot of catching up to do.

[rvida]

Also, for Arasan14 test suite (5s+1ply / position) the results are so much in favour of Strelka 5.5 that I am suspecting some oddity.
Kai

Arasan suite / 30s:
Solved 140/175 in 23m (Critter MP 127 in 37m)

Strelka 5.5 has a database of best-moves stored in the executable file for many popular test positions. (29643 positions in total).

When it reaches depth 5 in such position, it excludes all other moves and searches only the stored best-move.

In the 90's this could have been considered serious cheating (when people judged engines based on the ability to solve problems). And that was the reason why many testers kept their positions secret.

Are those positions flipped north/south west/east? What if a useless pawn is added?

Miguel

Positions are stored as hash keys. No flipping, no mirroring.

example:
rn1qr1k1/1p2bppp/p3p3/3pP3/P2P1B2/2RB1Q1P/1P3PP1/R5K1 w - - bm Bxh7+; id "01. Polugaevsky-Torre (London) 1984";

Strelka outputs Bxh7 at depth 6 with a huge negative score. Only at depth 16 it starts to see the move is winning.

Another one:
r3k2r/2q1b1pp/pp2p3/2pR4/P7/4BP2/1PP1Q1PP/5RK1 w kq - bm Bf4; id "arasan13.30"; c0 "Rozentalis-Dokhoian, USSR 1986";

again, Bf4 at depth 6 with much lower score than Rd3 at depth 5.

If these positions are flipped, it takes much much more plies to find the solution.

[Dr.Wael Deeb]

For what it's worth.........

i7 940 @ 3.75
F11 / 512 MB / ponder off / no egtb / klo_50 openings

Time Control : 4 mins + 2 secs

Strelka 5.5 x32 .................57.5 ............ +32,-17,=51

Komodo x64 .....................42.5

Impressive never the less....

I assume that this Komodo 4,right

[gleperlier]

Strelka 5.5 has a database of best-moves stored in the executable file for many popular test positions. (29643 positions in total).

When it reaches depth 5 in such position, it excludes all other moves and searches only the stored best-move.

Is it only a way to twink the engine results in test position or really a way to improve strenght ?

Gab

[LudiBuda]

Thanks for the info Richard.
I was so impressed when I saw Strelka solve some of my favourite WM test positions 10 plies sooner then any other engine

Shredder does something similar (internal opening book) when openning advisor is on, so if that's not wrong then what Strelka is doing is also fine.

[Dr.Wael Deeb]

Thanks for the info Richard.
I was so impressed when I saw Strelka solve some of my favourite WM test positions 10 plies sooner then any other engine

Shredder does something similar (internal opening book) when openning advisor is on, so if that's not wrong then what Strelka is doing is also fine.

Well,it's a nice cheating alright