testing engines with fixed depth

[BBauer]

anks for explaing

i was for a moment wondering why robbolito, stockfish and glaurung was so fast at fixed depth games.
philou took much more time for calculating at fixed depth play.

anyway the test can be usefull.

Yeah, it can use bandwidth.

[Togga]

thanks for explaining

i wanted to write:)


if you can, you can use every test for conclusion.

compare different versions of stockfish against philou at fixed depth.

the best stockfish version is the one with the best result against philou.

look at the result of stockfish and glaurung against philou.

stockfish was better.

[bob]

no Program Elo + - Games Score Av.Op. Draws

62 Stockfish 1.6 w32 1CPU 2980 14 14 1600 65.0% 2872 33.0%

251 Glaurung 2.2 w32 1CPU 2805 11 11 2430 47.6% 2822 32.4%

http://www.husvankempen.de/nunn/40_4_Ra ... liste.html

My data is not distorted with opening books, learning, etc. I've simply played several million games using a large (4,000) set of starting positions, and alternating colors. No learning kept between games, no book learning at all as there are no books, etc.

[Togga]

do you have own written chess engine?

[benkidwell]

(post edited to remove misunderstanding of post-response sequence.)

[Togga]

ok

[benkidwell]

I have read that in the past there was a fairly linear correlation between search depth and engine strength. I understand with modern pruning and selective search this is no longer true, but is the relationship still fairly linear for any given engine? Or do some engines "scale better" or worse, showing more or less relative gain in palying strength per ply?

[Togga]

when one program is elo 2980 and other elo 2805 then the result can be 10:0 like stockfish - glaurung .

right?

must not.

i also do not use book learning for testing.

i use these 5 opening position for my minimally testing.

[Fguy64]

Hi Paul

I have a stripped down version of my engine that does a fixed depth search (meaning no iterative deepening, no extensions, no qSearch, no special pruning) It is only a plain vanilla alphaBeta search with some MVV/LVA ordering on captures. It can be easily run with or without hashing.

I fund it useful for testing my base algorithm, and also I entertain myself with solving tactical problems.

Anyways, If this interests you, perhaps we can compare notes, i'd be interested in knowing how my base algorithm stacks up with others in terms of performance, with all of the bells and whistles removed.

[govert]

do you have own written chess engine?

Are you wondering if Prof. Hyatt has written his own engine?