OpenChess

I hear comparing engines via limiting the depth of search is completely unreliable since search() differs and depth reported varies widely btwn engines. For what, then, is "go depth" useful?

benstoker wrote:I hear comparing engines via limiting the depth of search is completely unreliable since search() differs and depth reported varies widely btwn engines. For what, then, is "go depth" useful?

Timing measurements. or program chances. I use a specific depth when making changes to see how a change affects the overall size of the tree. Or I use it to evaluate parallel search efficiency. I never use it to play games.

Цель поиска - выбор хода, ведущего к лучшей оценке на текущей глубине.
Если вы работаете над оценочной функцией, то тесты go depth помогут вам построить совершенную оценочную функцию.

jury_osipov wrote:The purpose of the search - Select the course leading to a better assessment of the current depth. If you are working on the evaluation function, the tests go depth will help you build a perfect evaluation function.

Oh, that's what you meant...

It's also useful for analysis of games, to keep the analysis uniform, and save time in easy positions like recaptures, while focusing (giving more time) to difficult positions.

Usually, if the engine takes more time to reach depth 19 in a given position (say, it reaches it in positions that are close in 10 seconds, but in this one it's taking more than a minute), it's probably a critical position (or one is in the horizon), winning and defending games sometimes is about finding such critical positions.