Stockfish depth vs. others; challenge

[Henk]

Large search depths says nothing about program strength. Semi-Cheating is very easy. For instance if I reduce depth in all my branches in the main search with N instead of 1 [except for the one which checks for a king capture, or the one that checks whether a position is encountered before ] my search depth will be increased with almost factor N.

[lkaufman]

Large search depths says nothing about program strength. Semi-Cheating is very easy. For instance if I reduce depth in all my branches in the main search with N instead of 1 [except for the one which checks for a king capture, or the one that checks whether a position is encountered before ] my search depth will be increased with almost factor N.

Of course it's easy to inflate search depth artificially, but Stockfish doesn't do that, it's all open source. They do reduce more than others, but not enough to explain the huge gap in reported search depth.

[Daniel Shawul]

Large search depths says nothing about program strength. Semi-Cheating is very easy. For instance if I reduce depth in all my branches in the main search with N instead of 1 [except for the one which checks for a king capture, or the one that checks whether a position is encountered before ] my search depth will be increased with almost factor N.

Of course it's easy to inflate search depth artificially, but Stockfish doesn't do that, it's all open source. They do reduce more than others, but not enough to explain the huge gap in reported search depth.

If I reduce by 2 plies from the second move onwards, i get close depths as stockfish, which ofcourse depends on type of position. But the point is you can choose to reduce by a huge margin and then try to fill up holes by either going rogue with all sorts of pruning, filling up the void with extensions (singular etc), or by doing other things that will cover up the weakness. In your case, Komodo may be designed witha a cuatios style (no aggresive reduction) from the beginning and any changes you make to try and increase bakfires because it is not designed from the start to use aggresive reductions. It is a choice of style. For me I always try what nominal maximum depth I can get before every release without hurting strength. If I am successful, then it would go into the release, because usually a larger nominal depth performs better at long TC. There is no concrete answer because stockish probably had this as a goal from the start. You can get a pretty strong tactical engine if/when it works, which I belive is what is scaring you, a valid concern But you should not expect to get answers that will get you high depths, and not loose some elos, because Komodo probably was designed with caution from the beginning.

[lkaufman]

Large search depths says nothing about program strength. Semi-Cheating is very easy. For instance if I reduce depth in all my branches in the main search with N instead of 1 [except for the one which checks for a king capture, or the one that checks whether a position is encountered before ] my search depth will be increased with almost factor N.

Of course it's easy to inflate search depth artificially, but Stockfish doesn't do that, it's all open source. They do reduce more than others, but not enough to explain the huge gap in reported search depth.

If I reduce by 2 plies from the second move onwards, i get close depths as stockfish, which ofcourse depends on type of position. But the point is you can choose to reduce by a huge margin and then try to fill up holes by either going rogue with all sorts of pruning, filling up the void with extensions (singular etc), or by doing other things that will cover up the weakness. In your case, Komodo may be designed witha a cuatios style (no aggresive reduction) from the beginning and any changes you make to try and increase bakfires because it is not designed from the start to use aggresive reductions. It is a choice of style. For me I always try what nominal maximum depth I can get before every release without hurting strength. If I am successful, then it would go into the release, because usually a larger nominal depth performs better at long TC. There is no concrete answer because stockish probably had this as a goal from the start. You can get a pretty strong tactical engine if/when it works, which I belive is what is scaring you, a valid concern But you should not expect to get answers that will get you high depths, and not loose some elos, because Komodo probably was designed with caution from the beginning.

Thanks. Komodo is actually pretty aggressive with reductions though, but also with extensions so the search depth isn't impressive. But I was hoping that someone would compare SF to Ivanhoe (or another Ippo version that is open source). If there is something about the design of the two engines that explains the difference, it should be possible to point it out.