PK wrote: ↑Sun Sep 01, 2019 3:43 pmDifferent playing styles, and ability to tweak them, should be a huge selling point, and one that is independent from playing strength. Why nobody does that?
Because it doesn't increase sales in practice?
Komodo 1.0 tried it and allowed people to create personalities with it and such, and it was ditched because it was clear customers didn't care much about it (or for those that cared, Komodo 1 was strong enough and stylistic enough to deliver at the time, so stronger versions for personalities wasn't necessary.)
In general, commercial developers have decided to rid their engines of configurable options for users. Learning was a major setting removed from commercial engines that used to have it, such as Rybka, Houdini and Shredder. What I gather is that adding more options does not increase sales, and removing options doesn't seem to decrease sales, so we only have to blame the people with the money interested in chess engines for what they buy, that dictates what options are kept or added, and engine personalities doesn't seem like one of them.
Except this also affects free engines as well, just look at Stockfish 14051722's settings:
Believe it or not, these few parameters were enough to "have fun" and change the playing style of the engine "drastically" (relative to other Stockfishes.)
So, how did the
very next version look like?
BAM! Overnight, something of value was lost, and... I think it has been difficult to find someone else that knows or cares about this, because Stockfish is an engine used by people looking for strong moves, and any tweaking of those settings would only make it weaker. While people that didn't care about strong moves and only style were focused on engines that could be tweaked further, such as Rodent. But I reckon Stockfish 14051722 with Pawn Structure = 0 and Aggressiveness=200 was the most stylistic engine from this codebase, even surpassing the best Glaurung.
My dream is that one day someone comes with a tool to "equalize" chess engines, so that they all have a performance of 50% against each other, and then we'll be able to see the actual chess being played regardless of how "weak" they play, and it'll manage to show their true playing style, and we'll be able to compare them. And only then, we'll be able to know if Rodent or some other engine has the most exciting one, and if tweaking the parameters can create the ultimate chess entity that would blow us away with its moves every single time it plays, but not necessarily because they're strong.
Like so many things, it seems I jumped on the ship right as it was sinking. Hopefully the main reason for finishing the project is boredom, and one day you can come back with a new engine, with beautiful source code, that leaves the Rodent project in shambles, by comparison.
Your beliefs create your reality, so be careful what you wish for.