Hiarcs 14 WCSC or Deep Shredder 13, to analyse like a human?

Hi all.

I like playing with UCI Engines that have a "human" style of play, rather than an absolute perfect, but very computer-like, play. That's why I generally play against Maia.

However, to help me analyse, I want to make use of something different, very strong, but still with some sort of Human personality in selecting their candidate moves and plans.

My choice would be between Hiarcs 14 WCSC and Deep Shredder 13.

As I'm not too familiar with those UCI engines, I'm seeking opinions and suggestions by those who have tried either of the two, or even better, can toss in some comments about both of the engines.

Thanks beforehand.

AG

IMO, neither engine, playing at full strength, will appear "human" to an actual human. Their play is indistinguishable from God Almighty to all but a few and I'm skeptical about the few.

They appear human only when they begin to make noticeable mistakes - and that happens only when they're intentionally made to play weaker.

That said, HIARCS has a reputation for playing more "human-like" moves than other UCI engines. I don't buy it, but people say this. (Maybe it was true 15-20 years ago.) I can tell you, however, that HIARCS, dumbed down (running on the HIARCS Chess Explorer software) provides a reasonable illusion that you're playing another human being.

That's only an opinion and I do own both programs.

@agatti1970 - Perhaps you will have look for LC0 with MAIA. I like the playing style. 👍

Hello @Crappov, thank-you for sharing your thoughts. On my side, I actually don't plan to PLAY against any of those two. For that purpose, I mentioned Maia in the original post. I'd like to use either Hiarcs or Deep Shredder to assist me in the analysis. Some time ago, I had the chance to use Hiarcs for a while, and I did a test, letting it analyse some positions from my chess books, next to Stockfish and other engines. In a number of occasions, Hiarcs was coming up with a line that was identical, or very close, to the book's real players' games. The same could not be said of Stockfish, who probably was giving me the very best line on earth, but absolutely not human, if you know what I mean.

That's why I started getting interested in engines that may help in analysing games, with their "human-like" line of thinking. For playing, in the absence of a good friend at the club, or of a player here (or on lichess!), than Maia is my go-to engine.

That said, thank-you for suggesting Hiarcs, point taken.

AG

Yes, that's what I do for playing purposes. For assisted analysis, Maia is unusable, because she' not showing any PV. Her node count is limited to 1, so basically unusable.

Thank-you @DesperateKingWalk. You're right, setting an engine's personality does affect its playing style, and can make it play somehow more "humanly" :-) or "human-like". However, in the past I found that tweaking personalities does not always give the desired result. For example, I could tweak an engine to be an extremely strong attacker, or to value much more his knights rather than his bishops, but can this be called a "human-like" approach?

Chess.com's bots, the ones I'm up to, are not really playing human style. The dumb-down variants are really poor players, that sometimes make "funny" blunders that a human would never make.
In terms of tweaking personalities, my favorites are Rodent III and Rodent IV. They both have a number of presets personalities, plus they give the possibility to create infinite ones.

Wrong forum.

Engines, not really.

I once opened a thread on the topic of engines in the general section, and have been warned that I should have posted here. Where are these threads to be opened then?

Oh wow, interesting. Thank-you for sharing. And I enjoyed the game as well!

Here.

Andrea, in addition to Maia, I would say Shredder. It's available as an app on the Google Play Store. It has a good "human-type" style, and it adusts its level based on whether you won or lost against it. I think it's one of the best engines for that.

Like you, I often play against engines, so I look for those with intelligent skill adjustment. I have an old version of Komodo that I spar against (Komodo 13) and it is good for that (25 skill levels). Of the free engines, Texel is a good engine with a very wide elo range. Another free engine is Arasan, which starts at 1000 elo.

I agree with you about Rodents 3 and 4. They are still great engines to play against as sparring partners. Rodent is one of the most fun and accessible chess engines ever, a real legend!

There was an interesting discussion in talkchess last month about strong neural nets generated from human games. Dkappe, the creator of several nets for Lc0 (BadGyal, GoodGyal, TinyGyal) which were all based on lichess.org games plus a percentage of Stockfish games, recommended his Harmon net, which is based on games from Kingbase. It uses the original NNUE format (20 MB) used in Stockfish 12 and 13, and also Cfish 12 and other clones. However, it maybe best to use an engine that will allow you to choose pure NNUE for the evaluation intead of hybrid.

Cfish 12 paired with the Harmon net and set to use pure NNUE is about as strong as Deep Shredder 13 (about 3300 elo, CCRL , single-core), so it maybe a good alternative to Shredder for analysis, since Shredder 13 still uses hand-coded evaluation. It's up to you then to decide which is more human-like in their evaluation recommendations.

No human has performed at a 3300-level, so Shredder and Cfish may not be able to provide moves that will be consistently judged as more likely to have been made by a human. Some players (Carlsen, Caruana) though, have reached TPRs of 3000+, which is about the level of Hiarcs 14. So maybe Hiarcs can provide a more human-like evaluation for analysis. You'd have to dumb down Shredder and Cfish (via nodestime, perhaps) to reach Hiarcs 14's level. Then you'll have three possible evaluations to choose from during analysis. Two using the classic hand-coded evaluation, and one using a neural net trained on human games.

@Lotus960 and @taychoe, many thanks both for the thoughtful posts. Yes, Shredder 13 seems a very strong, while still "human" choice. Probably I will go with Shredder, tuned to have a lower strenght than its full potential, to benefit from its human approach, but still with the option to present me candidate moves and plans that I'm able to appreciate and understand. There's no point for me to have the absolutely very best line from it, if I am so far away from understanding it, or having them play some moves against me, which I will very unlikely get from my usual opponents. Thanks! - AG

I forgot to mention the weights files of DKappe in my previous post.

As @taychoe said, they are a variety of nets created from human games on Lichess mixed in with some Stockfish. They are used with the Leela engine, just like the Maia weights.

I downloaded them all last year, and I have played against them. They make good sparring partners as they have a human style.

Quote from Dkappe:

<<Training the Ideal Sparring Partner

If you like to watch engines play against each other, these nets are not for you. They are designed to give humans good sparring partners. ("Bad Gyal" is Jamaican for "bad girl.")

How were they trained? By taking lichess games and producing Q and policy data from low depth SF10 searches. These are then trained against q+z. Basically we combine flawed human play with sf10 smarts (but not too smart) to give us a more swashbuckling and occasionally positional opponent.

Good Gyal and Evil Gyal

I’m training nets on the same data, but with different q-ratios. Good Gyal is at a q-ratio of 0.75, which means it gets more influence from Stockfish, while Evil Gyal is at a q-ratio of 0.25, so more influence from sleazy human blitz play.>>

Its so bizarre that people insistently advising sparring partner hence OP frequently mentioning that he is looking for an engine which can analyze like a human!? Which part you don't get and why? I am frankly curious about it.