Wish list for next version of Komodo Dragon

[cma6]

LK, in the current version, it is cumbersome to change between checked "use MCTS" and unchecked (do not use MCTS). In my Aquarium GUI, I have to first switch to a different engine, then go into engine parameters to make the switch. It would be great if one could switch back and forth without needing to go into engine parameters.

Hopefully, others can add to the wish list for next version of Komodo Dragon.

[carldaman]

A learning function similar to what ShashChess and SugaR offer (free SF derivatives).

It's hard to believe there is 'not enough interest' in a feature that would store useful analysis in a learning file, and also make Komodo Dragon a stronger entity over time.

[Marek Soszynski]

LK, about MCTS:
What is the difference between "MCTS Hash" and the hash set by the GUI? And what is the best balance between them?

About Komodo and Aquarium:
Your printed books involved the use of Komodo within Aquarium's IDeA. But that stopped with the last one, I believe, and I don't remember your giving an explanation for abandoning IDeA (if you did). What happened?

[Cornfed]

The "arms race" has begun.

The 'net' must be the size of FF2.
These act as 'defacto' opening/middlegame books...just 'positions' without moves to get there...the search gets you there.

Beyond that...surprise us.

But please, stop spending time on the gimicks like odds play and 'Chess 360'. Chess isn't played out...but if a programmer seems forever mired in 2nd/3rd place to freeware, I can see where one might like to spread that gospel.

[lkaufman]

LK, about MCTS:
What is the difference between "MCTS Hash" and the hash set by the GUI? And what is the best balance between them?

About Komodo and Aquarium:
Your printed books involved the use of Komodo within Aquarium's IDeA. But that stopped with the last one, I believe, and I don't remember your giving an explanation for abandoning IDeA (if you did). What happened?

In MCTS mode, the normal hash table is used by the short searches done at each MCTS node, so there is no point in using a large size; even 32 MB is plenty. The MCTS Hash is used to store the results of these short searches; the default should be fine for blitz, but as the time per move increases and as the number of threads increases, MCTS Hash needs to be larger. For overnight analysis on one position on a machine with many cores I would use a pretty big value, at least several GBs. If you always stay in MCTS mode you can just keep a large value for MCTS Hash, but if you just switch to MCTS mode now and then, just use maybe double the default on a modern laptop, quadruple on a machine with 8 or more cores, unless you are doing quite deep analysis.
Regarding IDeA, it has been many years since we've had any contact with the Aquarium people, my last version is many years old. The need for the Aquarium minimaxing is much less now than it was a decade ago due to the tremendous increase in software strength and to the ability to look at multiple lines "for free" in Komodo MCTS mode and with Lc0 versions. It comes down to whether I'm writing more for correspondence players or for OTB and online players. For correspondence chess the Aquarium minimax still makes sense, but for OTB players I'm more interested in being able to look at multiple options than in finding the absolute "best" line, which is often just a draw.

[lkaufman]

The "arms race" has begun.

The 'net' must be the size of FF2.
These act as 'defacto' opening/middlegame books...just 'positions' without moves to get there...the search gets you there.

Beyond that...surprise us.

But please, stop spending time on the gimicks like odds play and 'Chess 360'. Chess isn't played out...but if a programmer seems forever mired in 2nd/3rd place to freeware, I can see where one might like to spread that gospel.

Well, OTB human chess isn't played out, humans make too many mistakes, but I agree with those who say that Correspondence chess is pretty much played out, and engine vs engine chess, on many cores at long time limits, is also played out unless bad openings are mandated. Even if we were clearly number 1, I would still advocate for ideas like chess 960, various forms of Armageddon, mandated bad openings, and rule changes to reduce draws, as these are all things that would make the elo gap between number 1 and number 2 larger. So as long as Dragon is number 2 (on most hardware), I am actually arguing against my own best interest when I advocate these ideas. I go into a lot of these ideas in my forthcoming "Chess Board Options" book.

[cma6]

[quote=lkaufman post_id=883230 time=1613614327 user_id=4773]
"Iif you just switch to MCTS mode now and then..."

For deep correspondence analysis (now and then) would that amount to something like 1024 MB (1 GB) for MCTS hash?

[lkaufman]

"Iif you just switch to MCTS mode now and then..."

For deep correspondence analysis (now and then) would that amount to something like 1024 MB (1 GB) for MCTS hash?

That's probably enough unless you have a lot of cores, like 32 or more.

[bastiball]

"Iif you just switch to MCTS mode now and then..."

For deep correspondence analysis (now and then) would that amount to something like 1024 MB (1 GB) for MCTS hash?

That's probably enough unless you have a lot of cores, like 32 or more.

Does the table memory affect the MCTS?

[JohnW]

I don't know if it makes sense or not but I would like to see an elo rating associated with the various personalities. For example, when I select "Human" , am I selecting GM or a club player?