Dragon 3.1 Released at KomodoChess.com

[lkaufman]

We often test on fixed depth on our Linux tester, and the results are always replicable on one thread (at full strength), so I can't readily explain what you are reporting. It's possible it is a Windows issue, since we haven't verified this test on Windows, but that would be strange. We may have to investigate this.

It has been quite a long time since I've used Windows myself. The screen output I shared are from running dragon-3.1-linux-avx2. For another data point, it won't hurt try Komodo 13.02, the most recent of the free versions.

Komodo 13.02 64-bit (C) 2019 Don Dailey, Larry Kaufman and Mark Lefler
using hardware POPCNT
using BMI2
info string Licensed to Komodochess.com
setoption name Threads value 1
setoption name Clear Hash
info string Clearing Hash
go depth 3
info depth 1 time 1 nodes 0 nps 0
info depth 1 time 1 nodes 1 score cp 4 nps 1000 tbhits 0 pv a2a3
info depth 1 time 1 nodes 2 score cp 20 nps 2000 tbhits 0 pv b2b3
info depth 1 time 1 nodes 4 score cp 73 nps 4000 tbhits 0 pv d2d3
info depth 1 time 1 nodes 5 score cp 90 nps 5000 tbhits 0 pv e2e3
info depth 2 time 1 nodes 20 nps 20000
info depth 2 time 1 nodes 43 score cp 31 nps 43000 tbhits 0 pv e2e3 e7e6
info depth 3 time 1 nodes 79 nps 79000
info depth 3 time 1 nodes 120 score cp 59 nps 120000 tbhits 0 pv e2e3 e7e6 b1c3
info time 1 nodes 142 nps 142000
bestmove e2e3 ponder e7e6
setoption name Clear Hash
info string Clearing Hash
go depth 3
info depth 1 time 1 nodes 0 nps 0
info depth 1 time 1 nodes 1 score cp 58 nps 1000 tbhits 0 pv b1c3
info depth 1 time 1 nodes 2 score cp 90 nps 2000 tbhits 0 pv e2e3
info depth 2 time 1 nodes 20 nps 20000
info depth 2 time 1 nodes 28 score cp 31 nps 28000 tbhits 0 pv e2e3 e7e6
info depth 3 time 1 nodes 64 nps 64000
info depth 3 time 1 nodes 84 score cp 59 nps 84000 tbhits 0 pv e2e3 e7e6 b1c3
info time 1 nodes 111 nps 111000
bestmove e2e3 ponder e7e6
setoption name Clear Hash
info string Clearing Hash
go depth 3
info depth 1 time 1 nodes 0 nps 0
info depth 1 time 1 nodes 1 score cp 58 nps 1000 tbhits 0 pv b1c3
info depth 1 time 1 nodes 2 score cp 90 nps 2000 tbhits 0 pv e2e3
info depth 2 time 1 nodes 20 nps 20000
info depth 2 time 1 nodes 28 score cp 31 nps 28000 tbhits 0 pv e2e3 e7e6
info depth 3 time 1 nodes 64 nps 64000
info depth 3 time 1 nodes 83 score cp 59 nps 83000 tbhits 0 pv e2e3 e7e6 b1c3
info time 1 nodes 110 nps 110000
bestmove e2e3 ponder e7e6
quit

Ah, so the behavior happens in older versions of your software too. The same move is selected in the end, but the search paths are different. Only slightly between runs 2 and 3, with both dramatically different from run 1.

If I were to hazard a guess, the data structure storing move ordering is not cleared, with that retrained information informing subsequent searches. In the few test cases examined, node count always decreases. This observation supports the notion of subsequent searches following a tree path with earlier, or more effective pruning, which can be shown to have no effect on the end result. But, I'm not an engine author; my knowledge of the inner workings is superficial.

Addendum. This might not mean a lot, but I tried a comparison between dragon-3.1-linux-avx2 against dragon-3.1-64bit-avx2.exe running under Wine. They give identical outputs. That is, even though the search path differs between Clear Hash commands as above, when given the same sequence of commands the results are identical between the native Linux and emulated Windows versions.

The reasonable conclusion is that single-threaded normal strength Komodo is deterministic, but that issuing Clear Hash does not reset the entirety of the program state, hence what we see. We'll blame Stockfish for skewing my expectations. Stupid fish!

That sounds right, it is consistent with what we observe. I reload the engine when I want to start fresh.

[kyt_gg]

New owner of Komodo Dragon. Larry, I just wanted to make sure I am not doing anything funky but it seems when I put Dragon in MCTS mode, the nps is different depending on the time control (as if it was adjusting itself to play with more nodes when shorter on time). I was wondering why nps was slow for infinite time (I'm mainly using Dragon to analyze my own games).

Is this normal or maybe I am setting up something wrong?

[lkaufman]

New owner of Komodo Dragon. Larry, I just wanted to make sure I am not doing anything funky but it seems when I put Dragon in MCTS mode, the nps is different depending on the time control (as if it was adjusting itself to play with more nodes when shorter on time). I was wondering why nps was slow for infinite time (I'm mainly using Dragon to analyze my own games).

Is this normal or maybe I am setting up something wrong?

This is normal. With more time per move, we use some of the extra time to look at more MCTS nodes, but also use some of the extra time to look deeper at each MCTS node. So with more time, it looks at more total nodes, but fewer nodes per second. But we do draw the line at some point, so infinite time should show the same NPS as some time limit somewhere in the Rapid range (I forget the exact number). Changing number of threads also causes similar behavior. The entire MCTS search in Dragon is quite unique (not used in any other non-GPU engine I believe), and is already very complicated and sophisticated, but I am sure it could be improved dramatically somehow, it's just not obvious how. I would like it to become the main Komodo Dragon search, but that will require some major new insight, not mere tweaking. We have no other engines to get ideas from, since our MCTS is a hybrid between A/B and MCTS and therefore not similar to Lc0 or any GPU based program.

[kyt_gg]

Thanks for the answer Larry! I bought this new laptop to do some overnight analysis, so I'm glad to hear that my 14 threads are working as intended. Just didn't understand why it appeared so slow.

I had another question. Does your team actively think about features that add value for human play or is it almost by coincidence? Like according to you, MCTS is supposed to be harder for humans to play against, but was that a goal or just a consequence of something you thought was going to make the engine stronger anyway?

[lkaufman]

Thanks for the answer Larry! I bought this new laptop to do some overnight analysis, so I'm glad to hear that my 14 threads are working as intended. Just didn't understand why it appeared so slow.

I had another question. Does your team actively think about features that add value for human play or is it almost by coincidence? Like according to you, MCTS is supposed to be harder for humans to play against, but was that a goal or just a consequence of something you thought was going to make the engine stronger anyway?

Well, we made the MCTS mode because MCTS was looking so promising in the NN, GPU-based versions. We hoped it might ultimately prove stronger than A/B, but we also knew that it should be more human-like and also much more friendly to MultiPV use. So I figured that even if it didn't end up as strong as regular Komodo (as is so far the case), it might be a viable product for these other two reasons if it was at least not too much weaker than A/B, as is also so far the case. So multi-purpose, of which 2 out of 3 have so far come true.

[Fuddur]

I am posting this for Larry Kaufma, who is traveling now:

" KomodoChess.com has released Dragon 3.1, which today won the World Computer Chess Championship after a tiebreaking match with
Lc0. It is essentially the same version that won the WCCC, except for adjustments to limited Elo levels. It fixes some issues in
Dragon 3 (occasional time forfeits in standard mode and crashes in MCTS mode), and so is free for anyone who bought Dragon 3 (or got
it by subscription). It has a newer net and search improvements compared to Dragon 3, which make it about ten elo stronger with
normal openings (at CCRL blitz tc, one thread), twelve elo stronger at FRC (chess960), and twenty elo stronger with "unbalanced human
openings". It also has a user-settable minimum time per move called "Skill Time" when using the LimitStrength option, and now allows for Elo setting up
to 3499 (3500 means full strength). As usual, it is free for subscribers, and 20% discounted for purchasers of Dragon 2 thru 2.6. No
price changes from Dragon 3 "

Note I will be sending out copies to rating list testers starting tomorrow.

Hello sir,
Curiously waiting for Android version of dragon
From India!
Thanks..

[Werewolf]

Random thought: is there merit in having a "play function" which starts off in MCTS but switches automatically to AB once a certain advantage (say +3.00) has been built up? The idea is that once the position is winning the engine just wants to terminate the opponent quickly.

[lkaufman]

I am posting this for Larry Kaufma, who is traveling now:

" KomodoChess.com has released Dragon 3.1, which today won the World Computer Chess Championship after a tiebreaking match with
Lc0. It is essentially the same version that won the WCCC, except for adjustments to limited Elo levels. It fixes some issues in
Dragon 3 (occasional time forfeits in standard mode and crashes in MCTS mode), and so is free for anyone who bought Dragon 3 (or got
it by subscription). It has a newer net and search improvements compared to Dragon 3, which make it about ten elo stronger with
normal openings (at CCRL blitz tc, one thread), twelve elo stronger at FRC (chess960), and twenty elo stronger with "unbalanced human
openings". It also has a user-settable minimum time per move called "Skill Time" when using the LimitStrength option, and now allows for Elo setting up
to 3499 (3500 means full strength). As usual, it is free for subscribers, and 20% discounted for purchasers of Dragon 2 thru 2.6. No
price changes from Dragon 3 "

Note I will be sending out copies to rating list testers starting tomorrow.

Hello sir,
Curiously waiting for Android version of dragon
From India!
Thanks..

There are problems for us to make new Android versions, due to changes in policies by Android that make it impractical for us now.