How i s Komodo getting on? (now, end of December 2016)

[Hurnavich]

Hi,

Just downloaded Komodo 10.3

[Werewolf]

I hope this means that it will not lose ever again to Stockfish,

Do you have any idea what you're talking about?

[Jesse Gersenson]

It should also have the board presence as good as Carlsen has! (to know the best ways to probe the other side's weaknesses).

UCI engines do not know they have an opponent.

[Werner]

We have had some nice gains from the last release (Komodo 10.3).
We do expect a scaling with more processors improvement over earlier Komodo versions, but we have not had enough machine time to characterize it for more than 4-8 threads/cpus.

10.3?
Version 10.3 is availble now

[mjlef]

We have had some nice gains from the last release (Komodo 10.3).
We do expect a scaling with more processors improvement over earlier Komodo versions, but we have not had enough machine time to characterize it for more than 4-8 threads/cpus.

10.3?
Version 10.3 is availble now

I meant gains from Komodo 10.2. Sorry.

[mjlef]

It should also have the board presence as good as Carlsen has! (to know the best ways to probe the other side's weaknesses).

UCI engines do not know they have an opponent.

There is a UCI option called "UCI_Opponent" in the UCI specs, but I do not think I have seen it supported in any GUI. From the spec:

= UCI_Opponent, type string
With this command the GUI can send the name, title, elo and if the engine is playing a human
or computer to the engine.
The format of the string has to be [GM|IM|FM|WGM|WIM|none] [|none] [computer|human]
Example:
"setoption name UCI_Opponent value GM 2800 human Gary Kasparow"
"setoption name UCI_Opponent value none none computer Shredder"


It certainly would be nice to have engines modify play based on the type or name of the opponent (human/computer, elo etc). Well, assuming we can decide what might help in those cases. Contempt should probably automatically increase against humans and weaker programs.

[Ferdy]

It should also have the board presence as good as Carlsen has! (to know the best ways to probe the other side's weaknesses).

UCI engines do not know they have an opponent.

There is a UCI option called "UCI_Opponent" in the UCI specs, but I do not think I have seen it supported in any GUI. From the spec:

= UCI_Opponent, type string
With this command the GUI can send the name, title, elo and if the engine is playing a human
or computer to the engine.
The format of the string has to be [GM|IM|FM|WGM|WIM|none] [|none] [computer|human]
Example:
"setoption name UCI_Opponent value GM 2800 human Gary Kasparow"
"setoption name UCI_Opponent value none none computer Shredder"


It certainly would be nice to have engines modify play based on the type or name of the opponent (human/computer, elo etc). Well, assuming we can decide what might help in those cases. Contempt should probably automatically increase against humans and weaker programs.

Arena supported this in a buggy way.

2016-12-22 22:56:29.494<--1:info string Deuterium v14.3.34.130 64bit POPCNT
2016-12-22 22:56:29.525-->1:uci
2016-12-22 22:56:29.525<--1:id name Deuterium v14.3.34.130
2016-12-22 22:56:29.525<--1:id author Ferdinand Mosca
2016-12-22 22:56:29.525<--1:option name Hash type spin default 64 min 8 max 2048
2016-12-22 22:56:29.525<--1:option name EvalHash type spin default 16 min 4 max 32
2016-12-22 22:56:29.525<--1:option name PawnHash type spin default 4 min 1 max 8
2016-12-22 22:56:29.525<--1:option name KingAndPawnHash type spin default 4 min 1 max 8
2016-12-22 22:56:29.525<--1:option name Log type check default false
2016-12-22 22:56:29.525<--1:option name Ponder type check default false
2016-12-22 22:56:29.525<--1:option name OwnBook type check default false
2016-12-22 22:56:29.525<--1:option name BookFile type string default <empty>
2016-12-22 22:56:29.525<--1:option name KingSafetyPercent type spin default 100 min 0 max 1000
2016-12-22 22:56:29.525<--1:option name MidGameMobilityPercent type spin default 100 min 0 max 1000
2016-12-22 22:56:29.525<--1:option name EndGameMobilityPercent type spin default 100 min 0 max 1000
2016-12-22 22:56:29.525<--1:option name OffensivePercent type spin default 100 min 100 max 10000
2016-12-22 22:56:29.525<--1:option name DefensivePercent type spin default 100 min 100 max 10000
2016-12-22 22:56:29.525<--1:option name FutilityPruning type check default true
2016-12-22 22:56:29.525<--1:option name LateMoveReduction type check default true
2016-12-22 22:56:29.525<--1:option name NullMove type check default true
2016-12-22 22:56:29.525<--1:option name Razoring type check default true
2016-12-22 22:56:29.525<--1:option name GoodEvaluationPruning type check default true
2016-12-22 22:56:29.525<--1:option name RepetitionScore type spin default 0 min -500 max 500
2016-12-22 22:56:29.525<--1:option name InsufficientMaterialScore type spin default 0 min -500 max 500
2016-12-22 22:56:29.525<--1:option name FiftyMoveScore type spin default 0 min -500 max 500
2016-12-22 22:56:29.525<--1:option name StaleMateScore type spin default 0 min -30000 max 30000
2016-12-22 22:56:29.525<--1:option name TimeBufferMilliSec type spin default 60 min 0 max 1000
2016-12-22 22:56:29.525<--1:option name CheckExtension type check default true
2016-12-22 22:56:29.525<--1:option name EndingExtension type check default true
2016-12-22 22:56:29.525<--1:option name RecaptureExtension type check default true
2016-12-22 22:56:29.525<--1:option name PawnTo7thRankExtension type check default true
2016-12-22 22:56:29.525<--1:option name UseScorpioBitBases type check default false
2016-12-22 22:56:29.525<--1:option name AvailableScorpioEgbbMen type spin default 5 min 3 max 6
2016-12-22 22:56:29.525<--1:option name EgbbMinimumProbeDepth type spin default 8 min 1 max 128
2016-12-22 22:56:29.525<--1:option name ScorpioBitBasesPath type string default <empty>
2016-12-22 22:56:29.525<--1:option name ScorpioBitBasesCache type spin default 16 min 8 max 512
2016-12-22 22:56:29.525<--1:option name ScorpioBitBasesLoadType type combo default 4MEN var NONE var 4MEN var SMART var 5MEN
2016-12-22 22:56:29.525<--1:option name MultiPV type spin default 1 min 1 max 300
2016-12-22 22:56:29.525<--1:option name UCI_ShowCurrLine type check default false
2016-12-22 22:56:29.525<--1:option name UCI_ShowRefutations type check default false
2016-12-22 22:56:29.525<--1:option name Engine_LimitStrength type check default false
2016-12-22 22:56:29.525<--1:option name Engine_Elo type spin default 2800 min 1000 max 2800
2016-12-22 22:56:29.525<--1:option name UCI_AnalyseMode type check default false
2016-12-22 22:56:29.525<--1:option name UCI_Opponent type string default none 2800 computer Android
2016-12-22 22:56:29.525<--1:option name ClearHash type button
2016-12-22 22:56:29.525<--1:uciok
2016-12-22 22:56:29.525-->1:setoption name Hash value 128
2016-12-22 22:56:29.525-->1:setoption name OwnBook value true
2016-12-22 22:56:29.525-->1:setoption name UseScorpioBitBases value true
2016-12-22 22:56:29.525-->1:setoption name AvailableScorpioEgbbMen value 4
2016-12-22 22:56:29.525-->1:setoption name ScorpioBitBasesPath value C:\myfiles\chess\egbb4men
2016-12-22 22:56:29.525<--1:info string Hash is set to 128mb
2016-12-22 22:56:29.525-->1:isready
2016-12-22 22:56:33.644<--1:readyok
2016-12-22 22:56:51.020-->1:ucinewgame
2016-12-22 22:56:51.020-->1:isready
2016-12-22 22:56:51.036<--1:readyok
2016-12-22 22:56:51.052-->1:setoption UCI_Opponent value none 2400 human Ferdinand
2016-12-22 22:56:51.052-->1:position startpos moves d2d4
2016-12-22 22:56:51.052-->1:go wtime 66000 btime 60000 winc 6000 binc 6000
2016-12-22 22:56:51.052<--1:info depth 1 seldepth 1 score cp 10 time 0 nodes 27 nps 0 tbhits 0 pv g8f6
2016-12-22 22:56:51.067<--1:info depth 2 seldepth 6 score cp -53 time 0 nodes 185 nps 0 tbhits 0 pv g8f6 c1f4

It sends to the engine this,

Code: Select all

setoption UCI_Opponent value none 2400 human Ferdinand

"name" is missing after setoption, but I let Deuterium understand this.

When someone create a tournament in Arena and fill up the rating for the participants that rating is sent to the engine.

[S.Taylor]

It should also have the board presence as good as Carlsen has! (to know the best ways to probe the other side's weaknesses).

UCI engines do not know they have an opponent.

I actually meant in general. That Carlsen seems to try the whole time to rock the boat, which i thought can be done in a general way, to give the best chances for ANY opponent in general, man or machine, to go wrong.
Why, would the way Carlsen does it against humans, be useless if he were playing vs engines?

[S.Taylor]

I hope this means that it will not lose ever again to Stockfish,

Do you have any idea what you're talking about?

I didn't mean totally. But almost.
If that was the main way komodo was getting improved was to lose less games, but not to win much more than it does, then obviously it has to draw MANY more than it has been doing, vs SF. (i.e. not let SF win).

[lkaufman]

I hope this means that it will not lose ever again to Stockfish,

Do you have any idea what you're talking about?

I didn't mean totally. But almost.
If that was the main way komodo was getting improved was to lose less games, but not to win much more than it does, then obviously it has to draw MANY more than it has been doing, vs SF. (i.e. not let SF win).

The improvement should mean more wins in general, and fewer draws against weaker opponents. Long time limits make for more draws regardless of who is playing. Color becomes more important with more time. With enough time (and cores) the stronger engine should almost never lose with White if normal (GM approved) openings are used.