Komodo 1.3 Official Release Available

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
User avatar
Don
Posts: 5106
Joined: Tue Apr 29, 2008 2:27 pm

Re: Komodo 1.3 Official Release Available

Post by Don » Wed Dec 29, 2010 2:52 pm

fern wrote:Thanks Dan for your detailed answer. Respect to my idea of an evaluator, i was thinking JUST in something of that sort for YOUR engine, not to every engine because of the reason you said.

In my opinion, if in a complicated tactical position the human player choose the best move, that for itself is enough to put him in a certain level of proficiency. If then he play badly missing a more or less obvious move, then you have another criteria to evaluate him. In my opinion -perhaps Larry thinks the same- the fact that makes the difference between a master of any class and a good club player is that this last one is inconsistent, he can play very good moves but then he ruin everything with a silly one.

What is preposterous in any and every case is that an expert level player that anyway will lose almost every game to a 3000 engine be considered a 800 player due to that reason. It is not fair. same for club players, etc.

I like your idea of an engine going up or down in his capabilities according to results. that would be a "camarade" engine, or "Partner".

Finally, and I think we already discussed this, I believe that an engine can be weakened as much as his number of ply diminish. In fact that is probable one of the factor that makes the difference between players. Looking at the PV as I play, I have seen many times that the move I chose was considered good -for me- until ply 5,6 or 7 and then the engine goes on and get the real value of it at around ply 11 or 12 and beyond. What that it means? It means we middle level players does not go deep enough and so an engine, without committing obvious mistakes, can be weakened just in that manner.


My best
Fern
The easiest way to weaken the engine is to make it play fixed depth. A 1 ply search is pretty weak, but a 5 ply search will beat most players, but not good tournament players, etc. But there is a very big jump going from one ply to the next so this is rather crude.

So one way is to have an intermediate step doing something like what I said earlier, playing the occasional somewhat losing game.

One concern of mine however is that even a 1 ply search tends to play very good positional moves. In other words, it doesn't play the same way as a human of the same approximate strength.

However I think have a general solution, I can play games based on fixed node levels. The GUI would hide the details of this from you. Then I have fine control over the ELO level. This is basically how Rexchess worked. You could set it to play at some specified ELO level and it adjusted itself by computing how many nodes to look at per move.

User avatar
fern
Posts: 8755
Joined: Sun Feb 26, 2006 3:07 pm

Re: Komodo 1.3 Official Release Available

Post by fern » Wed Dec 29, 2010 4:05 pm

Surely you are right, you are the programmer ..:-)
In any case, I will insist in this more refined idea to evaluate players.
I suppose that some kind of equation should be considered.
By example:
results against the level the machine played + number of best moves played in difficult positions + number of bad moves played in easy positions + number of second best moves in tactical positions + best or bad moves in positional situations

You can make it more complex or not, but in any case I bet an evaluator of this kind should be more adequate that simple taking into account results.
No need the player evaluate himself in the setti8gs of the program. Just a game show more or less which your caliber is, more or less...

BTW, I have REX, a curious program. Like me, mixing good with bad moves. Well, more good moves than me...

Fern

User avatar
Don
Posts: 5106
Joined: Tue Apr 29, 2008 2:27 pm

Re: Komodo 1.3 Official Release Available

Post by Don » Wed Dec 29, 2010 4:32 pm

fern wrote:Surely you are right, you are the programmer ..:-)
In any case, I will insist in this more refined idea to evaluate players.
I suppose that some kind of equation should be considered.
By example:
results against the level the machine played + number of best moves played in difficult positions + number of bad moves played in easy positions + number of second best moves in tactical positions + best or bad moves in positional situations

You can make it more complex or not, but in any case I bet an evaluator of this kind should be more adequate that simple taking into account results.
No need the player evaluate himself in the setti8gs of the program. Just a game show more or less which your caliber is, more or less...

BTW, I have REX, a curious program. Like me, mixing good with bad moves. Well, more good moves than me...

Fern
I can provide an analysis mode that evaluates the quality of your moves. But it is virtually impossible to estimate a rating based on this. Some players are not very good but never make errors, and some are very creative and brilliant but make too many errors and they may be the same strength as measured over many games. If you are an error making, it cuts into your strength enormously.

When I first started playing tournament chess I knew very little about chess but thought I did. I was a kid and easily got outplayed by almost everyone in the chess club I went to and usually play USCF C and D players.

When I started playing in tournaments the first thing I noticed is that I was not actually getting outplayed that much - I would just blunder quite frequently. And my few wins were based on blunders of my opponent. Not weak moves, just outright blunders in both cases.

I made it my resolve to NEVER blunder again, at least in shallow obvious ways. Of course I still would make the occasional blunder, but now I was aware of the problem and never made a move without carefully checking it for anything that was an obvious blunder. In about a year I was suddenly a 1900 player. I'm sure some of that was due to study, but I attribute most of it to not beating myself.

Some of the early computers achieved ratings that were hard to reconcile with their obvious weaknesses. It would have a rating of 1800 or 1900 and you would play a few moves and notice that it didn't seem to be playing anywhere near 1800 strength. The reason why it seemed this way is that people never consider the blunder factor. It's a lot harder to beat an opponent who never makes an obvious error and who never misses YOUR obvious errors.

Players would lose due to an error, then heap verbal abuse upon the poor machine. It was funny to watch someone who just lost to an opponent loudly proclaim how horrible they played - they just make fools out of themselves. Deep down inside nobody believes those games should count. "It's not like me to do that!" A lot of people back then felt they were jinxed by computers, they would say stupid things like, "I don't know what it is about playing computers, but I seem to blunder all the time when playing them."

The truth of the matter is that they DID blunder all the time and just didn't realize it because weak human players often simply miss your blunders.

I guess what I'm saying is that I don't know how to convert number of blunders or second best moves to a rating. But I believe just 1 or 2 errors per game is going to reduce your ELO by 300 ELO or more. Let me define "error" as a bad move that is well within your ability to recognize, you just didn't.

What I COULD provide is an analysis tool that assigns some kind of score to the quality of the moves you played or at least provides statistics. Of course this would be based on the opinion of the program, which is far from infallible! But it would probably work well for players significantly weaker than the computer, which is most of us here.

User avatar
fern
Posts: 8755
Joined: Sun Feb 26, 2006 3:07 pm

Re: Komodo 1.3 Official Release Available

Post by fern » Wed Dec 29, 2010 7:27 pm

Seems to me a great idea.
yes, we human beings are prone to blunder in chess as in everything.
problem is that a mistake use to be more consequential than many hits. Life...


fern

User avatar
fern
Posts: 8755
Joined: Sun Feb 26, 2006 3:07 pm

Re: Komodo 1.3 Official Release Available

Post by fern » Wed Dec 29, 2010 7:39 pm

apropos of number of errors, yes, one or two are enough. i have observed on my games against strong engines that always the history is the same; I play reasonably well but around move 15 or at most 20 i commit one mistake, the first, sometimes a full pawn, usually a positional mistake and that is enough, the machine get a rapid advantage, I disappoint with myself and so I play more uncarefully so other mistakes comes and what seemed to be, at move 15, a game really well played becomes very fast in a defeat at move 25 or so, perhaps 40 or 50 if I fight the same, determinate to sell expensively my defeat. .
Another thing that get steam out of you is that you know well that combinations or traps or "brilliant" moves will not deliver. The bloody iron monster will see each of those stylish manoeuvrings and nothing will come of it, except unhinging your position.

Fern

Darkmoon
Posts: 191
Joined: Mon Dec 14, 2009 11:48 pm

Re: Komodo 1.3 Official Release Available

Post by Darkmoon » Thu Dec 30, 2010 6:16 am

Forgive me if this is a repetitive request-and, I would imagine you have had others ask this of you-

What is preventing you from developing your own website? I think it would be great to read up on your progress- and other relevant computer chess issues that you and Larry might be interested in expounding upon.

If you don't have the time and if you are perhaps still actively teaching or, at the very least, in the proximity of a university- you could easily find a knowledgeable student willing to do it for you.

Robert Flesher
Posts: 1246
Joined: Tue Aug 18, 2009 1:06 am

Re: Komodo 1.3 Official Release Available

Post by Robert Flesher » Fri Dec 31, 2010 6:16 pm

fern wrote:One feature that no program implement is a real rating evaluation tool. It is not realistic to be evaluated like a 800 Elo player because you loses every game against an engine even if you played lot of very good moves, proper of a, say, 1800 or more player.For doing what i propose it is enough with not only take into account the result, but the number of moves that the player did that coincided with the best move calculated by the engine.
So you could lose every game and still to be considered, say , a decent chess player.
We all need that incentive.....

My best and waiting your simple,. intuitive and ORIGINAL gui.
Fern

This is a very good point Fernando, I to would love to see this feature in a GUI. I am approx 2100-2150 strength, yet after losing many games to 3000+ engines my rating plummets, to silly levels. Also, perhaps a replay of your game(s), suggesting better moves, and ideas.

User avatar
fern
Posts: 8755
Joined: Sun Feb 26, 2006 3:07 pm

Re: Komodo 1.3 Official Release Available

Post by fern » Sat Jan 01, 2011 3:23 pm

Happy to see, Robert, that my logic on this is not completely lost or silly.
I am around your same rating and it is not funny to see me as a 800 player because, as it is expect-able, I lose every serious game -no takebacks- against a 3000 monster.
That is to add injury to insult, as it is said...

happy new year
Fern

Mincho Georgiev
Posts: 454
Joined: Sat Apr 04, 2009 4:44 pm
Location: Bulgaria
Contact:

Re: Komodo 1.3 Official Release Available

Post by Mincho Georgiev » Sat Jan 01, 2011 4:23 pm

Robert Flesher wrote:
fern wrote:One feature that no program implement is a real rating evaluation tool. It is not realistic to be evaluated like a 800 Elo player because you loses every game against an engine even if you played lot of very good moves, proper of a, say, 1800 or more player.For doing what i propose it is enough with not only take into account the result, but the number of moves that the player did that coincided with the best move calculated by the engine.
So you could lose every game and still to be considered, say , a decent chess player.
We all need that incentive.....

My best and waiting your simple,. intuitive and ORIGINAL gui.
Fern

This is a very good point Fernando, I to would love to see this feature in a GUI. I am approx 2100-2150 strength, yet after losing many games to 3000+ engines my rating plummets, to silly levels. Also, perhaps a replay of your game(s), suggesting better moves, and ideas.
Perhaps you could try Pawny for that level and I would be really glad to know if it fits into that boundaries of strength.

Congrats Don for the new version!

Post Reply