What is stronger than Houdini 3 for what?

[MM]

Now that we have tested Houdini 3 to an extent, by now,

Can we get a clearer idea to this question?

Is Houdini 3 the strongest engine in all cases at this point in time?

_I_ got the impression that there is one case that there is a stronger engine, e.g.

Houdini 1.5 vs Rybka 4.1.

Is this indeed stronger than

Houdini 3 vs 4.1?

(in general i refer to equal hardware, plus, longer time controls than blitz).

Thank you, that is an interest argument.

It is evident that more that the time control increases and more Houdini loses its strenght. It is written in the lists and in the tests games and it is completely logical.

That is because Houdini has a well known approach to chess, curiousily very similar to robbo-ippo era, then very fast search to find good moves, especially tactical moves, in a short time.

That's a huge advantage in blitz and in medium time control but at long time control almost every engine at a high level has enough time to see what Houdini often sees in a very short time.

I think there are other engines that play much better from a point of view of positional play, or if you prefer, with ''armony'', like the former world champion Smyslov used to say. But their problem is the tactics, especially in short time control (for example Shredder).

I think that Rybka is the best competitor at long time control because it has a deep way to analyze and it has also skill in tactics, if it has enough time. More, Rybka has a very good knowledge about endgames and all this hurts Houdini.

I wouldn't be surprised if at 120/40 repeated Rybka should be less than 30 elo to Houdini or even better.

Also Komodo would be a good competitor, i think probably the MP version of the actual best version of Komodo should be very close to Houdini at 120/40.

I think so because the difference in strenght reduces so much moving from 40/4 to 40/20 or 40/40 and from 1 core to 4/6 core that, going with this trend, at 120/40 on MP probably Houdini would lose most of its advantage.




Best Regards

Nonsense. From all what I saw, Houdini 3 is one of the best scaling with time top engine. Houdini 2 did scale worse than some top engines, but this is corrected with Houdini 3, which is undisputed leader at long time controls and probably scales even better than some other top engines. Look up the lists, matches, hardware and so on before writing your outlandish statements.

Kai

Thank you for the nice words. I wonder why you don't post more often so everyone could understand how to be polite and learn many other things from you.


From CCRL list, 40/4 - pure list (do you know what it is?)

http://www.computerchess.org.uk/ccrl/40 ... _pure.html

Houdini 3 - 4 cpu 3353
Critter 1.6a - 4 cpu 3254 (-99)
Stockfish 2.3.1 - 4 cpu 3247 (-106)
Rybka 4.1 - 4cpu 3243 (-110)


From CCRL list 40/40 - pure list


http://www.computerchess.org.uk/ccrl/40 ... _pure.html

Houdini 3 - 4 cpu 3248
Stockfish 2.2.2 - 4 cpu 3183 (-65)
Critter 1.2 - 4 cpu 3180 (-68)
Rybka 4.1 - 4 cpu 3171 (-77)

So Stockfish has +41, Critter has +31, Rybka has +33.


Don't like pure list?

No problem.

CCRL 40/4 complete list

http://www.computerchess.org.uk/ccrl/40 ... t_all.html

Houdini 3 - 4 cpu 3334
Critter 1.6a - 4 cpu 3235 (-99)
Stockfish 2.3.1 - 4 cpu 3230 (-104)
Rybka 4.1 - 4cpu 3217 (-117)

CCRL 40/40 complete list

http://www.computerchess.org.uk/ccrl/40 ... t_all.html

Houdini 3 - 4 cpu 3287
Critter 1.6a - 4 cpu 3207 (-80)
Rybka 4 - 4 cpu 3199 (-88)
Stockfish 2.2.2 - 4 cpu 3199 (-88)

Critter has +19, Rybka has +29, Stockfish has +16.


It is evident (not for you i think) that Houdini 3's advantage on the other engines diminishes moving from 40/4 to 40/40.

Some versions of Critter, Stockfish and Rybka seem to perform better than others at longer time control but this doesn't move the problem then all main competitors of Houdini 3 gain several points when the time control increases.

40/40 is not certainly a long time control, long time control is 120/40, for example, so it's logical to believe that the gap should diminish more.




Not enough?

CEGT 40/4 2 cpu (Houdini 3 - 4 cpu is not in the list)

http://www.husvankempen.de/nunn/40_4_Ra ... liste.html


Houdini 3 - 2 cpu 3134
Critter 1.4 - 2 cpu 3023 (-111)
Stockfish 2.2.2 - 2 cpu 3015 (-119)
Rybka 4 - 2 cpu 3008 (-126)



CEGT 40/20 ponder OFF 4 cpu

http://www.husvankempen.de/nunn/40_40%2 ... liste.html

Houdini 3 - 3154
Critter 1.6 - 3061 (-93)
Stockfish 2.3 - 3053 (-101)
Rybka 4.1 3047 (-107)

So Critter has +18, Stockfish +18, Rybka +19.

Of course 40/20 is rapid but you can see (?) that the gap diminishes moving from 40/4 to 40/20.

In this case too some versions of Stockfish, Rybka and Critter perform better than others but in any case their best version reduces the gap.


Is it still a nonsense?

http://www.husvankempen.de/nunn//40120n ... liste.html

This is, as far as i know, the most recent rating list at 40/120.

Houdini 3 is not included but look how Stockfish 2.2.2 is even +1 to Houdini 1.5 when in blitz Houdini 1.5 destroys easily Stockfish and all the other engines.

If you see (?) all top engines are very close.

Note (?) that also Komodo 5, Critter 1.6, Stockfish 2.3.1 and Rybka 4.1 are not in the list.

Still nonsense?

If this is not enough i invite you to observe sometimes the games, perhaps you will understand that the tactical ability of Houdini loses its strenght more that the time control increases, exactly as it happens in human games. Who plays chess (do you?) knows that there are many chess players very strong in bullet and blitz but sometimes their strenght diminishes dramatically when the time control becomes longer.


The reason is the same.


Best wishes, really.

Look up at the number of games, error margins, other matches played and posted here and elsewhere. You selectively choose what is convenient (where is CEGT 40/20 ponder ON and many other results?). Keep in mind that 100 points at 40/4 translate (with equal scaling) in some maybe 80 at 40/40 because the number of draws increases. Equal scaling doesn't mean that 40/4 and 40/120 results have the same Elo difference, and it is better to talk of time handicap to get the engines equal in strength. And I am talking specifically of Houdini 3, which scales better than Houdini 2, and probably on a par with Rybka, Critter, etc.

As to your chess abilities to judge the subtleties in quality of Houdini and Rybka play at long and short TC, they must be awesome.

Kai

I posted the most relevant informations i have, i didn't choose, i noticed and posted.
I posted also the links, so everyone can investigate.
The error bar ''savage'' cannot be applied when all (all) the indications drive to the same conclusions. It is just a question of ''how much'' not ''if''.

I could post also the recent match of Clemens Keck:

http://www.clemens-keck.de/base.htm

I cannot know how many points H3 could lose at 120/40 unless someone will play 10 thousand games at 120/40 but i' totally confident that the gap actually present in 40/4 would be reduced dramatically and i'm sure of it not only cause the indications that i posted but especially because it is absolutely logical and i already explained why so i won't repeat.

I didn't post ponder on of CEGT because it would have been a distortion, because all the other data was ponder OFF and i could not make a comparison because cegt has ponder ON only for 40/20, not 40/4.

Which subtleties are you talking about? No need (for anyone) to play at the level of Houdini to understand in which way he wins, draws or loses. Just watch a game and see, score helps. For example you understand, even if you are a weak player, if Houdini wins with an attack to the king, with a won endgame with a plus pawn, with a good knight against a bad bishop ect.

As regards Houdini 3 i think it scales better than 2.0 but i didn't investigate on that, it is an argument in which i'm not interested, but i don't think it is at Rybka level. There are some things that cannot be reduced to ''data'', ''games'', ''numbers''. There must be a minimum of knowledge and understanding. Houdini plays mainly tactically, tactics is good especially in fast games, it helps less in longer games, where strategy and or positional play becomes very important.

That's why generally Komodo has better scores or had better scores in the recent past with a longer time controls.

Regards

[syzygy]

Still nonsense?

Yes.
It's unsound to cherry-pick rating lists to make a point.
It's unwise to discuss rating list results without even once mentioning their error margins.

And the two reinforce each other: once you start cherry-picking, error margins lose their relevance. Even if you use a 99% confidence interval, 1 in 100 experiments will on average return incorrect results. Cherry-picking that experiment will turn the 99% confidence into something like 1% confidence.

[MM]

Still nonsense?

Yes.
It's unsound to cherry-pick rating lists to make a point.
It's unwise to discuss rating list results without even once mentioning their error margins.

No way, i posted the links, all data is available. I could say it's not elegant to enter a discussion only to try to defend a position, trying to hide the truth written on a paper.

It's not certainly me to have an interest to say this or that and i don't talk randomly.

The error margins apply, in this case, just in order to figure ''how much'', not ''if''.

Best Regards

[MM]

It is clear, from TCEC (games which you can see very clearly, in the archives of Chessbomb and elswhere), that Houdini 1.5 is clearly superior to Rybka 4.1, at long time controls.
There is no doubt about it. THAT Houdini is far better than Rybka 4.1 in a match between the two. Endgame or not, anything else or not.

But from all you have said, i don't yet see proof that Houdini 3 is better than Houdini 1.5, or from Houdini 2, or from every other program, and that with just a little tweak from something, something else can clearly dominate Houdini 3.

Is this true, or not?

Or will Houdini survive everything with a few elo points handy?

If none of the rating lists can convince you, what will?

Robert

First you basically mean that i'm not accurate to say something that i estrapolate from CEGT and CCRL with links and now you say that the same lists and others show the truth about the superiority of Houdini 3?

Why don't you treat these two cases in the same way?

Regards

[Laskos]

That's why generally Komodo has better scores or had better scores in the recent past with a longer time controls.

Regards

I am pretty tired of you blabbering of "facts". I would just say: that's why Houdini 3 will win comfortably a long match (consisting of many games) against Komodo 5 or Rybka 4.1 at every chosen time control (no matter how long) on any chosen equal hardware. And I mean - comfortably.

Kai

[syzygy]

First you basically mean that i'm not accurate to say something that i estrapolate from CEGT and CCRL with links and now you say that the same lists and others show the truth about the superiority of Houdini 3?

Why don't you treat these two cases in the same way?

Just look at what was being said:

But from all you have said, i don't yet see proof that Houdini 3 is better than Houdini 1.5, or from Houdini 2, or from every other program, and that with just a little tweak from something, something else can clearly dominate Houdini 3.

[MM]

That's why generally Komodo has better scores or had better scores in the recent past with a longer time controls.

Regards

I am pretty tired of you blabbering of "facts". I would just say: that's why Houdini 3 will win comfortably a long match (consisting of many games) against Komodo 5 or Rybka 4.1 at every chosen time control (no matter how long) on any chosen equal hardware. And I mean - comfortably.

Kai

I'm tired too to read you that quote me without reason and telling that i blabber. What is your point exactly? Komodo doesn't have better scores? Komodo does have horrible scores? What about Komodo? What about me?
Can you explain please?

I contributed this discussion with infos and links, i explained my point of view but you have always something to say on everything. Why?

[S.Taylor]

It is clear, from TCEC (games which you can see very clearly, in the archives of Chessbomb and elswhere), that Houdini 1.5 is clearly superior to Rybka 4.1, at long time controls.
There is no doubt about it. THAT Houdini is far better than Rybka 4.1 in a match between the two. Endgame or not, anything else or not.

But from all you have said, i don't yet see proof that Houdini 3 is better than Houdini 1.5, or from Houdini 2, or from every other program, and that with just a little tweak from something, something else can clearly dominate Houdini 3.

Is this true, or not?

Or will Houdini survive everything with a few elo points handy?

If none of the rating lists can convince you, what will?

Robert

The rating lists don't always show what is about to errupt. YOU would be the one who knows better.
Doesn't Houdini 1.5 make some moves stronger than Houdini 3, when playing Rybka 4.1, or which Rybka 4.1 can detect weaknesses in?

And do you say that Maurizio Maglio makes wrong conclusions from rating lists? I would believe what YOU say.

[Houdini]

First you basically mean that i'm not accurate to say something that i estrapolate from CEGT and CCRL with links and now you say that the same lists and others show the truth about the superiority of Houdini 3?

Why don't you treat these two cases in the same way?

Regards

You cherry-pick and extrapolate rating list results to make unverifiable claims about very long TC behavior.
I ask the OP to take a look at all the rating lists to see whether there is any ground for a claim that Houdini 3 is far superior to Houdini 1.5a.
There is a methodical difference, I hope you can spot it.

Robert

[MM]

It is clear, from TCEC (games which you can see very clearly, in the archives of Chessbomb and elswhere), that Houdini 1.5 is clearly superior to Rybka 4.1, at long time controls.
There is no doubt about it. THAT Houdini is far better than Rybka 4.1 in a match between the two. Endgame or not, anything else or not.

But from all you have said, i don't yet see proof that Houdini 3 is better than Houdini 1.5, or from Houdini 2, or from every other program, and that with just a little tweak from something, something else can clearly dominate Houdini 3.

Is this true, or not?

Or will Houdini survive everything with a few elo points handy?

If none of the rating lists can convince you, what will?

Robert

The rating lists don't always show what is about to errupt. YOU would be the one who knows better.
Doesn't Houdini make some moves stronger than Houdini 3, when playing Rybka 4.1, or which Rybka 4.1 can detect weaknesses in?

And do you say that Maurizio Maglio makes wrong conclusions from rating lists? I would believe what YOU say.

Please correct me if i understood bad. I understood that you basically doubt that Houdini 3 is indeed stronger than Houdini 1.5a when playing at LTC against Rybka 4.1?

Best Regards