Houdini 3 released

[tomgdrums]

Big congratulations to Mr. Houdart. It's clear that the tremendous strength improvement didn't happen by tuning values, but by adding new chess knowledge, search improvements, etc.

I don't see any tremendous strenght improvement, for now.
Basing on the most reliable (in my opinion) rating list, ipon chess, actually H3 is rated 38 elo better than H2. Of course it would be a good gain as well but i wouldn't call it in that way, at least for now.

http://www.inwoba.de/index.html

Best Regards

It's too early to conclude based on the IPON results. But at the very top even +38 ELO is a tremendous improvement. It's very difficult to achieve and requires a lot of work and ridiculous amounts of testing.

I think you are certainly right when you say that at this high level it is hard to improve the strenght of an engine.

But IMO i think that, considering 13 months and especially considering the ''relative'' positional play weakness of Houdini 2, H3 could have been a bit stronger than what it appears.

Of course it is too early to jump to conclusions and perhaps it's me to want too much, but i expected more.
Best Regards

How much stronger do you want it? It is (at the moment at least) scoring over 60% against every opponent on the IPON.

[fern]

Kostas, Mr Houdart just run a software little company, not the European central bank. It is there where you should throw your complain just to see if your petitions can be taken into account. Perhaps it will be scarcely noted in the middle of so many coming from your side of the world due to the unfair treatment you have received, but we never know.

Just a tip

Fern.

[bob]

Hello Robert:

First of all, congratulations for the huge improvement of Houdini 3 over Houdini 2.0c: it is outstanding! Houdini 3 is probably the most wanted engine of the year.

If I go to www.cruxis.com I still read Houdini Chess Engine (version 2.0) instead version 3 or 3.0... just an observation.

I think this is not the best subforum to ask it, but here I go. I quote a paragraph from Houdini site:

The engine evaluations have been carefully recalibrated so that +1.00 pawn advantage gives a 80% chance of winning the game against an equal opponent at blitz time control. At +2.00 the engine will win 95% of the time, and at +3.00 about 99% of the time. If the advantage is +0.50, expect to win nearly 50% of the time.

I guess that I must understand this assesment as the winning ratio (wins/games) without including the draw ratio (draws/games), just in view of if the advantage is +0.50, expect to win nearly 50% of the time: with +0.5 evals, Houdini 3 should score around 50% + draw_ratio/2 under the conditions you wrote.

IIRC, I read something like the following a while ago, written by you: the winning probability µ = (wins + draws/2)/games were similar to a logistic function with Houdini evals: µ ~ 1/[1 + 10^(-0.75*eval)]... has the approximated equation of µ change a lot in Houdini 3? Thanks in advance and sorry for all this stuff in the General Topics subforum.

Seems to fit now 1/[1 + 10^(-eval * 2/3)].
Curious, is this stable over all stages of a game, i.e. opening, middlegame, endgame? It seems to me that 0.70 in the opening is usually better than 0.70 in the endgame with past engines.

Ajedrecista.

I was actually thinking exactly the opposite. +0.70 in the opening can be caused when white castles and black can't, just yet. But eventually black does and that +.70 (or much higher, even) dissipates. But in the endgame, you are talking basically about pawn structure, things like candidates or passers, or distant passers, and those advantages tend to be more "real" than "imagination".

Just my $.02 of course...

[Leto]

Big congratulations to Mr. Houdart. It's clear that the tremendous strength improvement didn't happen by tuning values, but by adding new chess knowledge, search improvements, etc.

I don't see any tremendous strenght improvement, for now.
Basing on the most reliable (in my opinion) rating list, ipon chess, actually H3 is rated 38 elo better than H2. Of course it would be a good gain as well but i wouldn't call it in that way, at least for now.

http://www.inwoba.de/index.html

Best Regards

It's +46 now after 1152 games at IPON. I'd say that's a tremendous strength improvement.

[Ponti]

Komodo 5 runs with just 1 core and is only 25 points behind Houdini.

When will Komodo multicore version be released ?

[Uri Blass]

Komodo 5 runs with just 1 core and is only 25 points behind Houdini.

When will Komodo multicore version be released ?

Not correct.
Komodo5 is clearly more than 25 elo behind houdini3 and if you mean only 25 elo behind houdini2 then you should mention the version number because people who read you think about the latest version.

[tomgdrums]

Komodo 5 runs with just 1 core and is only 25 points behind Houdini.

When will Komodo multicore version be released ?

If you see below the IPON playing conditions state that all engines are running on 1 core!

So Whatever gain Houdini has on Komodo it is all based on 1 core at this time on the IPON list.

Playing conditions:

No anonymus engines or engines with deliberatly 'false' programmer names are accepted!

Ponder : ON
CPU : 3.2 GHz Phenom2 (x6)
Operating system : XP-64
Cores : All Engines with 1 core
Hash : 256 MB
Table-/Shredder-/Bitbases : 4 pieces
Cache for Tbs : 64 MB
Time control : 5 min/game + 3 sec/move (mean game length ~ 16 Minutes)
Openings : 75 different opening positions (no books) (50 positions before 2012.01.28)
Positions : Changing colors (150 games engine vs engine)
Learning : OFF
Bayeselo : mm 0 1 (draw rate consideration)

[notyetagm]

+48 after 1247 games out of 2550.

[lkaufman]

Komodo 5 runs with just 1 core and is only 25 points behind Houdini.

When will Komodo multicore version be released ?

As soon as MP version passes our tests it will be released within a week, but I can't predict how soon that will be. The current SP version is equal in strength to Houdini 2, as nearly as can be measured.

[Uri Blass]

After only 10 minutes, I received the mail with the link and Serial-Code - nice.

Test-gauntlet for my LS-rankinglist is in progress. Final result (hopefully) on sunday (evening)...

Nodes/second seems little bit increased (+5-10%).

Best - Stefan

I ordered today and I am not so lucky to get a link in 10 minutes.
After 1 hour I stilll did not recieve a link to download houdini and it is not in my spam filter.

I only recieved an email from paypal with the title
"Receipt for Your Payment to Robert Houdart".