WCCC 2013 in Yokohama

[hgm]

Why Robert opinion? 30 fold speed handicap is about 300 Elo points, more than the advantage Houdini has.

Indeed, in the CCRL 40/40 list Houdini 3 is at 3170, while Deep Junior 13 4CPU is at 3041, Deep Shredder 12 4CPU at 3031, and Hiarcs 13.2 4CPU at 2997. (And you can bet that the latter three will play on 8-core machines!) So there is hardly any rating difference to start with, and that does not even account for the effect of the opening book...

[overlord]

CCRL rating list is very obscure. I don´t trust it. There is definitely much bigger ELO difference according to IPON list: between Houdini 3 and Fritz 13 (253 points), Shredder 12 (259 points), Hiarcs 14 (271 points) and Junior 13.3 (328 points).

[Graham Banks]

CCRL rating list is very obscure. I don´t trust it. There is definitely much bigger ELO difference according to IPON list: between Houdini 3 and Fritz 13 (253 points), Shredder 12 (259 points), Hiarcs 14 (271 points) and Junior 13.3 (328 points).

Strange remark.

All of those engines mentioned have a decent number of games in the CCRL 40/40 rating list.
Also don't forget that IPON testing is 1CPU.

Code: Select all

CCRL 40/40 Rating List - Custom engine selection
460238 games played by 1328 programs, run by 18 testers
Ponder off, General books (up to 12 moves), 3-4-5 piece EGTB
Time control: Equivalent to 40 moves in 40 minutes on Athlon 64 X2 4600+ (2.4 GHz)
Computed on July 6, 2013 with Bayeselo based on 460'238 games
Tested by CCRL team, 2005-2013, http://computerchess.org.uk/ccrl/4040/

Rank                Engine                 Elo   +    -   Score  AvOp  Games
1 Houdini 3 64-bit 4CPU                3249  +16  -16  69.7% -127.6  1301
  Houdini 3 64-bit                     3170  +14  -14  64.2%  -92.4  1692
2 Hiarcs 14 4CPU                       3075  +19  -19  43.4%  +38.5   833
3 Deep Junior 13.3 64-bit 4CPU         3037  +20  -21  40.8%  +60.6   746
4 Deep Shredder 12 64-bit 4CPU         3031  +12  -12  48.9%   +7.1  2422

[overlord]

I have nothing against CCRL rating list, but engine testing without ponder on is simply not correct. It would be very interesting to make statistics how often e.g. Houdini predicts opponents next move and the same statistics for the other engines. I don´t havy any significat data, but my personal feeling is that one of te Houdini strengths is that it predicts opponents move with higer probability (so ponder on testing is essential - when you are playing human OTB you also think on opponents time...).

[Laskos]

I have nothing against CCRL rating list, but engine testing without ponder on is simply not correct. It would be very interesting to make statistics how often e.g. Houdini predicts opponents next move and the same statistics for the other engines. I don´t havy any significat data, but my personal feeling is that one of te Houdini strengths is that it predicts opponents move with higer probability (so ponder on testing is essential - when you are playing human OTB you also think on opponents time...).

Just click on the engine in CCRL lists, it will show ponder-hits. Ponder ON tests are just a waste of resources which are badly needed for testing many engines with as many games as possible.

[geots]

Funny tournament. Houdini running on low-end smartphone would crush them all in 25 moves

It may crush them, but probably not in 25 moves. But Houdini for Andrioid doesn't exist (yet).

Ray, the problem is it ought to be discontinued. It has turned into nothing more than a sick joke. Junior won last time around. In my beta testing for Don, at reasonable time controls- his WEAKEST beta that I tested in 30 game matches, in just wins and losses, beat Junior 13.3 and Deep Fritz 13 by a combined score of 40-0! Neither engine even got a win in 60 games- and I am still amazed they figured out how to draw 20 games between them. Guess they got lucky.

The best analogy I can come up with is that it is time for the World Cup championship game- and both sides come down with the flu. So instead of cancelling it- they run in 2 junior high school soccer teams to play each other for the World Cup.Kids 10 and 11 years old, and still have the gall to insult people's intelligence by attempting to drum up interest for it. And the winner advertises itself as the best team in existence in the world. Which do you do- laugh at them or cry.


gts

[overlord]

After the tournament Hiarcs web page will proudly demonstrate that this engine is chess world champion...some people has sense for good jokes

[hgm]

I don´t havy any significat data, but my personal feeling is that one of te Houdini strengths is that it predicts opponents move with higer probability (so ponder on testing is essential - when you are playing human OTB you also think on opponents time...).

That shows again how little you understand of these matters. If the opponents are much weaker, as you claim, Houdini should have great difficulty predicting their moves. Because Houdini will expect the opponent to play what is the best move in its own opinion...

[overlord]

It is just your opinion, nothing more. Just run the match between Houdini and Stockfish and you will see...

[overlord]

Please don´t forget that if Houdini doesn´t hit weaker opponents move there is high probability that this move is weak. Moreover, ponder on is important, because if both engines miss opponents move, they start from nothing...low searching depths (or limited time as you wish) is favourable for stronger engine (it is well known that weaker engines scores better in long time controls).