Why is Houdini 4 so clueless?

[sicilianquake87]

In my opinion, Houdini 4 will not lose a game in the manner described by the OP, given good hardware, reasonable amount of time and EGTBs.
Is what I'm saying too complicated to understand ?

What's difficult to understand is why you consider that a reasonable statement to make. Given infinite time any engine plays perfect chess.

Was Houdini playing on inferior hardware compared to Stockfish? No. Was it playing at time-odds? No. Was stockfish using EGTBs while Houdini wasn't? No. Was it at a disadvantage from Stockfish? No.
In this particular game (and perhaps under these particular conditions) Stockfish played better than Houdini. Houdini is very strong, but clearly not unbeatable - it misevaluates sometimes and it does miss tactics. That should be obvious anyway, since it would occasionally lose games (to Stockfish, for instance). If that happens more now this is because Stockfish improved and is better tuned than it used to be.

Also fair to remember: Houdini 4 does not improve, it is static. The "current" version of Stockfish has seen an enormous improvement since the version of Stockfish that was current when Houdini 4 came out. Comparing current Stockfish to Houdini 4 is comparing apples and oranges; you should compare with the current development version of Houdini, which no-one has access to (and which may not have improved as much as Stockfish since it's hard to beat the computer resources available to the development of Stockfish, but that's impossible to say for sure).

TCET will give us a verdict soon.

That is unfair to Houdini's development version to call TCEC a verdict. TCEC is not designed for that reason. To give a reliable rating, or verdict on any program. Houdini development version could be the strongest engine but still lose TCEC.

Let me guess: maybe this is the reason I wrote a verdict and not the verdict?

A verdict is the formal finding of fact. The verdict is the outcome of such fact. And yet with all your subtleties, you are not wise.

[Dhanish]

I tried Finalgen, and it says the final position is won for White:
[pgn]
[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "-"]
[White "?"]
[Black "?"]
[Result "*"]
[FEN "8/8/3k4/1B3p1p/3K1P1P/6P1/8/3b4 w - - 0 1"]
[SetUp "1"]

1. Bd3 Bg4 2. Bc2 Ke6 3. Kc5 Bf3 4. Bb1 Bd1 5. Ba2+ Ke7 6. Kd5 Kf6 7. Kd6 Ba4 8. Bc4 Bd1 9. Kd7 Ba4+ 10. Kd8 Bc6 11. Be2 Kg6 12. Ke7 Ba4 13. Bc4 Kg7 14. Ke6 Kg6 15. Bd3 Bb3+ 16. Ke5 Bd1 17. Bxf5+
[/pgn]

[Michel]

I tried Finalgen, and it says the final position is won for White:

That's impressive since it is an 9 piece endgame!

Although I can see how it is done (to prove a win for white one only has to consider certain pawn slices, with 4 movable pieces, which makes it tractable), it is still nice that Finalgen can do this in an automatic fashion!

PS. I now see that your pv ends with white capturing the f pawn. Did you simply ask Finalgen to prove that white can capture a pawn?

[Dhanish]

Did you simply ask Finalgen to prove that white can capture a pawn?

There is no such option in Finalgen. I cut off the PV when a pawn was won. Actually, Finalgen calculates till a pawn is promoted. In this position, it said "White wins in 34", and did not calculate the moves till mate.

Of course, Finalgen is a great programme. It took almost 7 hrs to solve this position. But it is 32 bit and can utilize only one core at a time. The author tried to make it multiprocess, but failed.

Hope somebody will take up this project and develop it further!

[shrapnel]

I am confused, at what point and speed does Houdini 4 stop losing games to Stockfish?

And how much of a speed advantage does Houdini 4 need over Stockfish?

And what is the fastest time control that Houdini 4 can play at to avoid making mistakes against Stockfish?

All the answers to your questions can be found here http://talkchess.com/forum/viewtopic.php?t=50943.
As is obvious, Houdini 4 makes BETTER use of top-class Hardware, like that of Riddick, compared to Stockfish. Also, at the LTC that Riddick uses, Houdini is clearly better..... Stockfish also gets more time like Houdini 4, but Houdini 4 is +3 up at this stage. Clear ?
Keep playin' your Blitz games if it makes you happy , but as you can see from Riddick's results, your games are perfectly meaningless actually !

[mwyoung]

Keep playin' your Blitz games if it makes you happy , but as you can see from Riddick's results, your games are perfectly meaningless actually !

So if I would stop testing only blitz games then I would see that Houdini 4 is clearly better then Stockfish.

Because blitz results are meaningless now, I guess because Stockfish is now winning against Houdini 4 in blitz.

Sounds like a man grasping at straws.

What about this results for CEGT played at 40/120 this has the old Stockfish DD ahead of Houdini 4.

This was not played at blitz time controls.

Code: Select all

CEGT 40/120 Ratinglist

1 Komodo TCEC x64 3070 25 25 300 61.0% 2993 58.7% 
2 Stockfish DD x64 3068 23 23 300 60.7% 2993 64.0% 
3 Houdini 4.0 x64 3065 26 26 300 60.2% 2994 56.3% 

or this match results from CCRL played at 40/40. Stockfish beat Houdini 4 head to head 54.5 to 42.5.

Code: Select all

 Stockfish DD - Houdini 4 64-bit 4CPU 3267 +20 −20 (+20) 54.5 − 42.5(+29−17=51) 

Looks like you are a H4 fan only, and your statements are clearly skewed to what every favors H4 in your mind.

[ernst]

Thanks Anil for pointing to that topic. Looking at the last post on page one and I see exactly what I am talking about, but now in a slow game.

In the 27th round, Houdini probably lost due to a bug. After 49...Qb7 evaluation and depth were abnormal until 56...Nxe5:

49.Nd4 -0.28/32 3:23 Qb7 -1.63/40 2:02 50.Kh3 -0.11/32 2:28 Rc5 -1.63/42 4:58 51.Nb3 -0.11/35 2:05 Rc2 -1.59/44 10:04 52.Qf3 -0.00/35 2:31 Ra2 -1.71/43 7:11 53.Bd4 -0.00/35 2:47 Nc6 -1.91/37 1:23 54.Bc3 -0.00/41 2:50 Ra3 -4.95/40 1:19 55.Nd2 -0.11/57 3:49 Bb4 -5.31/42 2:13 56.Qe2 -4.59/30 9:24 Rxc3 -6.12/41 1:31

Also in the output engine window were displayed two moves only, from 52.Qf3 to 55.Nd2. Houdini's depth jumped from 35 to 57 (impossible depth in this position for Houdini) in less than 4 minutes! I think that something is wrong... What do you think about it?

The 28 games about Bird Opening are finished.

HOUDINI: Points = 13.5/28; Average Speed = 19550 kN/s; Average Plies = 29.3.
STOCKFISH: Points = 14.5/28; Average Speed = 12598 kN/s; Average Plies = 39.0.

[shrapnel]

The attitude of some people here reminds me......" You can take a horse to the water, but you can't make it drink ! "

[shrapnel]

http://www.talkchess.com/forum/viewtopic.php?t=50943
Houdini 4 just increased its lead to + 4 ; that too with the BLACK pieces

GO GO GO Houdini ! One MAN is worth more than a 100 little boys !
ALL HAIL ROBERT HOUDART AND RIDDICK'S TOP-CLASS HARDWARE ! Stockfish Development Team should maybe take up fishing as a hobby and leave computer chess development to to the Professionals like Houdart and Kaufmann !

[michiguel]

http://www.talkchess.com/forum/viewtopic.php?t=50943
Houdini 4 just increased its lead to + 4 ; that too with the BLACK pieces

GO GO GO Houdini ! One MAN is worth more than a 100 little boys !
ALL HAIL ROBERT HOUDART AND RIDDICK'S TOP-CLASS HARDWARE ! Stockfish Development Team should maybe take up fishing as a hobby and leave computer chess development to to the Professionals like Houdart and Kaufmann !

[MODERATION]
It is nice not to be calling other members little boys, fan boys etc. etc. It generates a bad atmosphere and things escalate. We have seen it.
A little friendly taunting once in a while is fine, but if it becomes a pattern it abusive.
We received complaints already, from people that are refraining to escalate.

Miguel