Houdini 5 dev in TCEC

[Ozymandias]

A good prediction for the score after 100 games would be something to the effect of 20-5-75 in favour of Stockfish. It won't be some two or three point difference.

That's assuming Houdini will improve no more than SF. I bet for something like 15-5-80.

[JJJ]

Indeed Houdini improved faster than Stockfish or Komodo in almost a year. So maybe it can still improve faster. Who knows. We were all surprise to see Houdini reaching the superfinal. Will we be more surprised ?

[shrapnel]

I don't know why people claim such stupid things. We watch one stage of the final of TCEC and think we can say just how much stronger Houdini development is relative to Houdini 4. 140 Elo is absolutely ludicrous. Houdini definitely looks stronger; we still need to wait for its eventual release before we can test the bad boy for ourselves to see the true gain. The new Stockfish regression test is showing +65 Elo over Stockfish 7 which is absolutely massive. I would not expect Houdini to put up much of a fight in the final - such as was the case with Komodo. A good prediction for the score after 100 games would be something to the effect of 20-5-75 in favour of Stockfish. It won't be some two or three point difference.

Hi Adam
I fully agree. Chances are high that things will turn out EXACTLY as you say.
BUT, I have a hunch Houdini will win !
R Houdart is one of the most talented Chess Programmers in the world and he has the advantage of being able to watch everything that the Stockfish Team is doing and so use it in his own Engine...and maybe even improve on some ideas ; he is so talented.
While Stockfish has no access to any of HIS ideas !
In War and Chess, keeping your weapons secret till the last minute is half the Battle !
See what happened to an over-confident Komodo in that memorable first game against Houdini !

[Henk]

Maybe some engines have made an agreement that they only play draws against each other. Like Petrosian, Geller and Keres did in 1962 5th Candidates Curaçao. So better play matches.

[Dann Corbit]

Code: Select all

  # Name                             1        2        3        4        5        6        7        8   Score   Buch  Sommb
---------------------------------------------------------------------------------------------------------------------------
  1 Stockfish 160716          ******** ====1=1= ==1===== ===1=1=1 =11==1=1 1=1==11= 1=11==1= =1=1=1=1  39.0/56 1480.0 1008.50
  2 Houdini 200716            ====0=0= ******** ======1= =======1 ==1=1=1= =11=11== 11==1=1= ==11=1=1  35.5/56 1508.0 907.75
  3 Komodo 10.1               ==0===== ======0= ******** 1======= =11101== 11====1= ===1111= 1===1=11  34.5/56 1516.0 885.25
  4 Fire 5                    ===0=0=0 =======0 0======= ******** 1======= ==1====1 =====1== =1=1====  28.5/56 1564.0 753.00
  5 Andscacs 0.872b           =00==0=0 ==0=0=0= =00010== 0======= ******** ====0=== =1=11=0= 10==1===  23.5/56 1604.0 624.50
  6 Jonny 8                   0=0==00= =00=00== 00====0= ==0====0 ====1=== ******** ==10=010 1=======  22.0/56 1616.0 590.50
  7 Gull 3                    0=00==0= 00==0=0= ===0000= =====0== =0=00=1= ==01=101 ******** =====0==  20.5/56 1628.0 559.00
  8 Rybka 4.1                 =0=0=0=0 ==00=0=0 0===0=00 =0=0==== 01==0=== 0======= =====1== ********  20.5/56 1628.0 555.00

[reflectionofpower]

Go Houdini! Nice thread.

[Henk]

Code: Select all

  # Name                             1        2        3        4        5        6        7        8   Score   Buch  Sommb
---------------------------------------------------------------------------------------------------------------------------
  1 Stockfish 160716          ******** ====1=1= ==1===== ===1=1=1 =11==1=1 1=1==11= 1=11==1= =1=1=1=1  39.0/56 1480.0 1008.50
  2 Houdini 200716            ====0=0= ******** ======1= =======1 ==1=1=1= =11=11== 11==1=1= ==11=1=1  35.5/56 1508.0 907.75
  3 Komodo 10.1               ==0===== ======0= ******** 1======= =11101== 11====1= ===1111= 1===1=11  34.5/56 1516.0 885.25
  4 Fire 5                    ===0=0=0 =======0 0======= ******** 1======= ==1====1 =====1== =1=1====  28.5/56 1564.0 753.00
  5 Andscacs 0.872b           =00==0=0 ==0=0=0= =00010== 0======= ******** ====0=== =1=11=0= 10==1===  23.5/56 1604.0 624.50
  6 Jonny 8                   0=0==00= =00=00== 00====0= ==0====0 ====1=== ******** ==10=010 1=======  22.0/56 1616.0 590.50
  7 Gull 3                    0=00==0= 00==0=0= ===0000= =====0== =0=00=1= ==01=101 ******** =====0==  20.5/56 1628.0 559.00
  8 Rybka 4.1                 =0=0=0=0 ==00=0=0 0===0=00 =0=0==== 01==0=== 0======= =====1== ********  20.5/56 1628.0 555.00

Maybe some engines have made an agreement that they only almost play draws against each other. Playing only draws would be noticed too easily.

Or maybe they made an agreement to let other engine win. So playing individual matches is better.

[JJJ]

are you trying to be the best ARB here ?

[Henk]

I am only making a statement that if there are (commercial) interests, results of tournaments can not be trusted. TCEC also doesn't check source code. Don't know if they do similarity checks.

[shrapnel]

I am only making a statement that if there are (commercial) interests, results of tournaments can not be trusted.

Very true.
Unfortunately, the Santa Claus and Fairy Tale believers will call you ARB !!
Actually, I find their innocence and simple faith very Touching !