Page 1 of 1

1000-game H6 vs SF-dev LTC match complete

Posted: Thu Dec 21, 2017 12:50 am
by Dan Cooper
50m+10s tc

Score of Stockfish 170912 BMI2 vs Houdini 6.03 C0: 80 - 84 - 836 [0.498] 1000
ELO difference: -1.39 +/- 8.71

https://www.dropbox.com/s/lhnk9iguqaabi ... 6.zip?dl=0

Re: 1000-game H6 vs SF-dev LTC match complete

Posted: Thu Dec 21, 2017 10:08 am
by Ozymandias
Dan Cooper wrote:50m+10s tc
:o How much time did that take?

Re: 1000-game H6 vs SF-dev LTC match complete

Posted: Thu Dec 21, 2017 7:02 pm
by Jouni
Obviously c=0 helps Houdini? With default c=2 it always loses by 10-30 ELO. May be I must re-test with c=0.

Re: 1000-game H6 vs SF-dev LTC match complete

Posted: Mon Dec 25, 2017 7:59 pm
by Jouni
Some rapid games and Houdini with c=0 now:

Code: Select all

                                    
1   Stockfish 191217 64 BMI2   +41  +36/=98/-18 55.92%   85.0/152
2   Houdini 6 x64-pext         -41  +18/=98/-36 44.08%   67.0/152


Re: 1000-game H6 vs SF-dev LTC match complete

Posted: Wed Dec 27, 2017 7:55 am
by beram
Jouni wrote:Obviously c=0 helps Houdini? With default c=2 it always loses by 10-30 ELO. May be I must re-test with c=0.
I examined a big difference in results also in matches between latest SF171219 and Houdini 6 with contempt 0
H6 default got 39,5% where H6 c=0 got 47%(!)

Code: Select all

Houdini 6 - SF171219bf 14x AMD Ryzen 1800, Blitz 3m+2s                                      
1   Brainfish 191217 64 POPCNT   +74  +24/=73/-3 60.50%   60.5/100
2   Houdini 6 Pro x64-popc       -74  +3/=73/-24 39.50%   39.5/100

Code: Select all

Houdini 6(c=0) - SF171219bf 14x AMD Ryzen 1800, Blitz 3m+2s                                      
1   Brainfish 191217 64 POPCNT   +21  +16/=74/-10 53.00%   53.0/100
2   Houdini 6 Pro x64-popc       -21  +10/=74/-16 47.00%   47.0/100

Re: 1000-game H6 vs SF-dev LTC match complete

Posted: Thu Dec 28, 2017 7:33 am
by beram
beram wrote:
Jouni wrote:Obviously c=0 helps Houdini? With default c=2 it always loses by 10-30 ELO. May be I must re-test with c=0.
I examined a big difference in results also in matches between latest SF171219 and Houdini 6 with contempt 0
H6 default got 39,5% where H6 c=0 got 47%(!)

Code: Select all

Houdini 6 - SF171219bf 14x AMD Ryzen 1800, Blitz 3m+2s                                      
1   Brainfish 191217 64 POPCNT   +74  +24/=73/-3 60.50%   60.5/100
2   Houdini 6 Pro x64-popc       -74  +3/=73/-24 39.50%   39.5/100

Code: Select all

Houdini 6(c=0) - SF171219bf 14x AMD Ryzen 1800, Blitz 3m+2s                                      
1   Brainfish 191217 64 POPCNT   +21  +16/=74/-10 53.00%   53.0/100
2   Houdini 6 Pro x64-popc       -21  +10/=74/-16 47.00%   47.0/100
another 100 games same private testset of 25 lines
Houdini c=0 now 46,5%
Please release a new official Stockfisch 9 :)

Code: Select all

Houdini 6(c=0) - SF171226bf 14x, Blitz   3m+2s                                    
1   Brainfish 261217 64 POPCNT   +24  +14/=79/-7 53.50%   53.5/100
2   Houdini 6 Pro x64-popc       -24  +7/=79/-14 46.50%   46.5/100

Re: 1000-game H6 vs SF-dev LTC match complete

Posted: Thu Dec 28, 2017 2:46 pm
by JJJ
It's still a great score for Houdini against Stockfish dev.

Re: 1000-game H6 vs SF-dev LTC match complete

Posted: Sat Dec 30, 2017 2:32 pm
by Jouni
Houdini with c=0 is able to beat SFdev in 100 games:

Code: Select all

                                   
1   Houdini 6 x64-pext         +21  +16/=74/-10 53.00%   53.0/100
2   Stockfish 191217 64 BMI2   -21  +10/=74/-16 47.00%   47.0/100

4 cores, bullet and shortopenings.pgn