I Am Running Strelka 5.6 tests- Promise Accurate Results!

[Dan Astrachan]

Thanks Georges for the info. I'm eager to play with it too

[Uri Blass]

I use the same Sedat generic book for all- with a 12 move limit. But Yuri has come up with an extremely novel way of playing the opening. I have never seen anything like it ever done before. I tell you this because I got more hung up watching that work than anything else.

Best regards,

gts

If the novel way is using a hidden opening book based on database then I do not care about the results because it seems that they are not relevant for correspondence games and the only relevant way to test is to give it positions that people never played as starting positions(for example the position after 1.a3 a6 2.h4 h5).

[geots]

I use the same Sedat generic book for all- with a 12 move limit. But Yuri has come up with an extremely novel way of playing the opening. I have never seen anything like it ever done before. I tell you this because I got more hung up watching that work than anything else.

Best regards,

gts

If the novel way is using a hidden opening book based on database then I do not care about the results because it seems that they are not relevant for correspondence games and the only relevant way to test is to give it positions that people never played as starting positions(for example the position after 1.a3 a6 2.h4 h5).

Uri, who woke you up? I bet you were resting well. ( It doesn't have an opening book- hidden or not.) Try to get some more rest.



Take care-

gts

[geots]

I very much apologize, and time has gotten away from me. I have been in the bed a number of hours with my back acting up again. It would not be so bad except for the fact early on the electr. was off for about 30 min. and ruined my matches, unless 12 to 16 games would suffice.

I will have some interesting results- please have a bit of patience with me here. It is very aggravating.



Best,

gts

[geots]

Strelka 5.6 vs. Rybka 4.1, Stockfish 2.3.1, Critter 1.6a and Houdini 2.0c are now running with one important change. Why the change? Having time to reconsider, it just seemed like the thing to do: Time control was and still is 4'+2", and Rybka and Stockfish are still 100-game matches.

But maybe just based on ratings I felt the Critter and Houdini matches to be the most important to date, so I made the change and they are both now 200-game matches. So we shall see what we shall see.



Over and out-

gts

[geots]

As for Rybka, he is holding close as Rybka is known to do. Stockfish seems to be having a bit of trouble as we speak.

But it is the other 2 matches that have my attention. As I said, Strelka's matches with Critter and Houdini 2 are 200-game matches. I don't know how to say it without it sounding a bit strange, or a "lot" strange. I have been flipping back and forth from match to match with Critter and Houdini for a couple hours, and I wish I could stay up.

This is without doubt the first time I have seen Houdini hesitant, unsure of himself- and if he were human I would say he is content to play for a draw every game unless he has a huge clear-cut advantage. And Strelka has exposed some weaknesses in his game. Still, after all that- I just don't think Strelka can carry the mail for 200 games. Houdini is still Houdini, and watch out when he gets on a roll. And that is why I say that. As I type:

39 games have been played. Houdini leads Strelka : +6/-5/=28



OTOH, Critter seems more willing to get into tactical battles than Houdini does. Even with that- the draw rate is probably a bit higher in this match.
So here we are:

45 games played. Strelka leads Critter: +6/-4/=35


Prediction: I would not be surprised if Strelka won both matches, yet I would not be surprised if he lost both. (That is for sure hedging my bet)



Nite,

george

[geots]

These 2, 100-game matches have concluded. The 200-game matches with Houdini and Critter continue. I would not think there is much I need to add to these results, except I cannot remember the last time I ran a match where Rybka was just completely blown away.





XPS 8500 Intel i5-4 True Cores
Fritz 11 gui/Fritz 13 gui
1CPU/64bit
128MB hash
Bases=NONE
Ponder_Learning=OFF
Perfect 2012b.ctg w/12-move limit
4'+2"
Match=100 games

Code: Select all

XPS-Core-i5, Blitz:4'+2"  2/16/2013  

                     
Strelka 5.6 x64   +24    +26/-19/=55   53.50%   53.5/100
Rybka 4.1 x64     -24    +19/-26/=55   46.50%   46.5/100

Code: Select all

XPS-Core-i5, Blitz:4'+2"  2/16/2013  

                              
Strelka 5.6 x64            +108    +37/-7/=56   65.00%   65.0/100
Stockfish 2.3.1 pcnt x64   -108    +7/-37/=56   35.00%   35.0/100

Later with more-

george

[geots]

For the results, there is always a reason "they are what they are". Until I made a direct comparison based on CCRL's most recent results, I had no idea what we were really looking at. Try this:



1CPU, 40/40

Houdini 2.0c x64- 3141
Strelka 5.5 x64- 3120
Critter 1.6a x64- 3108

Now, when you are not an engine that Strelka can just overpower and destroy at 1' or 1 hour- he faces this problem:


1CPU, 40/4 (Very similar to the 4'+2" running now)

Houdini 2.0c x64- 3198
Critter 1.6a x64- 3170
Strelka 5.5 x64- 3136


He has increased 16 elo, but his problem is- look at the increase for Critter and Houdini he faces now.

Now he will increase in strength from 5.5 to 5.6, for sure. But will it be enough to compensate for the fact that Critter and Houdini rate this much better at blitz? We shall see.



gts

[geots]

I am not going to make it until the matches end. Too sleepy. Both have around 50 games left, and based on the scores- if it ended now- plus what I have watched- I think it is a conservative estimate to say Strelka 5.6 is 50 elo stronger than 5.5 at blitz- but probably more than that. But we shall see. Final results after sleep, and then a few more 4'+2" matches, and then we shall go to some of the longer time controls and see what he has in store for us there.



Spraying for bedbugs-

george

[geots]

Both matches have finished. Results soon! And if you believe in CCRL results- which I do- there is little doubt that Strelka is now firmly embedded as the #2 engine in the world. I might consider saying at longer controls it is possible, again "might" say it is possible he could be No.1- except for the fact that possibly being able to beat Houdini 2.0c at longer controls- which he hasn't done yet- doesn't get him there. There is this minor problem of Houdini 3.................... Oh, well- 2nd place aint bad. But he has accomplished something 5.5 could not.


gts