Stalling progress?

[melajara]

In absentia of TCEC season 10 to truly assess chess engine progresses, judging from the leader Stockfish 8, it looks like progress has stopped this year.
More precisely, one can judge this when watching the graph here http://spcc.beepworld.de/

Now Komodo 10.4 maintained the momentum. Can Komodo team confirm that progress has not slowed down? In that case, Komodo 10.5 might regain the lead.

Meanwhile, Houdini 5 was an impressive achievement. Now, 6 months later, is there any update to be expected soon?

Ditto for the followers, Fizbo 1.9 and Andscacs 0.9 are still more than 200 Elo behind and I don't see new contenders filling the void just behind the big 3.

I have a definite feeling of slow down in chess engine progress so far in 2017, am I alone feeling so?

Perhaps it is about time to revisit the concept of Giraffe with deep learning and reinforcement learning applied to self play, as exemplified with AlphaGo.

[mjlef]

I can confirm Larry and I still are making progress. You would think it would stop, but one thing leads to another and it gets better. How long this will last, I do not know. But it does not seem to be slowing down, although sometimes a whole week goes by without anything good enough to promote.

I have had some family health issues (both parents) to deal with which was slowing things down a bit this last week. But I am home again and able to work right now.

Mark

[Dann Corbit]

Last weekend's patches on Stockfish look very promising.

Look here:
http://abrok.eu/stockfish/
starting with "Doubled and supported pawns" on Sunday the 16th.

Another thing to keep in mind is that some improvements are more substantial for longer time control. Long time control test collections are hard to find.

Also, there are updated ASMfish (though they still lag a bit):
https://github.com/lantonov/asmFish/tre ... es/Windows

Currently a week behind the Stockfish engine updates. But a lot more NPS, so it is still probably a bit stronger.

[MikeB]

Last weekend's patches on Stockfish look very promising.

Look here:
http://abrok.eu/stockfish/
starting with "Doubled and supported pawns" on Sunday the 16th.

Another thing to keep in mind is that some improvements are more substantial for longer time control. Long time control test collections are hard to find.

Also, there are updated ASMfish (though they still lag a bit):
https://github.com/lantonov/asmFish/tre ... es/Windows

Currently a week behind the Stockfish engine updates. But a lot more NPS, so it is still probably a bit stronger.

+1 I agree 100% - what is especially nice is that the most recent patches scale, i.e, longer time control - very nicely

at least so so far in my limited number of 5 min 5 sec games

Code: Select all

Rank Name                      Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR 
---------------------------------------------------------------------------------------------------------
   1 Stockfish 170417 64 POPC   3111   0.0    9    9   258  136.5  52.9   28   13  217  10.9  84.1  3095 
   2 Stockfish 8 64 POPCNT      3097  14.2    9    9   255  125.0  49.0   17   22  216   6.7  84.7  3102 

The week before this , I wasn't seeing this at all. They test at super fast time controls of 10 sec with 0.1 second increment (STC) only the successful STC patches get tested at 60 sec with 0.6 increment , the process works, you can't argue with results , but sometimes the best patches at 10 sec /game or 60-/ sec per game are not the best patches at 5 min /game. Anyway, this past week may be the best week since the release of SF 8 ... time will tell.

[Jouni]

I also got after 500 games 60+0,6 SFDev vs. Houdini and Komodo +15 ELO to SF8. Nice, but it's almost 6 months now!

[pohl4711]

Last weekend's patches on Stockfish look very promising.

Look here:
http://abrok.eu/stockfish/
starting with "Doubled and supported pawns" on Sunday the 16th.

I just started the testrun of Stockfish 170417. Next Wednesday, the result will go online...Stay tuned !!

http://spcc.beepworld.de


Stefan

[melajara]

Thank you for the encouraging answers guys, much appreciated

[Lyudmil Tsvetkov]

In absentia of TCEC season 10 to truly assess chess engine progresses, judging from the leader Stockfish 8, it looks like progress has stopped this year.
More precisely, one can judge this when watching the graph here http://spcc.beepworld.de/

Now Komodo 10.4 maintained the momentum. Can Komodo team confirm that progress has not slowed down? In that case, Komodo 10.5 might regain the lead.

Meanwhile, Houdini 5 was an impressive achievement. Now, 6 months later, is there any update to be expected soon?

Ditto for the followers, Fizbo 1.9 and Andscacs 0.9 are still more than 200 Elo behind and I don't see new contenders filling the void just behind the big 3.

I have a definite feeling of slow down in chess engine progress so far in 2017, am I alone feeling so?

Perhaps it is about time to revisit the concept of Giraffe with deep learning and reinforcement learning applied to self play, as exemplified with AlphaGo.

this is definitely the end of the Framework, small issue however, bearing in mind that, statistically, there is about 3% chance a huge global war will not erupt within the next 2 years.