Andscacs - New version 0.87

[cdani]

I discovered that the new version is stronger in selfplay and against weaker opponents, so it is stronger in general but is not very interesting for me that is weaker against top engines.

So when I solve this I will publish a new version that should be stronger against stronger and weaker engines.

The previous 0.86189 had not this problem.

Was difficult to find the problem because it required long time control games against a gauntlet to show. In the standard stc/ltc games I run to test every little change, even at 80 seconds and in selfplay or against a gauntlet, this does not shows up.

I will not be surprised if something like this is what is happening to Stockfish. Simply some patches can be bad against strong engines, even if they are good in global and/or in selfplay.

Is not the first time that I observed this, but in general I was less worried about this than now

[Kyodai]

Thanks for the new version of Andscacs. I wonder if you have some
sort of estimation of the difference in elo between version 0.86 and 0.87?

Probably more than 20, but not sure as I have not done a serious tests against 0.86.

ok - thanks! Maybe it's a bit early to know for sure. My question was merely
connected to a private rating list of mine. +20 elos more would put Andscacs
slightly stronger than Critter 1.6a and Firenzina 2.4.3. This is progress - and
good luck also for the future!

[gotogo]

I will not be surprised if something like this is what is happening to Stockfish.

This is the new norm for Stockfish when playing TCEC. Everyone scratches head and never finds anything until to late! Everyone posts idea but nobody knows! All the compiled data and all the games played pointing to Stockfish possible win this year is thrown out the window miraculously. It has become the great Stockfish mystery that happens every TCEC season. I am going to call it the Stockfish Plague!

Watch them blame it on not scaling again. With all them draws Stockfish needs to depend on other engines win to survive this stage. It looking like Stockfish is doomed again!

only thing making TCEC watchable this year is the thought of Houdini 5 making a comeback.

[drj4759]

About 10 Elo increase from 0.86.189 TCEC version which is not enough to dislodge Fritz 15 from the rankings. See the gauntlet match here -> http://chessowl.blogspot.com/2016/06/an ... gines.html

[cdani]

ok - thanks! Maybe it's a bit early to know for sure. My question was merely
connected to a private rating list of mine. +20 elos more would put Andscacs
slightly stronger than Critter 1.6a and Firenzina 2.4.3. This is progress - and
good luck also for the future!

Thanks! I hope it continues improving!

It has become the great Stockfish mystery that happens every TCEC season. I am going to call it the Stockfish Plague!

As all we know, is normal that those things can happen. If you test for the objective A, you are not testing for the objective B. The problem is find ways to test for B, the real objective, and have some minimum guaranties.

About 10 Elo increase from 0.86.189 TCEC version which is not enough to dislodge Fritz 15 from the rankings. See the gauntlet match here -> http://chessowl.blogspot.com/2016/06/an ... gines.html

Many thanks for the test!! This version sure had not much improvements, was more a consolidation than anything.

[Sylwy]

http://talkchess.com/forum/viewtopic.ph ... 46&t=50060

[Jamal Bubker]

Thank you Daniel for releasing this new version.
Your engine is amazingly strong, now a Top Ten engine

Really a very tough opponent for any other top engines !

Please keep on working on it !