testing with or without book ?

[Daniel Anulliero]

I just wondered if it is better to test with a book openning or without ...
So , I decided for my self tests to play all games with all the first légal moves :
a3 a6
a3 a5
Etc...
This give me 400 diférents "opennings" and with reversed color , 800 games
Also I can see if my psqt's are not too bad lol
And you can see really funny games lol
What do you think about that ?
Bests
Dany

[Fabio Gobbato]

For my tests I made a little book of 6 plies. It's good enough to have different games and at the same time let the engine play the opening phase.

I chose only usual openings, I think it's better to improve the engine on those openings instead of improve it on positions that would never happen in a real game.

[brtzsnr]

Stockfish (and Zurichess) uses 2moves_v1/2moves_v2 opening books. All positions after 4 plies within +-80cp.

https://groups.google.com/d/msg/fishcoo ... dSlJdDmBEJ

[bob]

I just wondered if it is better to test with a book openning or without ...
So , I decided for my self tests to play all games with all the first légal moves :
a3 a6
a3 a5
Etc...
This give me 400 diférents "opennings" and with reversed color , 800 games
Also I can see if my psqt's are not too bad lol
And you can see really funny games lol
What do you think about that ?
Bests
Dany

I prefer without. I use a static set of starting positions and run the games with no opening books whatsoever. Book choices are problematic when there is either (a) randomness or (b) learning. That introduces another level of randomness and increases the number of games needed if you want to really estimate just the engine's strength, alone.