random evaluation perturbation factor

[Ras]

In my CT800 project, this feature is called "eval blurring", and the available options are "off" or "+/-10/30/50 centipawns".

[Gerd Isenberg]

By the 'Beal effect' are you referring to "Random Evaluations in Chess", Don F. Beal and Michael C. Smith, ICCA Journal [17(1):3-9], March 1994? Is a text available for that publication?

Yep, they demonstrated that root random (choosing a root move randomly) loses badly versus leaf random (random eval) the higher the search depth >= 1. To avoid favouring the leaf random look ahead approach due to mate scores, root random also applied a look ahead with the same search depth, but all leaves evaluated zero. Repetition and 50 move rule were ignored in the experiments. I don't have the paper and game scores actually handy, but will soon post a cpw page on that topic.

[Maarten Claessens]

By the 'Beal effect' are you referring to "Random Evaluations in Chess", Don F. Beal and Michael C. Smith, ICCA Journal [17(1):3-9], March 1994? Is a text available for that publication?

The article is also published in "Advances in Computer Chess 7", H.J. van den Herik, I.S. Herschberg and J.W.H.M. Uiterwijk (1994).

[Gerd Isenberg]

By the 'Beal effect' are you referring to "Random Evaluations in Chess", Don F. Beal and Michael C. Smith, ICCA Journal [17(1):3-9], March 1994? Is a text available for that publication?

A paper available online which refers the Beal/Smith paper:

Mark Levene, Trevor Fenner (2001). The Effect of Mobility on Minimaxing of Game Trees with Random Leaf Values. Artificial Intelligence, Vol. 130, No. 1

https://pdfs.semanticscholar.org/eb82/0 ... 1e5a2d.pdf

[Gerd Isenberg]

By the 'Beal effect' are you referring to "Random Evaluations in Chess", Don F. Beal and Michael C. Smith, ICCA Journal [17(1):3-9], March 1994? Is a text available for that publication?

Yep, they demonstrated that root random (choosing a root move randomly) loses badly versus leaf random (random eval) the higher the search depth >= 1. To avoid favouring the leaf random look ahead approach due to mate scores, root random also applied a look ahead with the same search depth, but all leaves evaluated zero. Repetition and 50 move rule were ignored in the experiments. I don't have the paper and game scores actually handy, but will soon post a cpw page on that topic.

https://chessprogramming.wikispaces.com ... eaf+Values

[D Sceviour]

Thank you for posting the page. The links are very helpful.

[Luis Babboni]

That random adition will be valid only for equal score positions or could even be enough to make a position with higher score than other with no randon having, with random, a lower score position than that other?

(Sorry for my Hollywood-Apache like english).